10

innovations
that could
change our
world
Volume 127 January/February
Number 1 2024

Display until March 4, 2024

 MIT TECHNOLOGY REVIEW’S

What Now?
NEW THIS YEAR
TWO LOCATIONS
SIGNATURE AI EVENT

The opportunities at the

intersection of
innovation and uncertainty

April May Save 25% with code

16 – 17, 2024 22 – 23, 2024 PRINTJF24
London, UK MIT campus EmTechDigital.com
02 From the editor

Happy
New Year!
The start of a new year offers such a great opportunity
to reflect while also thinking about what’s to come. That
is especially true for us, and for this issue. This year
marks the 125th anniversary of Technology Review. On
the back page of this issue, you can read the original
letter written in 1899 announcing the new publication.
That first issue, bound up with twine in our library, is
a remarkable artifact. And it’s just one of very many.
The MIT Technology Review archives bear witness
to the stunning advance of technological and scien-
tific progress across the 20th century and into the first
decades of the 21st. All year, we’re going to highlight
the most interesting stories and visuals found in those
back issues. This fall, we plan to dedicate an entire
issue to celebrating this milestone anniversary, look-
ing back in order to look ahead to the next 125 years.
We’re also launching a new monthly online series that
will explore how we’ve covered key technologies over
the years, how those technologies have evolved, and
Mat Honan
where they are going. is editor in
That last bit is key. We’re not interested in revisiting chief of
our history for its own sake. We’re instead thinking about MIT Technology
Review
this anniversary as a way to try to understand where
technology is headed next. And so it’s fitting that we
kick off the year, once again, with our annual look at society in fundamental ways. For this issue, rather than
10 Breakthrough Technologies (page 23). focusing on just one aspect of it, Will Douglas Heaven,
We began putting this list together in early summer. our senior editor for AI, stepped back to take in the big
We have met, talked, and argued over it ever since. picture (page 30) and ask where this powerful technol-
There was even one item (the robotaxi) that started ogy will go from here.
out on our list of breakthroughs but wound up instead We also have stories in this issue on the science of
as fodder for another newsroom tradition: a story on hunger and weight loss (featuring some truly deranged
the worst technologies of the year. (You can read why mice), the problems of greening cement and, by exten-
on our website.) sion, concrete (which I learned is the second-most-
There was one technology, however, that we never used material in the world by weight), solving the issue
doubted was having a genuine breakthrough moment: of noise in quantum computing (where tiny distur-
artificial intelligence. We’ve had AI on the list before, in bances = big problems), the progress toward earthquake
various capacities—including last year, when we high- prediction, the race to source rare earth elements, data
lighted its image-making capabilities. But 2023 was the poisoning, and much more.
year that it truly went mainstream. AI dominated our So I hope you enjoy this issue, and that you’ll keep
collective public consciousness, from the time OpenAI coming back to see what new advances and challenges
unleashed ChatGPT on November 30, 2022, through this year brings.
the entirety of 2023, which OpenAI closed out in a
tumult of executive and boardroom drama. Thank you,
ROBYN KESSLER

AI was at the center of strikes and lawsuits. It raised

new questions about the nature of human creativity Mat
and authorship, and it seems poised to radically alter mat.honan@technologyreview.com
 Realizing a steely vision for digital transformation.

www.technologyreview.com/thecloudhub
04
“There’s a sense of excitement and expectancy—and a feeling that
we’re making it up as we go.” –p. 30
Front
2 Letter from the editor
THE DOWNLOAD
7 A tool to thwart AI from using
artists’ work; a mouthguard
to protect against brain injury;
the ups and downs of nuclear
fusion; an online catalogue of
stolen heritage; an icy way to
save coral reefs; and why we
need a moonshot program for
computing.
EXPLAINED
18 How does a passive
house work?
This slowly growing building
trend is all about using less
energy in the first place.
By Patrick Sisson
PROFILE
20 Accounting for nature
Gretchen Daily is working
to make the environment
more of an element in eco-
nomic decision-making.
By Kathryn Miles
Cover
illustration by
Aaron Denton
Back
62 Before the ground shakes
Can machine learning solve
earthquake prediction?
The innovation issue By Allie Hutchison
68 Inside the race to make
23 10 Breakthrough Technologies 2024 critical materials
Every year, we look for promising technologies China has dominated the mar-
ket for rare earth elements.
poised to have a real impact on the world. Here are
The US is hoping to catch up.
the advances that we think matter most right now. By Mureji Fatunde
74 Displays are shrinking before
30 Six big questions for generative AI
our eyes
We’re still riding the generative-AI wave, but the
Could this be the moment
technology’s future—and ours—will be shaped by augmented-reality companies
what we do next. BY WILL DOUGLAS HEAVEN have been waiting for?
By Matthew Smith
38 Journey to the lair of the hunger demon
79 Recapturing the whimsy of the
Scientists have spent decades trying to unravel early internet
the intricate mysteries of the human appetite. Are The HTML Energy movement
they on the verge of finally determining how hunger aims to bring back the joys of
works? BY ADAM PIORE the early days. By Tiffany Ng
FIELD NOTES
46 Bring on the noise 82 What is a beautiful experiment?
Quantum computing is now confronting its biggest A new book invites us to mar-
technical stumbling block. BY MICHAEL BROOKS vel at the elegance of experi-
mentation. By Philip Ball
52 Climate’s hardest problem 125 YEARS
Cement is a climate nightmare. This startup is using
88 Happy anniversary
electricity to change that. BY CASEY CROWNHART For 125 years, Technology
Review has aimed to explain
the “always new and ever more
complex problems” we face.
 Masthead 05

Editorial Corporate Consumer marketing MIT Technology Review Insights

Editor in chief Chief executive officer and publisher Vice president, marketing and and international
Mat Honan Elizabeth Bramson-Boudreau consumer revenue Vice president, Insights and international
Alison Papalia Nicola Crepaldi
Executive editor, operations Director of retention marketing
Finance and operations Global director of custom content
Amy Nordrum Taylor Puskaric Laurel Ruma
Chief financial officer, head of operations
Executive editor, newsroom Director of acquisition marketing
Enejda Xheblati Senior manager of licensing
Niall Firth Alliya Samhat Ted Hu
General ledger manager
Editorial director, print Senior manager of acquisition marketing
Olivia Male Senior editor, custom content
Allison Arieff Courtney Dobson Michelle Brosnahan
Accountant
Editor at large Director of event marketing
Anduela Tabaku Senior editor, custom content
David Rotman Nina Mehta Kwee Chuan Yeo
Human resources director
Science editor Email marketing manager
Alyssa Rousseau Editor, custom content
Mary Beth Griggs Tuong-Chau Cai Teresa Elsey
Manager of information technology
News editor Circulation and print production manager
Colby Wheeler Senior project manager
Charlotte Jee Tim Borton Martha Leibs
Data analytics manager
Features and investigations editor
Christopher Doumas Project manager
Amanda Silverman Advertising sales Natasha Conteh
Managing editor Office manager
Senior vice president, sales and Director of partnerships, Europe
Timothy Maher Linda Cardinal
brand partnerships Emily Kutchinsky
Commissioning editor Andrew Hendler
Director of partnerships, Asia
Rachel Courtland Technology andrew.hendler@technologyreview.com
201-993-8794 Marcus Ulvne
Chief technology officer
Senior editor, MIT News
Drake Martinet Associate vice president, integrated
Alice Dragoon Board of directors
marketing and brand
Vice president, product
Senior editor, biomedicine Caitlin Bergmann Cynthia Barnhart, Cochair
Mariya Sitnova
Antonio Regalado caitlin.bergmann@technologyreview.com Alan Spoon, Cochair
Senior software engineer Lara Boro
Senior editor, climate and energy Executive director, brand partnerships
Molly Frey Peter J. Caruso II, Esq.
James Temple Marii Sebahar
Whitney Espich
Data engineer marii@technologyreview.com
Senior editor, AI Sanjay E. Sarma
Vineela Shastri 415-416-9140
Will Douglas Heaven David Schmittlein
Associate product manager Executive director, brand partnerships Glen Shor
Senior reporters
Allison Chase Kristin Ingram
Eileen Guo (features and investigations)
kristin.ingram@technologyreview.com
Jessica Hamzelou (biomedicine) Digital brand designer
415-509-1910
Customer service and
Melissa Heikkilä (AI) Vichhika Tep subscription inquiries
Tate Ryan-Mosley (tech policy) Executive director, brand partnerships
National
Stephanie Clement
Reporters Events stephanie.clement@
877-479-6505
Casey Crownhart (climate and energy)
Senior vice president, technologyreview.com International
Rhiannon Williams (news)
events and strategic partnerships 214-339-6115 847-559-7313
Zeyi Yang (China and East Asia)
Amy Lammers
Executive director, sales and brand Email
Copy chief
Director of event content and experiences partnerships customer-service@technologyreview.com
Linda Lowenthal
Brian Bryson Debbie Hanley
Web
Senior audience engagement editor debbie.hanley@technologyreview.com
Senior event content producer www.technologyreview.com/
Abby Ivory-Ganja 214-282-2727
Erin Underwood customerservice
Audience engagement editor Senior director, brand partnerships
Director of events Reprints
Juliet Beauchamp Ian Keller
Nicole Silva techreview@wrightsmedia.com
ian.keller@technologyreview.com
Creative director, print 877-652-5295
Event operations manager 203-858-3396
Eric Mongeon
Elana Wilner Licensing and permissions
Senior director, brand partnerships
Digital visuals editor licensing@technologyreview.com
Manager of strategic partnerships Miles Weiner
Stephanie Arnett
Madeleine Frasca Williams miles.weiner@technologyreview.com
Editorial fellows
Event coordinator
617-475-8078 MIT Technology Review
June Kim
Bo Richardson Senior director, digital strategy, planning, 196 Broadway, 3rd Floor
Abdullahi Tsanni
and ad ops Cambridge, MA 02139
Katie Payne 617-475-8000
katie.payne@technologyreview.com
Our in-depth reporting reveals what’s
Digital operations coordinator going on now to prepare you for what’s
Corrections to the November/December 2023 issue: coming next.
Brooke McGowan
On p. 13, the company mentioned in “Job title of the future” is brooke.mcgowan@technologyreview.com
Technology Review, Inc., is an indepen-
Persefoni (not Personifi). dent nonprofit 501(c)(3) corporation wholly
Media kit
On p. 14, a team of researchers studying captchas was said to be www.technologyreview.com/media owned by MIT; the views expressed in
based at the University of Illinois when they made their findings in our publications and at our events are not
2016. It was Columbia University. always shared by the Institute.

On p. 43, Kathryn Peters is the former executive director of CITAP.

We regret these errors.
 The Download 07

A new tool lets artists add invisible changes to the pixels in their
art before they upload it online so that if it’s scraped into an AI

The training set, it can cause the resulting model to break in chaotic
and unpredictable ways.
The tool, called Nightshade, is intended as a way to fight back

Download
against AI companies that use artists’ work to train their models
without the creator’s permission. Using it to “poison” this train-
ing data could damage future iterations of image-generating AI
models, such as DALL-E, Midjourney, and Stable Diffusion, by
rendering some of their outputs useless—dogs become cats,
cars become cows, and so forth. MIT Technology Review got an
exclusive preview of the research, which was submitted for peer

This new data- review at the computer security conference Usenix.

AI companies such as OpenAI, Meta, Google, and Stability

poisoning tool lets AI are facing a slew of lawsuits from artists who claim that
their copyrighted material and personal information was scraped

artists fight back without consent or compensation. Ben Zhao, a professor at the
University of Chicago who led the team that created Nightshade,

against generative AI
The tool, called Nightshade, messes up
training data in ways that could cause seri-
ous damage to image-generating AI models.
By Melissa Heikkilä
STEPHANIE ARNETT/MITTR | RIJKSMUSEUM, ENVATO
says the hope is that it will help tip the power balance back from
AI companies toward artists, by creating a powerful deterrent
against disrespecting artists’ copyrights and intellectual property.
Meta, Google, Stability AI, and OpenAI did not respond to MIT
Technology Review’s request for comment on how they might
respond.
Zhao’s team also developed Glaze, a tool that allows artists to
“mask” their own personal style to prevent it from being scraped
08 The Download
by AI companies. It works in a similar way to Nightshade: by
changing the pixels of images in subtle ways that are invisible
to the human eye but manipulate machine-learning models
into interpreting the image as something different from what
it actually shows.
The team intends to integrate Nightshade into Glaze, and
artists can choose whether they want to use the data-poisoning
tool or not. The team is also making Nightshade open source,
which would allow others to tinker with it and make their own
versions of it. The more people use it and make their own ver-
sions, the more powerful the tool becomes, Zhao says. The data
sets for large AI models can consist of billions of images, so the
more poisoned images can be scraped into the model, the more
damage the technique will cause.
A targeted attack
Nightshade exploits a security vulnerability in generative AI
models, one arising from the fact that they are trained on vast
amounts of data—in this case, images that have been hoovered
from the internet. Nightshade messes with those images.
Artists who want to upload their work online but don’t want
their images to be scraped by AI companies can upload them
to Glaze and choose to mask it with an art style different from
their own. They can then also opt to use Nightshade. Once AI
developers scrape the internet to get more data to tweak an
existing AI model or build a new one, these poisoned sam-
ples make their way into the model’s data set and cause it to
malfunction.
Poisoned data samples can manipulate models into learning,
for example, that images of hats are cakes, and images of hand-
bags are toasters. The poisoned data is very difficult to remove,
as it requires tech companies to painstakingly find and delete
each corrupted sample.
The researchers tested the attack on Stable Diffusion’s lat-
est models and on an AI model they trained themselves from
scratch. When they fed Stable Diffusion just 50 poisoned images
of dogs and then prompted it to create images of dogs itself, the
output started looking weird—creatures with too many limbs
and cartoonish faces. With 300 poisoned samples, an attacker
can manipulate Stable Diffusion to generate images of cats when
prompted to generate images of dogs.
Generative AI models are excellent at making connections
between words, which helps the poison spread. Nightshade
infects not only the word “dog” but all similar concepts, such as
“puppy,” “husky,” and “wolf.” The poison attack also works on
tangentially related images. For example, if the model scraped
a poisoned image for the prompt “fantasy art,” the prompts
“dragon” and “a castle in The Lord of the Rings” would similarly
be manipulated into something else.
Zhao admits there is a risk that people might abuse the
data-poisoning technique for malicious uses. However, he says
attackers would need thousands of poisoned samples to inflict
Concept Related prompts
A painting A castle in
Fantasy by Michael The Lord of
art Whelan A dragon the Rings
Clean
model
Poisoned
model
real damage on larger, more powerful models, as they are trained
on billions of data samples.
“We don’t yet know of robust defenses against these attacks.
We haven’t yet seen poisoning attacks on modern [machine-
learning] models in the wild, but it could be just a matter of time,”
says Vitaly Shmatikov, a professor at Cornell University who
studies AI model security and was not involved in the research.
“The time to work on defenses is now,” Shmatikov adds.
Gautam Kamath, an assistant professor at the University of
Waterloo who researches data privacy and robustness in AI mod-
els and wasn’t involved in the study, says the work is “fantastic.”
The research shows that vulnerabilities “don’t magically
go away for these new models, and in fact only become more
serious,” Kamath says. “This is especially true as these models
become more powerful and people place more trust in them,
since the stakes only rise over time.”
A powerful deterrent
Junfeng Yang, a computer science professor at Columbia
University who has studied the security of deep-learning sys-
tems and wasn’t involved in the work, says Nightshade could
have a big impact if it makes AI companies respect artists’ rights
more—for example, by being more willing to pay out royalties.
AI companies that have developed generative text-to-image
models, such as Stability AI and OpenAI, have offered to let art-
ists opt out of having their images used to train future versions
of the models. But artists say this is not enough. Eva Toorenent,
an illustrator and artist who has used Glaze, says opt-out poli-
cies require artists to jump through hoops and still leave tech
COURTESY OF THE RESEARCHERS
companies with all the power.
Toorenent hopes Nightshade will change the status quo.
“It is going to make [AI companies] think twice, because they
have the possibility of destroying their entire model by taking
our work without our consent,” she says. Q
Melissa Heikkilä is a senior reporter at MIT Technology Review.

Reducing the
risks of
traumatic
brain injury
A sophisticated new
mouthguard sheds new light
on major head impacts.
By David Hambling
When athletes or soldiers have a concus-
sion, the most beneficial course of action
is to simply get them off the playing field
or out of the action so they can recover.
Yet much about head injuries remains a
mystery, including the reasons why some
impacts result in concussion while oth-
ers don’t.
But new measuring devices are being
developed that could help deliver a wealth
of information about head impacts. By
giving an immediate warning that a per-
son needs to be removed from action or
play, they could help protect soldiers and
athletes alike from brain damage.
Appreciation of the real risks of head
injuries has been a long time coming.
“Even 10 years ago, if someone took a
big hit they were told to get up and play
or keep going,” says Mike Shogren, CEO
of Prevent Biometrics. “Now reducing
major head impacts and understanding
concussion risk is a major focus in sports
and the military.”
Prevent is one of several companies
developing new sensors to precisely mea-
sure and record head impacts, which would
help identify possible concussions and pro-
COURTESY OF PREVENT BIOMETRICS
vide data for studies of cumulative effects.
Scientists have been trying to measure
the forces involved in head trauma for a
long time, says Adam Bartsch, the compa-
ny’s chief science officer. “Decades ago,
scientists had to use Rube Goldberg con-
traptions to study head impact,” he says.
“Sometimes these were made from a dental
mold with a rigid plate and sensors big-
ger than dice, with a 10-meter-long cable
connecting it to a computer. The wearer
would drool and the data wasn’t perfect,
but it was the best they had.”
First conceived at the Cleveland Clinic,
Prevent’s device, the Impact Monitoring
Mouthguard (IMM), fits into the wearer’s
mouth, working as both a monitoring tool
and a functional mouthguard. It calculates
the force, location, direction, and number
of impacts and can then transmit data via
Bluetooth to other devices for assessment.
Prevent is using the IMM to study
parachute landing falls (or PLFs), a land-
ing technique that was developed by the
United States Army as part of its para-
trooper training program, using over 2,000
paratroopers as subjects. A correctly exe-
cuted PLF absorbs the shock of hitting the
ground as the parachutist lands feet first
and falls sideways, successively distrib-
uting the landing shock along the calves,
thighs, hips, and back. But an error can
whip the parachutist’s head backwards
and onto the ground. The IMM’s sensors
revealed that this occurs far more often
than anyone realized.
“We found a significant head impact in
about 5% of jumps,” says Bartsch. “That’s
about 30 times as much as the published
IMM charging
case
Solo
charging
case
The Download 09
incidence of concussion in paratroop-
ers.” A battery of tests confirmed that the
events the mouthguard registered as pos-
sibly causing concussions had in fact done
so. Paratroopers tend to just get up and
carry on after a bad landing, so the offi-
cial figures had previously reflected only
the injuries of those who were physically
unable to get up on their own.
In sports, similarly, athletes are often
encouraged to “get over it” rather than
report an injury. Prevent is carrying out
a large-scale project with World Rugby,
which will monitor players and allow
coaches to take injured players off the
field and have them assessed. (Several
other instrumented mouthguards—the
Biocore, the ORB, and HitIQ—are being
developed for other sports, including box-
ing and lacrosse.) In the future, Prevent
hopes to be able to evaluate the total effect
of lots of smaller shocks to see under what
circumstances they cause serious cumula-
tive injury. “Understanding total exposure
on top of just major impacts is also crit-
ical,” Shogren says. “It’s like in a boxing
match. The impact that knocks you out at
the end might not have knocked you out
on its own in the first round.” Q
David Hambling is a technology journalist
based in South London.
Web data
portal
Mobile app
Mouthguard
10 The Download

The Mirror
Fusion Test
Facility
A billion-dollar nuclear fusion
machine was shuttered the
day it was dedicated. It was
never turned on.
By Jon Keegan/BeautifulPublicData.com

On Friday, February 21, 1986, a group of

300 scientists and engineers gathered
for a dedication ceremony at Lawrence
Livermore National Laboratory (LLNL).
After final tests, the completion of Mirror
Fusion Test Facility-B (MFTF-B) was cele-
Above: The Right: The
brated, and a letter from John Herrington, installation of MFTF-B under
Ronald Reagan’s secretary of energy, was a magnet into the construction in
presented to program director T. Kenneth MFTF-B in 1981. 1983.

Fowler, extending his congratulations on

a job well done.
On the very same day, after nearly a
decade of development and close to a
billion dollars of funding (in 2023 dol-
lars, that is), the project was shut down,
the massive machine never having been
turned on.
“I want all of you to know how much I
regret the fact that, just as you complete
this remarkable new facility, the budget
pressures dictate that we must put it into
standby and not operate it as you might
have hoped,” wrote Herrington in the
letter to Fowler. “This is frustrating, and
“yin-yang” shape; its interior was cooled
by pumping liquid helium through the
vessel. Thirty miles of wire were wound
over the course of a year to make the
magnet’s conductor.
The construction was capable of gen-
erating magnetic fields 150,000 times
stronger than Earth’s.
The energy crisis of the 1970s had
motivated the US to throw lots of money a gut punch to the researchers. And
at alternative energy sources, and two Livermore’s Building 431, the project’s
major research pathways emerged in home, was demolished in 2005 after
the quest for nuclear fusion power: the failing to meet the threshold of histor-
doughnut-shaped tokamak design used ical significance to be protected on the
COURTESY OF LAWRENCE LIVERMORE NATIONAL LABORATORY

perhaps not the best use of our national
talent and resources, but we must bring
the deficit under control.”
I came across photos from the con-
struction of the facility’s components on
LLNL’s website and was captivated by
one from 1980 showing a strange twist-
ing mass of metal that at the time was
the largest superconducting magnet in
the world.
This 350-ton magnet was encased
in stainless steel built in a distinctive
by researchers at Princeton and MFTF-
B’s approach, which involved bouncing
superheated plasma off two opposing
“magnetic mirrors” at either end of a
linear vessel.
MFTF-B program director Fowler was
quoted as saying, “Building big machines
is a mixture of lead times, resources,
prudence, and gambling.” But the gam-
ble didn’t pay off.
The Reagan administration’s deci-
sion to mothball the machine came as
National Register of Historic Places.
With the recent news that LLNL’s
National Ignition Facility had achieved
the first nuclear fusion reaction to pro-
duce a net energy gain, the outlook for
fusion research looks brighter again.
The key to success turned out to
be containing the plasma within 192
high-powered lasers, focused on a tiny
pellet. Pounding it with 2 million joules
of energy created a fusion reaction that
lasted for 100 trillionths of a second. Q
 The Download 11

The 350-ton
“yin-yang”
magnetic mirrors
encased what
was the largest
superconducting
magnet in the
world in 1980.
12 The Download

includes, for the first time, Edo-centric

An online
catalogue of a
stolen heritage
The Digital Benin project
provides a central place to
see the Nigerian artifacts
scattered around the Global
North. Its organizers hope it
will be the first step toward
repatriation.
By Gouri Sharma
When British forces raided the African
kingdom of Benin in the late 19th century,
they took with them thousands of sculp-
tures dating back centuries. Sold to pri-
vate collectors and museums in the Global
North, the artifacts, known as the Benin
Bronzes, included ceremonial swords, rit-
ualistic statues, and musical instruments
that belonged to the Edo people. Their
looting had a profound impact on the
community, descendants of which live
in and around modern-day Benin City in
southern Nigeria. It effectively severed
the connection between them and their
cultural heritage.
Despite longstanding pressure, many
institutions, particularly in Germany and
the UK, have resisted calls to share infor-
mation about the Benin Bronzes in their
collection. This makes it difficult to deter-
mine how many there are outside Nigeria,
and where.
But the launch in November 2022 of a
knowledge, history, and language, includ-
ing traditional Edo names for the objects
and the locations where they originated.
The platform was built by a team based
in the US, Europe, and Nigeria with an
initial €1.2 million from a fine-art foun-
dation in Berlin called Ernst von Siemens
Kunststiftung. It has been designed spe-
cifically to ensure accessibility for people
who use the internet predominantly via
mobile, as most Nigerians do.
Anne Luther, a digital heritage spe-
cialist and one of Digital Benin’s principal
investigators, hopes the project will be a
model for other communities that want
to explore how technology can be used to
help them reconnect with and reclaim sto-
len artifacts. “Before Digital Benin, digital
restitution was not a conversation being
had by larger institutions,” Luther says.
In building the catalogue, the team
received more than 400 data sets from
131 institutions in 20 countries, which
they then drew from in order to make the
objects accessible, findable, and traceable. Leopard head hip
Luther says being able to assure institu- ornament from
tions that they would be able to retain their Benin (16th–19th
century).
rights to and ownership of any data sets
they shared meant that more institutions
were willing to do so. for research. It has been encouraging for
But she says the site is just a step on me to see people appreciate the informa-
the road to restoring the artifacts to the tion we have put out there.”
Edo. Pressure to make such restitution Luther and her colleagues are develop-
has been growing on a number of gov- ing a new project—contingent on fund-
ernments and institutions, including the ing—with the Living Arts Foundation
British Museum, which has more than 900 in Cambodia to preserve the country’s
objects. And there are signs the pressure performing arts traditions. Digital Benin,
is working: Germany, for example, has meanwhile, has expanded to include more
repatriated 21 Benin Bronzes formerly than 4,000 archival documents linked to
stored in its institutions. the Benin Bronzes.
Eiloghosa Obobaifo, an anthropologist But the team hopes the Digital Benin
based in Benin City, says the site has been model can grow to a global scale. “The
COURTESY OF METROPOLITAN MUSEUM OF ART

pioneering platform called Digital Benin
has changed that. Hosted by the Museum
am Rothenbaum (MARKK) in Hamburg,
Digital Benin is an online catalogue of
more than 5,000 Benin Bronzes. Its orga-
nizers say it is a form of digital restitu-
tion, using technology to reestablish a
community’s link to artifacts that were
looted from native lands. The catalogue
helping locals, including contemporary
bronze casters who make sculptures with
ancient techniques, reconnect with their
history. “Bronze casters have become very
dependent on the platform and use it to
see objects they could re-create,” says
Obobaifo, who carried out research for the
project. “It has also had an impact on edu-
cation, with students using the platform
long-term goal is to develop a database
that connects all objects in all institutions
worldwide,” Luther says. She estimates that
it would take $5 million over five years to
develop and maintain an expanded pro-
totype system. Q
Gouri Sharma is a journalist based in
Berlin.

Coral reefs around the
world are being stressed
by warming, more acidic
waters.
Putting coral on ice
Recent advances in cryopreservation
are a lifeline for rapidly vanishing reefs.
By Allison Guy
As sweltering ocean temperatures make graveyards of coral
reefs across the Caribbean and beyond, a team of scientists is
scrambling to cool corals down. Way down. To -200 °C.
The Coral Biobank Alliance, helmed in part by Smithsonian
Conservation Biology Institute biologist Mary Hagedorn, aims
to cryopreserve or otherwise keep in captivity the roughly 1,000
species of reef-building corals on the planet. Coral reefs, as an
ecosystem, are expected to go functionally extinct by 2035.
Hagedorn hopes to freeze enough coral sperm, larvae, and
adult polyps not just to support current conservation efforts—
such as breeding heat-resistant corals—but to reboot reefs in
COURTESY OF SMITHSONIAN’S NATIONAL ZOO AND CONSERVATION BIOLOGY INSTITUTE
the increasingly likely event of a marine mass extinction. Yet
major engineering hurdles have stood in the way of this plan.
Ice is perhaps the chief enemy of successful cryopreserva-
tion. “Ice is a crystal,” says Matthew Powell-Palm, a thermo-
dynamicist at Texas A&M University. “What we want instead
is to get the water inside the coral tissue to form a glass.” At a
molecular level, he explains, glass is akin to a solidified liquid,
with none of the jagged crystals that can rip apart delicate tis-
sues. To stave off ice formation, samples are doused in a glass-
promoting, cryoprotective solution and then rapidly cooled. A
separate process is needed for the coral’s symbiotic algae. And
any techniques must be developed with an eye to the limited
technological capacities of remote field stations in places such
as Australia or Indonesia.
Thawing is its own challenge. Corals are sickly and infec-
tion-prone in the first days and weeks after they are thawed, so
the team is developing a cocktail of antibiotics, probiotics, and
The Download 13
antioxidants to boost survival rates. “Think of a patient who’s
just had open-heart surgery,” Hagedorn says. “They’re not going
to walk out the door the next day.”
The difficulty of cryopreservation grows with the size, com-
plexity, and fat content of the tissue. Sperm are relatively simple,
Hagedorn says, while eggs, which are fatty, are viable for only a
few hours, making them “just too hard to preserve.” To prevent
extinctions, there need to be easy-to-use methods to freeze
larvae—the immature, free-swimming form corals take early
in their life cycle—along with fragments from mature colonies.
Hagedorn and her collaborators have been able to cryo-
preserve larvae since 2018, but the process was finicky, and
thawing them required sophisticated lasers to heat them at a
rate of more than 1,000,000 °C per minute. Now, Hagedorn’s
team has devised and tested a simplified method. The larvae
are dehydrated on a steel mesh, immersed in a cryoprotectant,
and plunged into liquid nitrogen. To thaw them, the mesh is put
in a room-temperature rehydration solution. If all goes well, the
larvae swim off within two hours. “That’s been a huge advance-
ment,” Hagedorn says. “And what that means is that we can stop
the extinction of species, and in a way that’s very user-friendly.”
There has also been progress preserving adult corals. In
August, Powell-Palm, Hagedorn, and others announced that
they had seen cryopreserved adult specimens survive one or
two days before falling prey to bacteria. The process they used,
called isochoric vitrification, employs a small, confined chamber
to prevent liquids from expanding into ice. “I always describe
it as low tech, high science,” Powell-Palm says. “The thermo-
dynamic processes at play within the chamber are wonderfully
complex ... but technologically, it’s unbelievably simple.”
If—and Hagedorn emphasizes the if—isochoric vitrification
is perfected, it would be a game changer. Rather than requiring
researchers to wait for the few nights a year that corals spawn
to catch their sperm and larvae, this method will allow them
to collect and cryopreserve corals whenever it’s convenient.
Thawed fragments would grow into larger colonies and one
day spawn as well, helping to restart dead or struggling reefs.
The most intractable problem, however, might be time. As
high temperatures, acidic water, pollution, and disease batter
corals, they may grow too sickly to withstand the stresses of
cryopreservation. Corals need to be collected as soon as pos-
sible, Hagedorn says, while their health is still mostly intact.
At the same time, she’s planning for a conservation program
that may last for decades—even centuries, because the oceans
may continue to warm for many generations, even in the event
of decisive climate action. Hagedorn is investigating the feasi-
bility of storing coral and other animal tissues on permanently
cool parts of the moon. If some disaster were to befall coral
repositories on Earth, samples tucked in lunar lava tubes or
shady craters could persist for centuries. Q
Allison Guy reports on health and the environment.
14 The Download
Book reviews
Uncertain: The Wisdom
and Wonder of Being Unsure
By Maggie Jackson (Prometheus
Books, 2023)
In a world where swift resolutions and
assurances are often sought, journalist
Maggie Jackson contends that embracing
uncertainty is precisely what enables us
to envision the unimaginable and find
strength in difference and difficulty. From
a surgeon’s operating room to the AI lab developing an “I don’t
know” robot, the scenarios described here invite readers to seek
the unsung value of not knowing.
White Holes
By Carlo Rovelli (Riverhead Books, 2023)
In explaining his hypothesis that black
holes can turn into white holes (theoretical
cosmic regions that nothing can enter),
Rovelli writes, “I do not know if it is
correct. I do not even know if white
holes actually exist.” For this theoretical
physicist, science is all about the journey,
not the destination, and his book is as
poetic and playful as it is rigorous: “Are they really out there?”
he asks. “Who knows … I would love it if it were the case.”
The Blue Machine:
How the Ocean Works
By Helen Czerski (Norton, 2023)
We have built a culture of ignoring the
realities of living on a finite planet, says
Czerski, a physicist and oceanographer.
She believes we have especially taken
the ocean for granted—at our peril.
Ocean systems are changing now, she
writes, and “the best immediate action
is to stop causing the damage. But longer term interventions
to repair the damage that’s already been done are going to be
complicated and we can’t run away from that complication.”
Women in Science Now: Stories
and Strategies for Achieving Equity
By Lisa M.P. Munoz (Columbia University
Press, 2023)
Recalling her first days as an engineering
student at Cornell in the mid-1990s,
science writer Munoz says, “I did not
feel out of place. I believed it was where
I belonged. Then I started my first-year
engineering courses and things changed.”
She argues that the current “leaky pipeline” metaphor needs to
be retired as a way of describing the forces at play for women
in STEM: “Women are not dripping through the holes in the
system; they are being pushed out of a system that historically
did not want them in the first place, even if it wants them now.”
Combining first-person accounts from scientists with empirical
research, Munoz aims to offer a path toward making the sciences
more inclusive and equitable.
The Dimensions of a Cave
By Greg Jackson (Macmillan, 2023)
In this unsettling novel, a modern-day
retelling of Joseph Conrad’s Heart of
Darkness, investigative reporter Quentin
Jones uncovers a covert government
program that is developing a lifelike
virtual reality—only to have the story he’s
written on it killed by higher-ups. Digging
deeper, he discovers VIRTUE, cutting-edge
technology that simulates reality during interrogation, and things
really get complicated as his physical and digital worlds collide.
Dust: The Modern World in a Trillion
Particles By Jay Owens (Abrams, 2023)
Dust, explains writer and researcher Jay
Owens, “is matter at the very limit point
of formlessness, the closest ‘stuff’ gets
to nothing.” She first started thinking
seriously about it while procrastinating on
her dissertation and would later come to
see that “the dust under your sofa contains
the world.” When she got caught in a dust
storm en route to the Mojave Desert, she realized that she saw
COURTESY OFTHE PUBLISHERS
in the substance “the ghost of stories I might come to tell—of
nuclear fallout, volcanic ash and air pollution, of climate modelling,
space science and geoengineering.” And also a new dissertation
topic. Ultimately, she writes, “dust offered a route into facing
the catastrophe of our time: the climate crisis and all its kindred
ruptures into the biogeochemical systems of our planet.”

Op-Ed
We need a moonshot
for computing
The US government aims to push
microelectronics research forward.
But maintaining competitiveness in
the long term will require embracing
uncertainty.
By Brady Helwig and PJ Maykish
In its final weeks, the Obama administration released
a report that rippled through the federal science and
technology community. Titled Ensuring Long-Term US
Leadership in Semiconductors, it warned that as con-
ventional ways of building chips brushed up against
the laws of physics, the United States was at risk of
losing its edge in the chip industry. Five and a half
years later, in 2022, Congress and the White House
collaborated to address that possibility by passing the
CHIPS and Science Act—a bold venture patterned
after the Manhattan Project, the Apollo program, and
the Human Genome Project. Over the course of three
administrations, the US government has begun to
organize itself for the next era of computing.
Secretary of Commerce Gina Raimondo has gone
so far as to directly compare the passage of CHIPS to
President John F. Kennedy’s 1961 call to land a man
on the moon. In doing so, she was evoking a US tradi-
tion of organizing the national innovation ecosystem
to meet an audacious technological objective—one
that the private sector alone could not reach. Before
JFK’s announcement, there were organizational chal-
lenges and disagreement over the best path forward
to ensure national competitiveness in space. Such is
the pattern of technological ambitions left to their
TIM HERMAN/INTEL
own timelines.
Setting national policy for technological develop-
ment involves making trade-offs and grappling with
unknown future issues. How does a government
The Download 15
account for technological uncertainty? What will the
nature of its interaction with the private sector be?
And does it make more sense to focus on boosting
competitiveness in the near term or to place big bets
on potential breakthroughs?
The CHIPS and Science Act designated $39 bil-
lion for bringing chip factories, or “fabs,” and their
key suppliers back to the United States, with an
additional $11 billion committed to microelectron-
ics R&D. At the center of the R&D program would
be the National Semiconductor Technology Center,
or NSTC—envisioned as a national “center of excel-
lence” that would bring the best of the innovation
ecosystem together to invent the next generation of
microelectronics.
In the year and a half since, CHIPS programs and
offices have been stood up, and chip fabrication facil-
ities in Arizona, Texas, and Ohio have broken ground.
But it is the CHIPS R&D program that has an oppor-
tunity to shape the future of the field. Ultimately, there
is a choice to make in terms of national R&D goals:
the US can adopt a conservative strategy that aims
to preserve its lead for the next five years, or it can
orient itself toward genuine computing moonshots.
The way the NSTC is organized, and the technology
programs it chooses to pursue, will determine whether
the United States plays it safe or goes “all in.”
Welcome to the day of reckoning
In 1965, the late Intel founder Gordon Moore famously
predicted that the path forward for computing involved
cramming more transistors, or tiny switches, onto
16 The Download
flat silicon wafers. Extrapolating from the birth of
the integrated circuit seven years earlier, Moore
forecast that transistor count would double regularly
while the cost per transistor fell. But Moore was not
merely making a prediction. He was also prescribing
a technological strategy (sometimes called “transistor
scaling”): shrink transistors and pack them closer and
closer together, and chips become faster and cheaper.
This approach not only led to the rise of a $600 bil-
lion semiconductor industry but ushered the world
into the digital age.
Ever insightful, Moore did not expect that tran-
sistor scaling would last forever. He referred to the
point when this miniaturization process would reach
its physical limits as the “day of reckoning.” In 2023,
the chip industry is very close to reaching that day,
if it is not there already. Costs are skyrocketing and
technical challenges are mounting. Industry road maps
suggest that we may have only about 10 to 15 years
before transistor scaling reaches its physical limits—
and it may stop being profitable even before that.
To keep chips advancing in the near term, the
semiconductor industry has adopted a two-part
strategy. On the one hand, it is building “accelera-
tor” chips tailored for specific applications (such as
AI inference and training) to speed computation. On
the other, firms are building hardware from smaller
functional components—called “chiplets”—to reduce
costs and improve customizability. These chiplets
can be arranged side by side or stacked on top of
one another. The 3D approach could be an especially
powerful means of improving speeds.
This two-part strategy will help over the next 10
years or so, but it has long-term limits. For one thing,
it continues to rely on the same transistor-building
method that is currently reaching the end of the line.
And even with 3D integration, we will continue to
grapple with energy-hungry communication bot-
tlenecks. It is unclear how long this approach will
enable chipmakers to produce cheaper and more
capable computers.
Building an institutional home
for moonshots
The clear alternative is to develop alternatives to con-
ventional computing. There is no shortage of candi-
dates, including quantum computing; neuromorphic
computing, which mimics the operation of the brain
in hardware; and reversible computing, which has
the potential to push the energy efficiency of com-
puting to its physical limits. And there are plenty of
novel materials and devices that could be used to
build future computers, such as silicon photonics,
magnetic materials, and superconductor electronics.
These possibilities could even be combined to form
hybrid computing systems.
None of these potential technologies are new:
researchers have been working on them for many
years, and quantum computing is certainly making
progress in the private sector. But only Washington
brings the convening power and R&D dollars to
help these novel systems achieve scale. Traditionally,
breakthroughs in microelectronics have emerged
piecemeal, but realizing new approaches to compu-
tation requires building an entirely new computing
“stack”—from the hardware level up to the algorithms
and software. This requires an approach that can
rally the entire innovation ecosystem around clear
objectives to tackle multiple technical problems in
tandem and provide the kind of support needed to
“de-risk” otherwise risky ventures.
Does it make more sense to focus on boosting
competitiveness in the near term or to place
big bets on potential breakthroughs?
The NSTC can drive these efforts. To be successful,
it would do well to follow DARPA’s lead by focusing
on moonshot programs. Its research program will
need to be insulated from outside pressures. It also
needs to foster visionaries, including program man-
agers from industry and academia, and back them
with a large in-house technical staff.
The center’s investment fund also needs to be
thoughtfully managed, drawing on best practices
from existing blue-chip deep-tech investment funds,
such as ensuring transparency through due-diligence
practices and offering entrepreneurs access to tools,
facilities, and training.
It is still early days for the NSTC: the road to suc-
cess may be long and winding. But this is a crucial
moment for US leadership in computing and micro-
electronics. As we chart the path forward for the
NSTC and other R&D priorities, we’ll need to think
critically about what kinds of institutions we’ll need
to get us there. We may not get another chance to
get it right. Q
Brady Helwig is an associate director for economy
and PJ Maykish is a senior advisor at the Special
Competitive Studies Project, a private foundation
focused on making recommendations to strengthen
long-term US competitiveness.
 Give the gift
of knowledge.
Gifting an annual subscription to MIT Technology Review will keep the
tech-obsessed people on your gift list inspired all year long.

A gift subscription includes:

• In-depth, thoughtful stories on AI, climate change, biotech, and more
• Expert insights on emerging technologies impacting the world
• Unlimited access to our website, app, and magazine archives
• Discounts on all our signature events

Scan this code to

purchase a gift subscription
or go to TechnologyReview.com/GiveAGift
18 Explained
This slowly growing building trend
is all about using less energy in the
first place.
How does
a passive house
work?
W
hen the Canadian
engineer Harold Orr
and his colleagues
began designing an
ultra-efficient home in
Saskatchewan in the late ’70s, respond-
ing to a provincial conservation mandate
during the oil embargo, they knew that the
trick wasn’t generating energy in a greener
way, but using less of it. They needed to
make a better thermos, not a cheaper cof-
fee maker.
The result was the 1978 Saskatchewan
Conservation House, a cedar-clad trape-
zoid that cut energy usage by 85%—and
helped inspire today’s globally recognized
passive-house standard for building design.
Adopted by thousands of buildings com-
prising tens of thousands of housing units,
this concept marries vernacular building
techniques, like orienting toward the sun,
with cutting-edge insulation and air cir-
culation systems. The formula for these
efficient homes, standardized and shared
By Patrick Sisson
Illustration by Arthur Mount
by the German physicist Wolfgang Feist
and the Swedish structural engineer Bo
Adamson beginning in 1988, also bestows
health benefits. With airtight exteriors and
better air circulation, these homes offer
improved interior air quality and signifi-
cant noise reduction.
It’s a marriage of efficiency and rigor-
ously applied physics, says Bronwyn Barry,
a passive-house pioneer and principal of
a Bay Area architecture firm. If homes are
machines for living, passive-house design
principles offer a blueprint for a better
machine, highlighting just how poorly con-
structed postwar suburban sprawl can be.
Passive design focuses on the exterior,
or envelope, which needs to be tightly insu-
lated to avoid allowing heat out or unwanted
heat in. This means using thick thermal insu-
lation and high-quality, often triple-pane
windows, which let in the sun’s light and
warmth but keep heat from escaping. Heat
loss (and, in warm weather, gain) through
standard windows necessitates 25% to 30%
1
1
High-performance windows
with orientation and shading
Heat loss and gain through
standard windows accounts
for 25% to 30% of residential
energy use.
of residential energy use. Construction also
eliminates thermal bridges, or breaks in
the envelope or insulation that allow heat
to drain out. Think “boxy but beautiful,”
as Barry once wrote: houses boast contin-
uous layers of insulation while minimiz-
ing the cantilevers, corners, dormers, and
other features that characterize the messy
rooflines of McMansions. These design
requirements result in airtight buildings,
as measured by a blower door test: after
a specially calibrated door-mounted fan
sucks air out of the house to lower the air
pressure inside, technicians look for gaps
and cracks where higher-pressure air from
the outside flows in.
While this single-minded focus on
efficiency, or building the best thermos,

2 3
2 3 4 5
Airtight building Climate-appropriate Thermal-bridge-free
envelope insulation levels design
Keeping air, and thus heat, Some homes are even built Weak points in a building
from leaking out or in entirely from insulating envelope that allow heat
further reduces the need materials to minimize the to drain out of a house are
for heating and cooling. need for energy supply. eliminated.
Passive leads to exceptional performance—up to
a 90% reduction in heating and cooling
houses now demand—passive houses can’t circulate
air like traditional builds. But heat recovery
account for ventilators or energy recovery ventilators
can address that problem, exchanging air
less than 1% without sacrificing interior heat.
Often considered a cold-climate
of multifamily approach, passive houses actually have
universal benefits. Warmer climates simply
construction. require different windows and exteriors,
and additional shading to combat exces-
sive heat gain.
The passive-house movement has
expanded well beyond single-family homes
and the German and Nordic regions where
it’s most popular. There are now 275-
plus finished multifamily projects in the
Explained 19
Continuous ventilation
with heat recovery
Fresh, filtered air enters
the home through a heat
recovery ventilation (HRV)
system.
US alone, including dorms at Cornell
University, scores of affordable housing
projects across New York City, and the
newly opened Winthrop Center, a 53-story
skyscraper in downtown Boston. But even
though the Passive House Network, an
educational organization for the building
industry, has found that costs are com-
petitive for these large-scale projects, and
incentives introduced by the Biden admin-
istration through the Inflation Reduction
Act could decrease costs even more, pas-
sive houses still account for less than 1%
of all multifamily construction in the US
in the past decade.
Patrick Sisson, a Chicago expat
living in Los Angeles, covers
technology and urbanism.
20 Profile

W
hat is the true value of
a honeybee? A moun-
tain stream? A man-
grove tree?
G re tc h e n D a i ly,
cofounder and faculty director of the
Stanford Natural Capital Project, has
dedicated her career to answering such
complex questions. Using emerging sci-
entific data and the project’s innovative
open-source software, Daily and her
team help governments, international
banks, and NGOs to not only quantify
the value of nature, but also determine
the benefits of conservation and ecosys-
tem restoration.
This marriage of ecological and eco-
nomic concerns may seem an unusual
one to some. But to Daily, it’s a union
as natural as the planet’s ecosystems
themselves.
Daily completed her doctoral work
in ecology at Stanford during the 1990s.
It was, she says, a revolutionary time for
interdisciplinary approaches to both eco-
nomic and ecological crises. Spurred by
a summit hosted by the Royal Swedish
Academy of Scientists, ecologists and
economists began coming together for
the first time to consider the benefits of
a joint approach to developing economic
and environmental policy.
“For so much of our history, humanity
had operated under the assumption that
nature was infinite,” says Daily. “We knew
that collapses of civilization were at least
in part because of the destruction of the
local environment, but nobody thought
that could happen at a planetary scale.”
Global climate change and its myriad

Accounting for
impacts changed all that. “That crisis forced
us all to rethink the assumptions on which
economic systems operate,” she says. “It

nature also revealed the frailties in different lines

Gretchen Daily is working
to make the environment more
of an element in economic
decision-making.
of inquiry that have built up for decades
and even centuries.”
In 1997, Daily edited Nature’s
By Kathryn Miles Services: Societal Dependence on Natural
Portrait by Patrick Tehan Ecosystems—one of the first books to intro-
duce the concept of ecosystem services,
a field that seeks to quantify the value of

resources such as clean water, fertile soil,
and species habitats. The release of that
book inspired unprecedented interdisci-
plinary collaboration on issues of ecology
and economics.
“I think many of us finally began to see
that, fundamentally, environmental prob-
lems are economic and social problems,”
she says. “We cannot maintain the vital-
ity and security of the biosphere without
valuing nature.”
That recognition, Daily says, inspired
her to create the Natural Capital Project
in 2005. More than anything, she adds,
the initiative was born out of the idea
that mapping and modeling the value
of nature would compel global leaders
to see the inherent benefits of conser-
vation as well.
A partnership between Stanford, the
Chinese Academy of Sciences, the Nature
Conservancy, the University of Minnesota,
and the World Wildlife Fund, the Natural
Capital Project now works with banks,
governments, and nonprofit organizations
around the globe.
The organization’s open-source soft-
ware model, called InVEST, combines
data gleaned from thousands of research-
ers working with techniques such as
satellite imaging, soil surveys, climate
modeling, and human development map-
ping to quantify and place a value on
natural resources. Recent advances in
this data collection, along with machine
learning and software modeling, allow the
Natural Capital team to evaluate ecosys-
tems at a level of detail and sophistica-
tion previously considered impossible.
In a recent project undertaken for the
Colombian government, for instance,
the Natural Capital Project assisted in
establishing a conservation plan for the
Caribbean Gulf of Morrosquillo and its
hinterlands. The region’s Rio Sinú is an
essential source of drinking water for
many downstream communities but also
originates in an area that depends upon
logging, ranching, and agriculture for its
financial security. Using InVEST, Daily
and her team were able to determine
“Many of us finally began to see
that, fundamentally, environmental
problems are economic and social
problems. We cannot maintain the
vitality and security of the biosphere
without valuing nature.”
the actual cost of silt deposition in the
river, particularly for drinking water and
hydropower, and the value of maintaining
upstream forests that would prevent that
congestion from occurring.
“We were able to show that commu-
nities in the region were benefiting from
this forest in ways they hadn’t necessarily
realized,” says Lisa Mandle, lead scientist
and director of science-software inte-
gration for the Natural Capital Project.
“We can never capture the total value
of a forest in terms of cultural and spir-
itual values or even biodiversity, but we
can say that it has measurable economic
values across dimensions that have not
been considered before.”
And that, says Mandle, has created
powerful incentives for the Colombian
government to think about how to sup-
port the communities within that cru-
cial forest.
A similar approach, also crafted by
the Natural Capital Project, helps coun-
tries determine their gross ecosystem
product, or GEP. Modeled after the gross
domestic product, the GEP index allows
nations to determine the monetary value
of their ecological systems. Daily and her
team piloted this index in 2014 on both
municipal and national scales in China,
and it was adopted by the United Nations
Statistical Commission in 2021.
“Just as the Great Depression exposed
the urgent need for better macroeconomic
performance metrics, our current ‘Great
Degradation’ of natural capital is mak-
ing it imperative that we track ecological
performance and use that information to
Profile 21
guide investments in revitalization and
regeneration,” says Daily, who predicts that
the GEP metric will be employed globally
within the next decade.
In the meantime, she and her team
are dedicated to streamlining their eco-
logical assessments in a way that makes
the final analysis and visualization easier
for political leaders, investors, and local
communities to use. Making that infor-
mation more accessible, she says, will be
crucial for fostering a cultural shift toward
recognizing humanity’s dependence upon
the biosphere.
In many ways, this idea of codifying the
value of nature has been 30 years in the
making. And it couldn’t have become real-
ity without Daily’s vision, says Qingfeng
Zhang, a senior director at the Asian
Development Bank, which now includes
a Natural Capital Lab inspired and sup-
ported by the Stanford project. This initia-
tive, which was launched in 2020, created
a platform for the bank to promote sus-
tainable finance with the help of tools that
Daily and her team developed.
“Gretchen’s work in the area of envi-
ronmental science and its implications
for public policy has been monumental,”
says Zhang. “Her InVEST model and GEP
concept are transforming the way gov-
ernments, corporations, and civil society
look at nature. We now have a tangible
economic basis to invest in protecting and
growing nature.”
Kathryn Miles is a journalist and the
author of five books including, most
recently, Trailed: One Woman's Quest
to Solve the Shenandoah Murders.
Discover what’s coming
next in technology.
Subscribe now • In depth reporting on AI, climate change, biotech & more
• Trusted insights you can’t find anywhere else
for access to: • Science & technology news shaping the future
• 6 print and digital issues a year
• Discounts on MIT Technology Review events

Scan this code to subscribe or

learn more at technologyreview.com/subscribe

Every year, we look for promising
technologies poised to have
a real impact on the world.
Here are the advances that we
think matter most right now.
Breakthrough
Technologies
2024
Illustrations By Aaron Denton
23
AI for everything
Generative-AI tools like ChatGPT
reached mass adoption in record
time, and reset the course of an
entire industry.
WHO
Google, Meta, Microsoft, OpenAI
WHEN
Now
When OpenAI launched a free web app
called ChatGPT in November 2022,
nobody knew what was coming. But that
low-key release changed everything.
By January, ChatGPT had become the
fastest-growing web app ever, offering
anyone with a browser access to one of the
most powerful neural networks ever built.
We were dazzled and disturbed.
And that was only the start. In February,
Microsoft and Google revealed rival plans
to combine chatbots with search—plans
that reimagined our daily interactions with
the internet.
Early demos weren’t great. Microsoft’s
Bing Chat went off the rails, quick to churn
out nonsense. Google’s Bard was caught
making a factual error in its promo pitch.
But the genie wasn’t going back in its bot-
tle, no matter how weird it was.
Microsoft and Google have since moved
beyond search to put chatbot-based assis-
tants into the hands of billions of people
via their office software. The tech prom-
ises to summarize emails and meetings;
draft reports and replies; generate whole
slide decks—titles, bullet points, and pic-
tures—in seconds.
Microsoft and Meta released image-
making models that let users generate
shareable images of anything with a click.
Cue a nonstop stream of zany mash-ups—
and dozens of posts about Mickey Mouse
and SpongeBob SquarePants flying a plane
into the Twin Towers.
Google’s new phones now use AI to
let you edit photos to a degree never seen
before, exchanging sad faces for happy
ones and overcast afternoons for perfect
sunsets.
24
because not enough people had a strong
reason to leave, or a place to go if they did.
Never has such radical new technology
gone from experimental prototype to con-
sumer product so fast and at such scale.
What’s clear is that we haven’t even begun
to make sense of it all, let alone reckon
with its impact.
Is the shine coming off? Maybe. With
each release, the astonishing becomes
more mundane. But 2023’s legacy is clear:
billions have now looked AI in the face.
Now we need to figure out exactly what’s
looking back. —Will Douglas Heaven
The first gene-
editing treatment
Sickle-cell disease is the first illness
to be beaten by CRISPR as the
technology reaches the market.
WHO
CRISPR Therapeutics, Editas Medicine,
Precision BioSciences, Vertex Pharmaceuticals
WHEN
Now
The first gene-editing cure has arrived.
Grateful patients are calling it “life
changing.”
It was only 11 years ago that scientists
first developed the potent DNA-snipping
technology called CRISPR. Now they’ve
brought CRISPR out of the lab and into
real medicine with a treatment that cures
the symptoms of sickle-cell disease.
Sickle-cell is caused by inheriting two
bad copies of one of the genes that make
hemoglobin. Symptoms include bouts of
intense pain, and life expectancy with
the disease is just 53 years. It affects 1 in
4,000 people in the US, nearly all of them
African-American.
So how did this disease become
CRISPR’s first success? A fortuitous fact
of biology is part of the answer. Our bod-
ies harbor another way to make hemo-
globin that turns off when we’re born.
Researchers found that a simple DNA
edit to cells from the bone marrow could
turn it back on.
A real Twitter competitor never broke out
Now they have both.
Many CRISPR treatments are in trials,
but in 2022, Vertex Pharmaceuticals, based
in Boston, was first to bring one to regula-
tors for approval. That treatment was for
sickle-cell. After their bone marrow was
edited, nearly all the patients who volun-
teered in the trial were pain free.
Good news. But the expected price
tag of the gene-editing treatment is $2
to $3 million. And Vertex has no imme-
diate plans to offer it in Africa—where
sickle-cell disease is most common, and
where it still kills children.
The company says this is because
the treatment regimen is so complex. It
involves a hospital stay; doctors remove
the bone marrow, edit the cells, and then
transplant them back. In countries that
still struggle to cover basic health needs,
the procedure remains too demanding. So
simpler, cheaper ways to deliver CRISPR
could come next. —Antonio Regalado
Heat pumps
Heat pumps are a well-established
technology. Now they’re starting
to make real progress on
decarbonizing homes, buildings,
and even manufacturing.
WHO
Daikin, Mitsubishi, Viessmann
WHEN
Now
We’ve entered the era of the heat pump.
Heat pumps are appliances that can cool
and heat spaces using electricity. Many
buildings today are still heated with fossil
fuels, specifically natural gas. Switching to
electric heat pumps that run on renewable
energy could help homes, offices, and even
manufacturing facilities cut their emissions
dramatically.
While heat pumps have been used in
buildings since the mid-20th century, the
technology is breaking through in a new
way. Global sales of heat pumps grew by
11% in 2022, the second consecutive year of
double-digit growth, though that rate may
have slowed in 2023. Europe saw the most
dramatic shift, with a 40% growth in heat
pump installations through 2022, largely
driven by the energy crisis stemming from
the Russia-Ukraine war and by efforts to
move away from natural gas.
Asia is another hot spot, with China lead-
ing global installations and China and Japan
together accounting for more than half of
new patents filed on heat pump technology
since 2010. New approaches are enabling
heat pumps to reach higher temperatures,
which could allow the technology to help
clean up industrial manufacturing by sup-
plying power to generate steam used in
food processing and paper making.
In total, heat pumps have the potential
to cut global emissions by 500 million
tons in 2030—as much as pulling all cars
in Europe today off the roads. That would
require the total number of heat pumps
installed to reach about 600 million by
the end of the decade. (That’s about 20%
of the heating needs for all the world’s
buildings.)
There are still big challenges ahead for
heat pumps, including ramping produc-
tion to meet rising demand and ensuring
that the electrical grid is robust enough
to supply electricity to these and other
climate-focused technologies. But all signs
indicate that heat pumps are entering their
heyday. —Casey Crownhart
Twitter killers
Millions of people have fled the “bird
site” and flocked to decentralized
social media services instead.
WHO
Bluesky, Discord, Mastodon, Nostr, Threads
WHEN
Now
For the better part of 17 years, the roiling,
rolling, fractious, sometimes funny, some-
times horrifying, never-ever-ending global
conversation had a central home: Twitter.
If you wanted to know what was happen-
ing and what people were talking about
right now, it was the only game in town.
But then Elon Musk purchased Twitter,
renamed it X, fired most of its employees,
and more or less eliminated its modera-
tion and verification systems. He put in
place a new financial structure that incen-
tivized creators to spread and amplify
lies and propaganda. Many people have
begun casting about for a replacement
service—ideally one that is beyond any
individual’s control.
Decentralized, or federated, social
media allows for communication across
independently hosted servers or plat-
forms, using networking protocols such
as ActivityPub, AT Protocol, or Nostr. It
offers more granular moderation, more
security against the whims of a corporate
master or government censor, and the
opportunity to control your social graph.
It’s even possible to move from one server
to another and follow the same people.
To be sure, the dream of a decentral-
ized Twitter-like service has been around
for years. History is littered with failed
attempts—most notably App.net and
Identi.ca. A real competitor never broke
out because not enough people had a
strong reason to leave Twitter, or a place
to go if they did. Now they have both.
According to Similarweb, X’s traffic is
down by nearly 20%, year over year. Another
study, by Apptopia, found that the number
of daily active users went from 141 million
to 120 million. Meanwhile, decentralized
services like Mastodon, Bluesky, and some
Nostr clients have surged in popularity. But
it’s Threads, from Meta, that’s been the big
winner. Meta disclosed in September that
Threads already had nearly 100 million
monthly users. (As of press time, Threads
has not yet implemented ActivityPub,
but it promises to do so.) Nerd favorite
Mastodon is a distant second at 1.5 million
active users but is growing, while the still
invite-only Bluesky, which runs on the AT
Protocol, is at 2 million.
And of course, the real Twitter killer?
That’s Elon Musk. —Mat Honan
 Geothermal heat, an abundant and carbon-
free energy source, offers an alternative
Enhanced to fossil fuels that doesn’t vary with the
geothermal weather or time of day. However, conven-
tional geothermal plants require specific
systems geological conditions—in particular, per-
meable rocks with water sources.
This advanced drilling technology Because of this, geothermal accounts for
could unlock the potential of less than 1% of global renewables capac-
geothermal energy in many more ity. But an emerging technology could let
places. us exploit even more of the heat beneath
our feet.
Enhanced geothermal systems have
WHO
AltaRock Energy, Fervo Energy,
Utah FORGE lab been in development since the 1970s.
Recent advances show that they could
WHEN
3 to 5 years dramatically increase production of renew-
able energy. Fervo Energy tested one such
system this year in Nevada and proved its
commercial viability. The company is build-
ing another project in Utah, with a goal of
providing constant, clean power by 2026.
With enhanced geothermal, compa-
nies can access geothermal heat in new
locations. Hydraulic fracturing tech-
niques—widely used by the oil and gas
industry—are now being used to crack
open relatively solid rocks, at depths much
greater than existing geothermal wells.
Water is then injected into these rocks to
generate steam, which subsequently drives
turbines to produce electricity.
Fervo also looks to use enhanced geo-
thermal techniques to create what are
essentially giant underground batteries
Geothermal accounts for less than
1% of global renewables capacity,
but an emerging technology could
let us exploit even more of the heat
beneath our feet.
for the grid. By building up or relieving
pressure in the wells, it can save up energy
when demand is low and boost generation
when it rises.
The technology is not without potential
risks. In particular, the scientific commu-
nity is divided on how hydraulic fracturing
could affect seismic activity. While some
believe earthquake risks are minimal, a
2017 incident in South Korea was linked
to an enhanced geothermal project.
Several other companies and labs are
now advancing pilot projects and research
in the field. Washington-based AltaRock
Energy is developing specialized techniques
to access extremely hot rocks, which could
dramatically increase energy output. Utah
FORGE, sponsored by the US Department
of Energy, is drilling a well that can act as
a test bed for enhanced geothermal tech-
nologies. Many of these projects are still
in the experimental phase, but it’s increas-
ingly clear that enhanced geothermal is a
hot topic in the energy world. —June Kim
Weight-loss drugs
Weight-loss drugs are wildly popular
and effective, but their long-term
health impacts remain unknown.
WHO
Eli Lilly, Novartis, Novo Nordisk, Pfizer,
Viking Therapeutics
WHEN
Now
One-third of US adults have obesity, a
condition that makes them more suscep-
tible to heart disease, diabetes, and cancer.
Anti-obesity drugs—including Wegovy
and Mounjaro—could help address this
public health crisis. Success stories are
everywhere online, from Reddit to TikTok.
Novo Nordisk, the company behind two of
the popular medications, has seen profits
soar, and pharmacies have struggled to
keep the drugs in stock.
These medications help people lose
weight by suppressing their appetite. Most
were originally developed to treat type 2
diabetes, but in June 2021, Wegovy became
the first drug to be approved for weight
management since 2014. Semaglutide,
the active ingredient in both Wegovy and
Ozempic (a diabetes drug that’s often pre-
scribed off-label for weight loss), mimics a
hormone that the intestine releases after
eating, causing you to feel full. Patients
inject the drugs once a week at home and
can lose about 12% to 15% of their body
weight (though many hit a plateau after that).
These drugs aren’t perfect—common
side effects include nausea, diarrhea, and
vomiting. Many patients must stay on
the drugs for life to keep the weight off,
and the long-term impacts of these treat-
ments remain unknown. The drugs are also
expensive, costing more than $1,000 per
month, and most insurance plans don’t
cover them for weight loss.
Nevertheless, the treatments could
improve the health of millions of people.
Some studies even suggest that they alle-
viate symptoms of heart failure. Dozens
of companies are now developing new
versions of these weight-loss medications,
some of which can be taken orally.
In November, the US Food and Drug
Administration approved Eli Lilly’s dia-
betes drug Zepbound for obesity. With
about 70 new obesity treatments in devel-
opment, six are now awaiting regulatory
review. In the coming year, expect to see
more companies entering the final stages
of trials and seeking approval as demand
skyrockets. —Abdullahi Tsanni
Chiplets
Chipmakers are betting that smaller,
more specialized chips can extend
the life of Moore’s Law.
WHO
Advanced Micro Devices, Intel,
Universal Chiplet Interconnect Express
WHEN
Now
Packaging. It may sound boring, but it’s
an essential part of building computer
27
systems. Now companies are defining
what that looks like for a new generation
of machines.
For decades, chipmakers have improved
performance by making transistors smaller
and cramming more of them onto chips.
The popular name for the trend is Moore’s
Law. But that era is ending. It’s gotten
immensely expensive to further shrink
transistors and manufacture the complex
chips that today’s high-tech industries
demand.
In response, manufacturers are turning
to smaller, more modular “chiplets” that
are designed for specific functions (such
as storing data or processing signals) and
can be linked together to build a system.
The smaller a chip, the fewer defects it’s
likely to contain, making manufacturing
less expensive.
Companies including Advanced Micro
Devices and Intel have been marketing
systems based on chiplets for years. But
whether chiplets can help the industry
maintain performance gains at the pace
of Moore’s Law will depend on pack-
aging, which entails placing them side
by side or stacking them, forming fast,
high-bandwidth electrical connections
between them, and encasing them in
protective plastic.
Manufacturers are still working out
the best way to balance cost with per-
formance. The $52.7 billion CHIPS Act,
the 2022 US legislation aimed at shoring
up the nation’s chip industry, directs $11
billion toward “advanced semiconductor”
research and creates a National Advanced
Packaging Manufacturing Program to
foster collaboration between academia
and industry.
So far, chiplet adoption has been
hindered by the lack of technical stan-
dards for packaging. That’s changing:
the industry has embraced an open-
source standard called Universal Chiplet
Interconnect Express. In theory, stan-
dards will make it easy to combine chip-
lets made by different companies, which
could give chipmakers more freedom in
fast-moving fields like AI, aerospace, and
automaking. —Mike Orcutt
28
Super-efficient
solar cells
Solar cells that combine
traditional silicon with cutting-
edge perovskites could push
the efficiency of solar panels to
new heights.
WHO
Beyond Silicon, Caelux, First Solar, Hanwha
Q Cells, Oxford PV, Swift Solar, Tandem PV
WHEN
3 to 5 years
In November 2023, a buzzy solar tech-
nology broke yet another world record for
efficiency. The previous record had existed
for only about five months—and it likely
won’t be long before it too is obsolete.
This astonishing acceleration in efficiency
gains comes from a special breed of next-
generation solar technology: perovskite
tandem solar cells. These cells layer the
traditional silicon with materials that share
a unique crystal structure.
In the decade that scientists have been
toying with perovskite solar technology, it
has continued to best its own efficiency
records, which measure how much of the
sunlight that hits the cell is converted into
electricity. Perovskites absorb different
wavelengths of light from those absorbed
by silicon cells, which account for 95% of
the solar market today. When silicon and
perovskites work together in tandem solar
cells, they can utilize more of the solar spec-
trum, producing more electricity per cell.
Technical efficiency levels for silicon-
based cells top out below 30%, while
perovskite-only cells have reached exper-
imental efficiencies of around 26%. But per-
ovskite tandem cells have already exceeded
33% efficiency in the lab. That is the tech-
nology’s tantalizing promise: if deployed
on a significant scale, perovskite tandem
cells could produce more electricity than
the legacy solar cells at a lower cost.
But perovskites have stumbled when
it comes to actual deployment. Silicon
Essentially, the Frontier
supercomputer can perform
as many calculations in one second
as 100,000 laptops.
solar cells can last for decades. Few per-
ovskite tandem panels have even been
tested outside.
The electrochemical makeup of per-
ovskites means they’re sensitive to sucking
up water and degrading in heat, though
researchers have been working to create
better barriers around panels and shifting
to more stable perovskite compounds.
In May, UK-based Oxford PV said it
had reached an efficiency of 28.6% for a
commercial-size perovskite tandem cell,
which is significantly larger than those
used to test the materials in the lab, and
it plans to deliver its first panels and ramp
up manufacturing in 2024. Other com-
panies could unveil products later this
decade. —Emma Foehringer Merchant
Apple Vision Pro
Micro-OLED technology has been
in development for more than a
decade, but the Vision Pro will be
the highest-profile demonstration
of its abilities to date.
WHO
Apple
WHEN
2024
History is littered with doomed face com-
puters. Google Glass, Microsoft HoloLens,
and even Meta’s Quest line all failed to
catch fire. Now, it’s Apple’s turn to try.
Later this year, Apple plans to start
shipping its new Vision Pro, the company’s
first mixed-reality headset. Its commercial
success is very much an open question, but
the Vision Pro is no doubt a breakthrough
device, with a display radically better than
any that has come before.
Apple revealed the headset (which it
calls a spatial computer) in June at its annual
developers’ event, pitching it as a better way
to watch films, experience photos, connect
with others, and even read and create.
Unlike virtual reality, mixed reality over-
lays digital content onto your real-world
surroundings. Cameras and sensors embed-
ded in the headset take what’s going on
around you in a room and re-create it on
a pair of screens—one for each eye. Then
the headset adds in whatever digital con-
tent you’ve selected—a surfing video, for
example, virtually projected right in front
of your actual couch.
The Vision Pro’s twin micro-OLED dis-
plays provide much higher resolution and
sharper contrast than the liquid crystal dis-
plays found in most VR headsets—which
means in addition to delivering mixed real-
ity, the headset can also provide the most
fully immersive experience we’ve ever seen.
One big remaining question: What
will people use it for? Apple has a suite
of apps at the ready, and developers can
create their own, but will it mostly be a
fancy video viewer? Another: Will people
actually wear it? Apple added virtual eyes
to the front to show where the wearer is
looking, but at the June event it only shared
videos of people wearing the headsets. How
good or weird will those virtual eyes look
in person? And at $3,499, the Vision Pro
will cost thousands of dollars more than
other headsets.
That said, no other company has been
as successful at developing and marketing
consumer technology over the years. Apple’s
investment may give mixed reality its best
chance yet to catch on. —Amy Nordrum
Exascale
computers
Computers capable of crunching
a quintillion operations per second
are expanding the limits of what
scientists can simulate.
WHO
Oak Ridge National Lab, Jülich Supercomputing
Centre, China’s Supercomputing Center in Wuxi
WHEN
Now
In May 2022, the global supercomputer
rankings were shaken up by the launch of
Frontier. Now the fastest supercomputer
in the world, it can perform more than 1
quintillion (1018) floating-point operations
per second. That’s a 1 followed by 18 zeros,
also known as an exaflop. Essentially,
Frontier can perform as many calculations
in one second as 100,000 laptops.
With the launch of Frontier, located
at Oak Ridge National Laboratory in
Tennessee, the era of exascale comput-
ing officially began. Several more such
exascale computers will soon join its
ranks. In the US, researchers are install-
ing two machines that will be about
twice as fast as Frontier: El Capitan, at
Lawrence Livermore National Laboratory
in California, and Aurora, at Argonne
National Laboratory in Illinois. Europe’s
first exascale supercomputer, Jupiter, is
expected to come online in late 2024.
China reportedly also has exascale
machines, although it has not released
results from standard benchmark tests.
Scientists and engineers are eager to
use these turbocharged computers to
advance a range of fields. Astrophysicists
are already using Frontier to model the
flow of gas in and out of the Milky Way;
in addition to simulating motion on the
scale of our galaxy, their model can zero
in on exploding stars. This application
showcases supercomputers’ unique ability
to simulate physical objects at multiple
scales simultaneously.
The progress won’t stop here. For the
last three decades, supercomputers have
gotten about 10 times faster every four
years or so. And the stewards of these
machines are already planning the next
models: Oak Ridge engineers are design-
ing a supercomputer that will be three to
five times faster than Frontier, likely to be
unveiled in the coming decade.
But one big challenge looms: the energy
footprint. Frontier, which already employs
energy-conserving innovations, draws
enough power even while idling to run
thousands of homes. Engineers will need
to figure out how to build these behemoths
not just for speed, but for environmental
sustainability. —Sophia Chen
 31

It was a stranger who first brought home for me how

big last year’s vibe shift was going to be. We were wait-
We’re still riding the
ing for a stuck elevator together one March morning,
generative-AI wave,
and she told me she had just used ChatGPT to help her
but the technology’s
write a report for her marketing job. She hated writing
future—and ours—
reports because she didn’t think she was very good at
will be shaped by what
it. But this time her manager had praised her. Did it
we do next.
feel like cheating? Hell no, she said. You do what you
can to keep up.
That stranger’s experience of generative AI is one

6
among millions. People in the street (and in elevators)
are now figuring out what this radical new technol-
ogy is for and wondering what it can do for them. In
many ways the buzz around generative AI right now
recalls the early days of the internet: there’s a sense of
excitement and expectancy—and a feeling that we’re
making it up as we go.
That is to say, we’re in the dot-com boom, circa
2000. Many companies will go bust. It may take years
before we see this era’s Facebook (now Meta), Twitter
(now X), or TikTok emerge. “People are reluctant to
imagine what could be the future in 10 years, because
no one wants to look foolish,” says Alison Smith, head
big of generative AI at Booz Allen Hamilton, a technology
consulting firm. “But I think it’s going to be something
questions wildly beyond our expectations.”
The internet changed everything—how we work and
for play, how we spend time with friends and family, how

generative we learn, how we consume, how we fall in love, and

so much more. But it also brought us cyber-bullying,
revenge porn, and troll factories. It facilitated geno-

AI
cide, fueled mental-health crises, and made surveil-
lance capitalism—with its addictive algorithms and
predatory advertising—the dominant market force of
our time. These downsides became clear only when
people started using it in vast numbers and killer apps
like social media arrived.
Generative AI is likely to be the same. With the
infrastructure in place—the base generative models
from OpenAI, Google, Meta, and a handful of others—
people other than the ones who built it will start using
and misusing it in ways its makers never dreamed of.
“We’re not going to fully understand the potential and
the risks without having individual users really play
around with it,” says Smith.
By Generative AI was trained on the internet and so
Will Douglas Heaven has inherited many of its unsolved issues, including

Illustrations
by Selman Design
32

those related to bias, misinformation, copyright

infringement, human rights abuses, and all-round
economic upheaval. But we’re not going in blind.
Here are six unresolved questions that we need
to bear in mind as we watch the generative-AI rev-
olution unfold. This time around, we have a chance
to do better.

Will we ever mitigate the bias problem?

1
Bias has become a byword for AI-related harms,
for good reason. Real-world data, especially
text and images scraped from the internet,
is riddled with it, from gender stereotypes to
racial discrimination. Models trained on that
data encode those biases and then reinforce them
wherever they are used.
Chatbots and image generators tend to portray
engineers as white and male and nurses as white
and female. Black people risk being misidentified
by police departments’ facial recognition programs,
leading to wrongful arrest. Hiring algorithms favor
men over women, entrenching a bias they were some-
times brought in to address.
Without new data sets or a new way to train mod-
els (both of which could take years of work), the root
cause of the bias problem is here to stay. But that
hasn’t stopped it from being a hot topic of research.
OpenAI has worked to make its large language models
less biased using techniques such as reinforcement
learning from human feedback (RLHF). This steers
the output of a model toward the kind of text that
human testers say they prefer.
Other techniques involve using synthetic data sets.
For example, Runway, a startup that makes generative
models for video production, has trained a version of
the popular image-making model Stable Diffusion on
synthetic data such as AI-generated images of people
who vary in ethnicity, gender, profession, and age.
The company reports that models trained on this data
set generate more images of people with darker skin
and more images of women. Request an image of a
businessperson, and outputs now include women in
headscarves; images of doctors will depict people
who are diverse in skin color and gender; and so on.
Critics dismiss these solutions as Band-Aids on
broken base models, hiding rather than fixing the

problem. But Geoff Schaefer, a colleague of Smith’s
at Booz Allen Hamilton who is head of responsible
AI at the firm, argues that such algorithmic biases
can expose societal biases in a way that’s useful in
the long run.
As an example, he notes that even when explicit
information about race is removed from a data set,
racial bias can still skew data-driven decision-making
because race can be inferred from people’s addresses—
revealing patterns of segregation and housing discrim-
ination. “We got a bunch of data together in one place,
and that correlation became really clear,” he says.
Schaefer thinks something similar could happen
with this generation of AI: “These biases across soci-
ety are going to pop out.” And that will lead to more
targeted policymaking, he says.
But many would balk at such optimism. Just because
a problem is out in the open doesn’t guarantee it’s
going to get fixed. Policymakers are still trying to
address social biases that were exposed years ago—
in housing, hiring, loans, policing, and more. In the
meantime, individuals live with the consequences.
Prediction ---------------------------------
Bias will continue to be an inherent feature
of most generative AI models. But workarounds
and rising awareness could help policymakers
address the most obvious examples.
How will AI change the way we
apply copyright?
2
Outraged that tech companies should
profit from their work without consent,
artists and writers (and coders) have
launched class action lawsuits against
OpenAI, Microsoft, and others, claiming
copyright infringement. Getty is suing Stability AI,
the firm behind the image maker Stable Diffusion.
These cases are a big deal. Celebrity claimants
such as Sarah Silverman and George R.R. Martin
have drawn media attention. And the cases are set
to rewrite the rules around what does and does not
count as fair use of another’s work, at least in the US.
But don’t hold your breath. It will be years before the
courts make their final decisions, says Katie Gardner,
a partner specializing in intellectual-property licens-
ing at the law firm Gunderson Dettmer, which rep-
resents more than 280 AI companies. By that point,
33
she says, “the technology will be so entrenched in the
economy that it’s not going to be undone.”
In the meantime, the tech industry is building
on these alleged infringements at breakneck pace.
“I don’t expect companies will wait and see,” says
Gardner. “There may be some legal risks, but there
are so many other risks with not keeping up.”
Some companies have taken steps to limit the
possibility of infringement. OpenAI and Meta claim
to have introduced ways for creators to remove their
work from future data sets. OpenAI now prevents
users of DALL-E from requesting images in the style of
living artists. But, Gardner says, “these are all actions
to bolster their arguments in the litigation.”
Google, Microsoft, and OpenAI now offer to pro-
tect users of their models from potential legal action.
Microsoft’s indemnification policy for its generative
coding assistant GitHub Copilot, which is the sub-
ject of a class action lawsuit on behalf of software
developers whose code it was trained on, would in
principle protect those who use it while the courts
shake things out. “We’ll take that burden on so the
users of our products don’t have to worry about it,”
Microsoft CEO Satya Nadella told MIT Technology
Review. At the same time, new kinds of licensing
deals are popping up. Shutterstock has signed a six-
year deal with OpenAI for the use of its images. And
Adobe claims its own image-making model, called
Firefly, was trained only on licensed images, images
from its Adobe Stock data set, or images no longer
under copyright. Some contributors to Adobe Stock,
however, say they weren’t consulted and aren’t happy
about it.
Resentment is fierce. Now artists are fighting
back with technology of their own. One tool, called
Nightshade, lets users alter images in ways that
are imperceptible to humans but devastating to
machine-learning models, making them miscate-
gorize images during training. Expect a big realign-
ment of norms around sharing and repurposing
media online.
Prediction ---------------------------------
High-profile lawsuits will continue to draw
attention, but that’s unlikely to stop compa-
nies from building on generative models. New
marketplaces will spring up around ethical
data sets, and a cat-and-mouse game between
companies and creators will develop.
34
How will it change our jobs?
3
We’ve long heard that AI is coming for
our jobs. One difference this time is that
white-collar workers—data analysts, doc-
tors, lawyers, and (gulp) journalists—look
to be at risk too. Chatbots can ace high
school tests, professional medical licensing examina-
tions, and the bar exam. They can summarize meet-
ings and even write basic news articles. What’s left for
the rest of us? The truth is far from straightforward.
Many researchers deny that the performance of
large language models is evidence of true smarts. But
even if it were, there is a lot more to most professional
roles than the tasks those models can do.
Last summer, Ethan Mollick, who studies inno-
vation at the Wharton School of the University of
Pennsylvania, helped run an experiment with the
Boston Consulting Group to look at the impact of
ChatGPT on consultants. The team gave hundreds of
consultants 18 tasks related to a fictional shoe company,
such as “Propose at least 10 ideas for a new shoe tar-
geting an underserved market or sport” and “Segment
the footwear industry market based on users.” Some of
the group used ChatGPT to help them; some didn’t.
The results were striking: “Consultants using
ChatGPT-4 outperformed those who did not, by a lot.
On every dimension. Every way we measured perfor-
mance,” Mollick writes in a blog post about the study.
Many businesses are already using large language
models to find and fetch information, says Nathan
Benaich, founder of the VC firm Air Street Capital
and leader of the team behind the State of AI Report,
a comprehensive annual summary of research and
industry trends. He finds that welcome: “Hopefully,
analysts will just become an AI model,” he says. “This
stuff’s mostly a big pain in the ass.”
His point is that handing over grunt work to
machines lets people focus on more fulfilling parts
of their jobs. The tech also seems to level out skills
across a workforce: early studies, like Mollick’s with
consultants and others with coders, suggest that less
experienced people get a bigger boost from using AI.
(There are caveats, though. Mollick found that people
who relied too much on GPT-4 got careless and were
less likely to catch errors when the model made them.)
Generative AI won’t just change desk jobs. Image-
and video-making models could make it possible to
produce endless streams of pictures and film without
human illustrators, camera operators, or actors. The
strikes by writers and actors in the US in 2023 made
it clear that this will be a flashpoint for years to come.
Even so, many researchers see this technology as
empowering, not replacing, workers overall. Technology
has been coming for jobs since the industrial revolution,
after all. New jobs get created as old ones die out. “I
feel really strongly that it is a net positive,” says Smith.
But change is always painful, and net gains can hide
individual losses. Technological upheaval also tends
to concentrate wealth and power, fueling inequality.
“In my mind, the question is no longer about
whether AI is going to reshape work, but what we
want that to mean,” writes Mollick.
Prediction --------------------------------
Fears of mass job losses will prove exaggerat-
ed. But generative tools will continue to pro-
liferate in the workplace. Roles may change;
new skills may need to be learned.
What misinformation will it
make possible?
4
Three of the most viral images of 2023 were
photos of the pope wearing a Balenciaga
puffy, Donald Trump being wrestled to
the ground by cops, and an explosion at
the Pentagon. All fake; all seen and shared
by millions of people.
These are three of the
most viral images of Using generative models to create fake text or
2023. All fake; all seen images is easier than ever. Many warn of a misinfor-
and shared by millions mation overload. OpenAI has collaborated on research
of people. that highlights many potential misuses of its tech for
fake-news campaigns. In a 2023 report it warned that
large language models could be used to produce more
persuasive propaganda—harder to detect as such—
FAKE at massive scales. Experts in the US and the EU are
already saying that elections are at risk.
It was no surprise that the Biden administration
made labeling and detection of AI-generated content
a focus of its executive order on artificial intelligence
in October. But the order fell short of legally requiring
tool makers to label text or images as the creations of
FAKE an AI. And the best detection tools don’t yet work well
enough to be trusted.
The US has said it will audit any AI that might
pose threats to national security, including election
FAKE
 35

interference. It’s a great step, says Benaich. But even

Here’s the
the developers of these models don’t know their full
capabilities: “The idea that governments or other inde-
pendent bodies could force companies to fully test

catch:
their models before they’re released seems unrealistic.”
Here’s the catch: it’s impossible to know all the
ways a technology will be misused until it is used.
“In 2023 there was a lot of discussion about slowing
down the development of AI,” says Schaefer. “But we

it is take the opposite view.”

Unless these tools get used by as many people in
as many different ways as possible, we’re not going

impossible
to make them better, he says: “We’re not going to
understand the nuanced ways that these weird risks
will manifest or what events will trigger them.”

to know all
Prediction --------------------------------
New forms of misuse will continue to surface
as use ramps up. There will be a few standout
examples, possibly involving electoral manip-

the ways a
ulation.

Will we come to grips with its costs?

technology
5
The development costs of generative AI,
both human and environmental, are also
to be reckoned with. The invisible-worker

will be problem is an open secret: we are spared

the worst of what generative models can
produce thanks in part to crowds of hidden (often

misused
poorly paid) laborers who tag training data and weed
out toxic, sometimes traumatic, output during testing.
These are the sweatshops of the data age.
In 2023, OpenAI’s use of workers in Kenya came
under scrutiny by popular media outlets such as

until it is Time and the Wall Street Journal. OpenAI wanted to

improve its generative models by building a filter that
would hide hateful, obscene, and otherwise offensive

used. content from users. But to do that it needed people

to find and label a large number of examples of such
toxic content so that its automatic filter could learn
to spot them. OpenAI had hired the outsourcing firm
Sama, which in turn is alleged to have used low-paid
workers in Kenya who were given little support.
With generative AI now a mainstream concern,
the human costs will come into sharper focus, put-
ting pressure on companies building these models to
address the labor conditions of workers around the
world who are contracted to help improve their tech.
36

The other great cost, the amount of energy required

to train large generative models, is set to climb before
the situation gets better. In August, Nvidia announced
Q2 2024 earnings of more than $13.5 billion, twice
as much as the same period the year before. The
bulk of that revenue ($10.3 billion) comes from data
centers—in other words, other firms using Nvidia’s
hardware to train AI models.
“The demand is pretty extraordinary,” says Nvidia
CEO Jensen Huang. “We’re at liftoff for generative AI.”
He acknowledges the energy problem and predicts
that the boom could even drive a change in the type
of computing hardware deployed. “The vast majority
of the world’s computing infrastructure will have to
be energy efficient,” he says.
Prediction --------------------------------
Greater public awareness of the labor and
environmental costs of AI will put pressure
on Big Tech. But don’t expect significant
improvement on either front soon.

Will doomerism continue to

dominate policymaking?

6
Doomerism—the fear that the creation of
smart machines could have disastrous, even
apocalyptic consequences—has long been
an undercurrent in AI. But peak hype, plus
a high-profile announcement from AI pio-
neer Geoffrey Hinton in May that he was now scared
of the tech he helped build, brought it to the surface.
Few issues in 2023 were as divisive. AI luminar-
ies like Hinton and fellow Turing Award winner Yann
LeCun, who founded Meta’s AI lab and who finds
doomerism preposterous, engage in public spats,
throwing shade at each other on social media.
Hinton, OpenAI CEO Sam Altman, and others
have suggested that (future) AI systems should have
safeguards similar to those used for nuclear weapons.
Such talk gets people’s attention. But in an article he
co-wrote in Vox in July, Matt Korda, project manager
for the Nuclear Information Project at the Federation
of American Scientists, decried these “muddled analo-
gies” and the “calorie-free media panic” they provoke.
It’s hard to understand what’s real and what’s not
because we don’t know the incentives of the people
raising alarms, says Benaich: “It does seem bizarre
that many people are getting extremely wealthy off

the back of this stuff, and a lot of the people are the
same ones who are mandating for greater control.
It’s like, ‘Hey, I’ve invented something that’s really
powerful! It has a lot of risks, but I have the antidote.’”
Some worry about the impact of all this fear-
mongering. On X, deep-learning pioneer Andrew
Ng wrote: “My greatest fear for the future of AI is if
overhyped risks (such as human extinction) lets tech
lobbyists get enacted stifling regulations that suppress
open-source and crush innovation.” The debate also
channels resources and researchers away from more
immediate risks, such as bias, job upheavals, and mis-
information (see above).
“Some people push existential risk because they
think it will benefit their own company,” says François
Chollet, an influential AI researcher at Google. “Talking
about existential risk both highlights how ethically
aware and responsible you are and distracts from
more realistic and pressing issues.”
Benaich points out that some of the people ringing
the alarm with one hand are raising $100 million for
their companies with the other. “You could say that
doomerism is a fundraising strategy,” he says.
Prediction --------------------------------
The fearmongering will die down, but the
influence on policymakers’ agendas may be
felt for some time. Calls to refocus on more
immediate harms will continue.
Still missing: AI’s killer app
It’s strange to think that ChatGPT almost didn’t
happen. Before its launch in November 2022, Ilya
Sutskever, cofounder and chief scientist at OpenAI,
wasn’t impressed by its accuracy. Others in the com-
pany worried it wasn’t much of an advance. Under the
hood, ChatGPT was more remix than revolution. It
was driven by GPT-3.5, a large language model that
OpenAI had developed several months earlier. But
the chatbot rolled a handful of engaging tweaks—in
particular, responses that were more conversational
and more on point—into one accessible package. “It
was capable and convenient,” says Sutskever. “It was
the first time AI progress became visible to people
outside of AI.”
The hype kicked off by ChatGPT hasn’t yet run its
course. “AI is the only game in town,” says Sutskever.
“It’s the biggest thing in tech, and tech is the biggest
37
thing in the economy. And I think that we will con-
tinue to be surprised by what AI can do.”
But now that we’ve seen what AI can do, maybe the
immediate question is what it’s for. OpenAI built this
technology without a real use in mind. Here’s a thing,
the researchers seemed to say when they released
ChatGPT. Do what you want with it. Everyone has
been scrambling to figure out what that is since.
“I find ChatGPT useful,” says Sutskever. “I use
it quite regularly for all kinds of random things.” He
says he uses it to look up certain words, or to help him
express himself more clearly. Sometimes he uses it
to look up facts (even though it’s not always factual).
Other people at OpenAI use it for vacation planning
(“What are the top three diving spots in the world?”)
or coding tips or IT support.
Useful, but not game-changing. Most of those
examples can be done with existing tools, like search.
Meanwhile, staff inside Google are said to be hav-
ing doubts about the usefulness of the company’s
rival chatbot, Bard. “The biggest challenge I’m still
thinking of: what are LLMs truly useful for, in terms
of helpfulness?” Cathy Pearl, a user experience lead
for Bard, wrote on Discord in August, according to
Bloomberg. “Like really making a difference. TBD!”
Without a killer app, the “wow” effect ebbs away.
Stats from the investment firm Sequoia Capital show
that despite viral launches, AI apps like ChatGPT,
Character.ai, and Lensa, which lets users create styl-
ized (and sexist) avatars of themselves, lose users
faster than existing popular services like YouTube and
Instagram and TikTok.
“The laws of consumer tech still apply,” says Benaich.
“There will be a lot of experimentation, a lot of things
dead in the water after a couple of months of hype.”
Of course, the early days of the internet were also
littered with false starts. Before it changed the world,
the dot-com boom ended in bust. There’s always the
chance that today’s generative AI will fizzle out and
be eclipsed by the next big thing to come along.
Whatever happens, now that AI is fully in the
mainstream, niche concerns have become every-
one’s problem. As Schaefer says, “We’re going to be
forced to grapple with these issues in ways that we
haven’t before.”
Will Douglas Heaven is a senior editor for AI
at MIT Technology Review.

O U H AV E N ’ T S E E N H U N G RY U N T I L
YOU’VE SEEN BRAD LOWELL’S MICE.
A few years ago, Lowell—
a Harvard University neuro-
scientist—and a postdoc,
Mike Krashes, figured out
how to turn up the volume
on the drive for food as high
as it can go. They did it by stimulating
a bundle of neurons in the hypothal-
amus, an area of the brain thought
to play a key role in regulating our
basic needs.
A video captures what happened
next. Initially, the scene is calm as
a camera pans slowly along a series
of plastic cages, each occupied by
a docile, well-fed mouse, reclining
on a bed of wood chips. None of the
eight mice shown are interested in
the food pellets arrayed above them
on the other side of a triangular metal
grate that drops down from the ceil-
ing. Which is not surprising, since
each mouse has just consumed the
rodent equivalent of a Thanksgiving
dinner.
But as the seconds displayed on a
timer at the bottom of the screen tick
away, half the mice begin to stir—the
first evidence that a chemical agent
designed to turn on specific neurons
associated with appetite is reaching
its targets.
39
Soon, the mice seem possessed.
Some stand on their hind legs, thrust-
ing their noses through the grates
above them at the inaccessible pellets.
Others climb the walls, hang from the
bars of the grate, or dig frantically
through the wood chips.
“It looks like they’re losing their
minds,” Lowell says.
Lowell, who is one of the world’s
leading experts on the circuits in
the brain that control hunger, sati-
ety, and weight regulation, some-
times references this video to make a
point: When you’re starving, hunger
is like a demon. It awakens in the most
ancient and primitive parts of the
brain and then commandeers other
neural machinery to do its bidding
until it gets what it wants.
Scientists have spent decades trying to
unravel the intricate mysteries of the human
appetite. Are they on the verge of finally
determining how hunger works?
Illustration by Portraits by
Francesco Francavilla Tony Luong
40
What might begin as a small sensation
quickly spirals. Intrusive thoughts pulled
from our memory centers burst into our
consciousness. Images of meatball sand-
wiches. The smell of bread. The imagined
taste of a cork-like food pellet. The moti-
vational and emotional areas of our brain
infuse the need to eat with a nonverbal
imperative that feels so powerful it eclipses
all else. Our prefrontal cortex kicks into
gear, considering how we might obtain
food. (If we are in a dangerous situation
like a war zone, we weigh how much dan-
ger we are willing to risk to get it.) Then
we mobilize our sensory and motor areas.
We steal a chicken, attempt to spear a fish
in a pond, raid the work refrigerator, or
hurl our body against a metal grate, hop-
ing to get a taste of a food pellet.
So by exciting the hunger neurons in
those mice, Lowell catalyzed a storm of
neural activity that spread to the cerebral
cortex and other higher-order processing
centers, leading directly to a chain of com-
plex goal-directed behaviors (ineffective
though they turned out to be).
It also drove home for Lowell just how
much we still have to learn.
“Sure, we managed to have the brain say
‘Go eat,’” he says. “But that’s not really an
explanation. How does that actually work?”
To answer that question, Lowell has
teamed up with Mark Andermann, a neu-
roscientist who studies how motivation
shapes perception (and who also happens
to occupy the office next to his at Boston’s
Beth Israel Deaconess Medical Center).
Together they are following known parts of
the neural hunger circuits into uncharted
parts of the brain, in some cases activat-
ing one neuron at a time to methodically
trace new connections through areas so
primitive that we share them with lizards.
Their work could have important impli-
cations for public health. More than 1.9
billion adults worldwide are overweight
and more than 650 million are obese, a
condition correlated with a wide range of
chronic health conditions, including diabe-
tes, fatty liver disease, heart disease, and
some types of cancer. Understanding the
circuits involved could shed new light on
the factors that have caused those num- coined a term for the elusive bundle of brain
bers to skyrocket in recent years. cells he is seeking: “Holy Grail” neurons.
And it could also help solve the mystery It might sound like a tired scientific
behind a new class of weight-loss drugs trope. But for the understated Lowell, the
known as GLP-1 agonists. Many in the field term is perfectly apt: what he’s seeking gets
of public health are billing these drugs, at the very heart of human will. Finding
which include Wegovy and Ozempic, as it would be the culmination of decades of
transformative, providing the first effective work, and something he never imagined
method of combating obesity, and allowing would become possible in his lifetime.
some individuals to lose more than 15% of
their body weight. They’ve also become The hunger mystery
something of a cultural phenomenon; in Brad Lowell likes to joke that he is the
the last three months of 2022, US health- token local at Beth Israel Deaconess
care providers wrote more than 9 million Medical Center. Born in the hospital next
prescriptions for the drugs. Yet no one can door to where he now conducts research,
explain precisely how and why they work. he grew up 25 miles north in the town
Part of the reason is that scientists still of Boxford and attended the University
haven’t decoded the complex neural machin- of Massachusetts, Amherst, a couple of
ery involved in the control of appetite. hours’ drive away.
“The drugs are producing the good Soon after arriving at UMass as an
effects, the satiety effects, through some undergrad in the late 1970s, he was
aspect of this larger system,” says Lowell, accepted into the physiological psychology
who has watched their emergence with lab of Richard Gold, a pioneering neurosci-
surprise and genuine fascination. “One of entist who was working to identify neural
the most important components in figuring structures involved in regulating appetite.
out how they work is to define what the Gold’s focus was the hypothalamus—a
system is. And that is what we are doing.” primitive structure deep in the brain that
But the ultimate goal for Lowell and hasn’t changed much through evolution. It
Andermann is far loftier than simply is thought to be responsible for keeping the
reverse-engineering the way hunger body in “homeostasis” by monitoring and
works. The scientists
are searching for the elu-
sive bundle of neurons “Sure, we managed to have the brain
that allow our instinc-
tual urge to eat to com- say ‘Go eat,’” Lowell says.
mandeer higher-order “But that’s not really an explanation.
brain structures involved How does that actually work?”
in human motivation,
decision-making, mem-
ory, conscious thought,
and action. They believe identifying these balancing important functions like body
neurons will make it possible to study how temperature, blood pressure, our need for
a simple basic impulse—in this case, a food and water, and other basic drives.
signal from the body that energy stores Gold suspected that the paraventric-
are beginning to run low and need to be ular hypothalamic nucleus (PVH), a tiny
replenished—propagates through the brain patch of roughly 50,000 neurons in the
to dominate our conscious experience and hypothalamus, played a role in controlling
turn into something far more complex: a appetite. By today’s standards, the tools
series of complicated, often well-thought- to study it back then were “stone age”—
out actions designed to get food. Lowell says he used a “retracting wire
This quest has so consumed Lowell in knife” to sever bundles of neuronal pro-
recent years that his graduate students have jections that emanated from the PVH and

Neuroscientist Brad Lowell has
spent decades trying to understand
the brain circuitry that explains
hunger.
41
connected to neurons outside it—but they
were effective. When the anesthetized
rodents Lowell had operated on woke up,
they were crazed with hunger, and they
quickly became obese.
The experience made a lasting impres-
sion. Lowell, then an athletic 19-year-old
soccer aficionado, had always assumed
that anyone who was overweight was just
“lazy.” The experiment suggested there
was likely far more to it than that. It also
convinced Lowell to become a scientist.
But further research into how pre-
cisely the brain worked to control hun-
ger and satiety had reached something
of an impasse.
“Gold and a few other labs put the PVH
on the map as a site required to restrain
what you eat,” Lowell explains. “But they
didn’t have the tools to look any further.”
Figuring out which of the 50,000 neu-
rons in the PVH were actually important
to appetite, the ones that could essentially
mute the hunger switch, was a challenge
that seemed insurmountable—akin to, as
Lowell puts it, trying to untangle a “huge
bowl of spaghetti.”
“How do you differentiate one strand
of spaghetti from another? These being
neurons, right?” he asks. “There’s no way.
They all look the same.”
When Lowell opened his own lab at
Beth Israel Deaconess Medical Center
in the early 1990s, after earning an MD
and PhD at Boston University, he studied
metabolism in tissues like muscle, organs,
and fat that were connected to the brain
through the peripheral nervous system.
But his undergrad experience in Gold’s
lab nagged at him.
“The brain is the Lord of the Rings,”
Lowell says. “It’s the one ring that rules
them all. And it was not that interesting to
study these other things with the master
player up there.”
The entry point
Early in his career, Lowell envied his col-
leagues who studied vision. For decades,
neuroscientists had been able to trace the
neural circuits involved in that function
by shining light into the eyes of mice,
42

Mapping the hunger- 3

satiety circuit 1

How do subconscious signals make it

to the conscious parts of the brain? 2

identifying which neurons lit up, and then 1

following them to map out the relevant
brain circuits. Lowell and his peers who Hypothalamus
were interested in hunger had never had
Released by fat cells Released by
a similar entry point. in proportion to fat stores empty gut
That changed in 1994, when Jeffrey Insulin,
Friedman, a researcher at Rockefeller glucose,
HUNGER/SATIETY
University, gave Lowell and others a way Leptin Ghrelin and other
HORMONES
absorbed
to identify the first important neurons—or
nutrients
individual “strands of spaghetti”—involved

excites

excites
in
in hunger regulation. hi
bi
ts
Back in 1949, scientists at the Jackson
Laboratory in Bar Harbor, Maine, had HUB 1 The ARC
bred mice with an unidentified genetic Satiety- Hunger- contains
mutation that caused them to grow mas- The ARC producing
inhibits
producing neurons that
The arcuate neurons neurons sense energy
sively obese. They hypothesized that the
nucleus levels.
obesity stemmed from the mice’s inability
exc

its
to produce a crucial protein involved in
es

ib
it

weight regulation. inh

Decades later, Friedman was the first to receptor SATIETY The PVH
HUB 2
apply cutting-edge genetic technologies to SWITCH plays a big role
clone the DNA sequences that were abnor- The PVH Melanocortin
High activity in metabo-
mal in the obese mice; he then confirmed causes satiety; lism and other
The paraventricular neurons
low activity autonomic
that their obesity was caused by an inabil- hypothalamic nucleus
causes hunger. functions.
ity to produce a key hormone released by
fat cells, which the brain uses to track the
excites

body’s available energy stores. Friedman 2

purified the hormone and named it leptin.
He also identified the DNA sequence
Brain stem
needed to make the leptin “receptor”—the ?????????????????
specialized proteins that stick out of brain ? ?
? ?
cells involved in appetite regulation like HUB 3 ? ? The PBN is the key way
microscopic antennae, sensing whenever ? ? station to higher-order
The PBN ?
“Holy Grail”
? brain areas. It also receives
leptin is present and kicking off a chemical ? neurons ?
The parabrachial input about digestion
cascade that promotes a sense of satiety. nucleus
? ?
from the gut.
? ?
The discovery added further evidence ? ?
to the idea that obesity was biologically ?????????????????
determined, and more specifically to the Tens of thousands of unmapped neurons—
concept of a “set point” when it comes which ones are connected to appetite?

to weight—a predetermined weight, fat

mass, or other measurable physiological
characteristic that the body will defend.
Appetite is the means by which the body Subcortical structures involved in emotion and reward help this information
eventually reach the cortex via direct and indirect projections.
performs “error correction” and mobilizes
to devote energy and attention to the task
of restoring homeostasis.
A “cure” for obesity suddenly seemed 3
within reach. The biotech firm Amgen
licensed the rights to leptin for $20 mil-
Cortex
lion, hoping to develop a drug that could CONSCIOUS,
mimic its effects. But the drug it came up ACTION-ORIENTED
with had very little effect on most people ACTIVITY

with obesity, suggesting that leptin was
only part of the story—a hypothesis that
seemed to be confirmed when other labs
discovered additional hormones and sig-
nals that seemed to be involved in hunger.
Further experiments showed that many
obese humans in fact had normal or high
levels of leptin.
It stood to reason, then, that somewhere
in the brain leptin was being combined with
other signals related to available energy, and
that this information would then have to be
compared with a homeostatic “set point.”
This suggested a highly complex set of
neurological circuits involved in hunger
regulation. Understanding how this pro-
cess worked would require a detailed wir-
ing diagram that might explain how all the
parts fit together. And while Friedman’s
discoveries regarding leptin didn’t answer
all the questions, they provided the entry
point that Lowell and the rest of the field
had been waiting for, allowing them to
begin to draw such a map.
Following the path of leptin, scientists
in other labs found the hormone’s first tar-
get, and therefore the first important way
station in the hunger circuit: a specific
patch of neurons known as the arcuate
nucleus (ARC). Located at the base of the
hypothalamus, the ARC, we now know,
integrates information coming from other
brain structures, as well as circulating
nutrients and hormones like leptin and
insulin. All of these inputs convey key
information about the current state of the
body, such as the level of existing energy
stores and nutrient availability.
Determining how the ARC worked—
and where it sent information after taking it
in—was the next question facing the field.
By then, Lowell had abandoned studies on
peripheral systems and joined the hunt.
Switching hunger on and off
In 1997, the next part of the puzzle fell into
place after Roger Cone, then a researcher
at Oregon Health and Science University,
discovered a key part of the switch that
essentially turned hunger on and off.
He bred mice with a gene mutation that
interferes with another class of key signaling
proteins, called melanocortins. Mice with
this mutation more closely resembled obese
humans than did mice with leptin muta-
tions: their obesity set in relatively late,
and they had diabetes-causing levels of
insulin and glucose. This particular muta-
tion prevented key receptors from detect-
ing melanocortin hormones, which in turn
interfered with the feeling of satiety and
caused mice to continue to eat. But when
these melanocortin receptors were func-
tioning normally, detecting the presence
of the melanocortin hormones seemed to
turn down appetite. In essence, Cone had
found the brain’s “satiety switches.”
This discovery was critical in helping
scientists determine how leptin worked its
magic in the ARC, the first stop in the hun-
ger circuit. It turned out that when leptin
reached the ARC, it set off a biochemical
chain reaction that caused more melano-
cortin hormones to be released, eventually
activating these “satiety switches.”
But these satiety switches were not pres-
ent just in the ARC; they were on neurons
distributed throughout the hypothalamus,
the hindbrain, and the forebrain, suggesting
that one of these areas was the next key hub
in the hunger circuit. So which one was it?
Some of these switches were in the
paraventricular hypothalamic nucleus—
the brain area Lowell had studied in the
lab of Richard Gold as an undergraduate.
Since Lowell had seen with his own eyes
that mice ate voraciously if you took it
offline, he had long believed the PVH to
be a stop in that circuit.
Now he had the tools to prove it. Over
the years, Lowell had developed an exper-
tise in cutting-edge genetic engineering
techniques that allowed him to target
and delete specific genes and create new
strains of “knockout” mice—meaning
specific genes had been knocked out in
an embryo, causing a mouse to be born
without a functional copy.
In 2005, Lowell and a colleague, Joel
Elmquist, engineered mice to carry a
genetic sequence that prevented them
from making functional copies of sati-
ety switches anywhere in the brain. As
expected, the mice grew obese.
43
Lowell and Elmquist then created pairs
of microscopic molecular scissors. Using
genes unique to neurons in the PVH as a
homing beacon, they programmed these
scissors to seek out only DNA associated
with PVH neurons and snip away the small
sequence that prevented the development
of functional satiety switches in that part
of the brain. In other words, they “fixed”
the satiety switches in the PVH, while they
remained disabled in the rest of the brain.
If the PVH was where the magic happened,
restoring the satiety switches there would
fix the problem of obesity.
Indeed, Lowell’s knockout mice were
effectively “cured” of obesity—confirm-
ing his hypothesis. He had proved that
the PVH was the next key relay point in
the hunger-satiety circuit.
For Lowell, confirming the PVH’s place
in the circuit was huge‚ but it still did not
answer perhaps the most fascinating ques-
tion of all: How did these signals eventu-
ally make it into the conscious parts of the
brain, the parts that could make an animal
take action to get food? How did hunger, in
other words, manage to commandeer the
neural machinery of those crazed mice?
How do intrusive thoughts of a meatball
sandwich compel someone to put on shoes
and a coat and track one down?
To find out, Lowell needed to determine
where the signals in the PVH led, in the
hopes that if he continued to follow the
string it would lead him to the gateway
to higher-order brain structures. This was
complicated by the fact that neurons in the
PVH sent signals to a number of different
areas, including the brain stem, regions
that affect thyroid function, and others.
Lowell was stymied. “We could knock
out these genes and then measure how
much food the mice ate or measure how
fat they got, but we couldn’t go much fur-
ther,” he says.
A magic “remote control”
In the summer of 2009, four years after the
PVH discovery, Lowell was visiting Colgate
in upstate New York with his high-school-
age son. Lying on the grass outside the
administrative building while his son did an
44
interview, he flipped open the latest issue
of the scientific journal Neuron. An article
detailed a new laboratory tool developed
by Bryan Roth at the University of North
Carolina, Chapel Hill: a “chemical-genetic
remote control” that could be used to turn
specific neurons on and off in mice. Lowell
recognized instantly it was the breakthrough
he had been waiting for his entire career.
Instead of just knocking out populations
of neurons permanently in mice, Lowell
could instead create new strains of mice that
were bred to have this genetic “remote con-
trol” switch, allowing him to turn distinct
populations of neurons on and off simply
by administering a chemical agent. (A sepa-
rate technique known as optogenetics also
allows him to do this by beaming a specific
wavelength of light into the brain through
a fiber-optic cable.) He could then observe
the behavioral effect of turning specific
neurons on and off in real time.
“Suddenly I was able to do things that
when I was an undergraduate I never
dreamed I’d be able to do,” he says.
In 2014, Lowell used the remote-control
tool to methodically turn each bundle
of neurons leading out of the PVH on
and off, to see which ones produced sati-
ety. Once he identified the neurons that
affected satiety, he followed them out of
the hypothalamus. It led him to an area
in the brain stem called the parabrachial
nucleus (PBN)—the third key hub involved
in the hunger-satiety circuit.
It was a scientific watershed. Lowell had
finally arrived at an area of the brain with
direct connections to higher-order brain
structures affecting all aspects of our con-
scious experience, including areas involved
in motivation, reward, emotion, processing
sensory stimuli, memory, selective atten-
tion, and a wide array of other functions.
Somewhere in that area of the brain
was the last way station, the “Holy Grail”
neurons: those finally telling the rest of
the brain to “go eat.”
Hunting for the Holy Grail
For the past eight years, Lowell and
Andermann have been looking for the
PBN neurons involved in hunger. It’s a
painstaking hunt—the PBN contains hun- be detected, using sophisticated optical
dreds of thousands of neurons. Lowell’s sensing technology, through a window in
lab is tracing the hunger-satiety circuit for- the skull and then reproduced on a com-
ward out of the PBN while Andermann’s puter screen.) This has allowed Lowell
lab works backwards toward it from the and Andermann to reduce the number of
insular cortex, an area associated with the candidate neurons he is considering from
conscious experience of bodily states like hundreds of thousands to about 10,000.
hunger. The goal is to meet in the middle. To further narrow down the possibili-
If they can trace this circuit, then they ties, Lowell spent three years sorting these
will begin to examine how it is that a sim- 10,000 neurons into different subtypes
ple signal—a signal that we are hungry— using their genetic signatures. He has
works to recruit higher-order brain areas identified 37 genetically distinct subtypes.
and focuses them on the completion of Now Lowell and Andermann are exper-
a task. They will have the opportunity to imenting with subtype after subtype to
develop a model of how animals translate see which ones are involved in the hun-
desire into action. Put simply, they might ger circuit.
be able to characterize a complex action To do so, they are exposing live mice to
from beginning to end. different conditions and watching to see
The sheer number of neurons in the which neurons fire in response. They can
PBN makes the task daunting. It’s made see if a neuron fires when, for instance, the
even more complicated by the fact that the mice are shown pictures they’ve learned
PBN isn’t just involved in sending hunger to associate with a tasty treat.
signals to higher-order brain processing Once they identify neurons that are
centers but is also the final stop for scores activated in the PBN by the food cue, they
of other impulses and needs. It is a huge are using other experimental techniques to
way station for all sorts of information, figure out which of the 37 distinct genetic
most of which has nothing to with hun- profiles these neurons carry.
ger—like sexual arousal; the sensations The process, which involves sacrific-
associated with pain; the detection of heat ing the mice and dissecting their brain
and cold, itches and nausea; and signals tissue, can be painstaking. But Lowell and
associated with a wide
array of autonomic func-
tions, including respira- It still did not answer perhaps the
tion, blood pressure, and
temperature regulation. most fascinating question of all:
Each one of these signals How did these signals eventually make it
likely has its own set of into the conscious parts of the brain?
dedicated, genetically
distinct neurons in the
PBN. Most of these neu-
rons have never been identified or studied. Andermann insist they are closing in on
And they all look identical. their target. They hope that within the next
At times, the researchers have had to five years they will have found the neurons
trace the path of nerve impulses one neu- they are looking for. From there, they can
ron at a time—activating a neuron they proceed into higher-order areas of the brain.
know is part of the hunger-satiety circuit The recent development of the new
using the “remote control” technologies, class of weight-loss drugs—and the expe-
and then watching to see which neurons riences reported by patients—tantalizingly
light up in response. (The DNA of the mice illustrate how much power the circuits they
he works with also contains sequences are tracing can have on those areas. Not
for fluorescent tracers that light up when only is the physical experience of hunger
certain neurons fire, and that light can absent—because the drugs seem to lower

For the past eight years,
neuroscientist Mark Andermann has
worked with Lowell to hunt for the
Holy Grail neurons.
45
the body’s “set point”—but everything
else that usually goes along with hunger
seems to fade away. Patients report that
they are no longer plagued by intrusive
thoughts of food. (These reports parallel
what Andermann and Lowell are seeing in
the lab. Using their neural imaging tech-
niques, the researchers can actually tell
when mice are thinking about visual cues
they have seen in the last minute or hour.)
It remains to be seen whether Lowell
and Andermann’s work will actually resolve
the intense debate in the field over how
these drugs work, and what parts of the
brain they act on. But the researchers hope
that by decoding the circuit, their findings
may inform the development of new gener-
ations of drugs that are even more effective
and lack side effects such as nausea, vomit-
ing, diarrhea, abdominal pain, and, in some
cases, pancreatitis and changes in vision.
Though this would be newsworthy, it’s
still not what excites Lowell the most. He
remains most committed to the idea that
his research could yield new insights into
motivation, decision-making, and a wide
array of other functions—into human will
and survival. To illustrate why he is excited,
he talks about a video he’s seen of a hungry
squirrel navigating a “Mission Impossible”
course to access food; the squirrel climbs
up a pole, hurls itself through the air and
lands on a windmill, and shimmies through
a small opening in a plastic barrier while
hanging upside-down from a clothesline.
“The squirrel isn’t operating on reflex,”
he says. “It’s a totally novel environment.
It has to use all of its higher processes to
achieve that goal.” How does this very sim-
ple system manage to take over?
“That’s the big question,” he says. “We
don’t know how any of that works, those
higher processes.”
Now that he’s finally equipped with all
the tools he needs to untangle the dizzyingly
complex bowl of neural spaghetti, it may just
be a matter of time before he finds out.
Adam Piore is a freelance journalist
based in New York. He is the author
of The Body Builders: Inside the
Science of the Engineered Human,
about how bioengineering is changing
modern medicine.
46

I
Quantum computing is now confronting its biggest n the past 20 years, hundreds of companies, including giants
technical stumbling block. like Google, Microsoft, and IBM, have staked a claim in the
rush to establish quantum computing. Investors have put in
well over $5 billion so far. All this effort has just one purpose:
creating the world’s next big thing.
Quantum computers use the counterintuitive rules that gov-
ern matter at the atomic and subatomic level to process informa-
By Photographs by
tion in ways that are impossible with conventional, or “classical,”
Michael Brooks Peter Garritano
computers. Experts suspect that this technology will be able to
make an impact in fields as disparate as drug discovery, cryp-
tography, finance, and supply-chain logistics. The promise is
certainly there, but so is the hype. In 2022, for instance, Haim
Israel, managing director of research at Bank of America, declared

Bring
that quantum computing will be “bigger than fire and bigger
than all the revolutions that humanity has seen.” Even among
scientists, a slew of claims and vicious counterclaims have made
it a hard field to assess.
Ultimately, though, assessing our progress in building useful
quantum computers comes down to one central factor: whether
we can handle the noise. The delicate nature of quantum systems

on
makes them extremely vulnerable to the slightest disturbance,
whether that’s a stray photon created by heat, a random signal
from the surrounding electronics, or a physical vibration. This
noise wreaks havoc, generating errors or even stopping a quan-
tum computation in its tracks. It doesn’t matter how big your
processor is, or what the killer applications might turn out to
be: unless noise can be tamed, a quantum computer will never

the
surpass what a classical computer can do.
For many years, researchers thought they might just have to
make do with noisy circuitry, at least in the near term—and many
hunted for applications that might do something useful with
that limited capacity. The hunt hasn’t gone particularly well, but
that may not matter now. In the last couple of years, theoretical
and experimental breakthroughs have enabled researchers to

noise
declare that the problem of noise might finally be on the ropes.
A combination of hardware and software strategies is showing
promise for suppressing, mitigating, and cleaning up quantum
errors. It’s not an especially elegant approach, but it does look
as if it could work—and sooner than anyone expected.
“I’m seeing much more evidence being presented in defense
of optimism,” says Earl Campbell, vice president of quantum
science at Riverlane, a quantum computing company based in
Cambridge, UK.
Even the hard-line skeptics are being won over. University of
Helsinki professor Sabrina Maniscalco, for example, researches
Opposite: Jay Gambetta
heads development of IBM’s the impact of noise on computations. A decade ago, she says,
quantum computers and led she was writing quantum computing off. “I thought there were
the initiative to put such
systems in the cloud. really fundamental issues. I had no certainty that there would
be a way out,” she says. Now, though, she is working on using
quantum systems to design improved versions of light-activated
47
48

cancer drugs that are effective at lower concentrations and can they asked was whether Gambetta was sure he could pull it off.
be activated by a less harmful form of light. She thinks the proj- “I said yes,” he says. “I thought, how hard can it be?”
ect is just two and a half years from success. For Maniscalco, the Very hard, it turned out, because IBM’s executives told
era of “quantum utility”—the point at which, for certain tasks, it Gambetta he had to get it done quickly. “I wanted to spend two
makes sense to use a quantum rather than a classical processor—is years doing it,” he says. They gave him a year.
almost upon us. “I’m actually quite confident about the fact that It was a daunting challenge: he barely knew what the cloud
we will be entering the quantum utility era very soon,” she says. was back then. Fortunately, some of his colleagues did, and they
were able to upgrade the team’s remote access protocols—useful
Putting qubits in the cloud for tweaking the machine in the evening or on the weekend—to
This breakthrough moment comes after more than a decade of create a suite of interfaces that could be accessed from anywhere
creeping disappointment. Throughout the late 2000s and the in the world. The world’s first cloud-access quantum computer,
early 2010s, researchers building and running real-world quan- built using five qubits, went live at midnight on May the 4th,
tum computers found them to be far more problematic than the 2016. The date, Star Wars Day, was chosen by nerds, for nerds.
theorists had hoped. “I don’t think anyone in upper management was aware of that,”
To some people, these problems seemed insurmountable. Gambetta says, laughing.
But others, like Jay Gambetta, were unfazed. Not that upper management’s reaction to the launch date was
A quiet-spoken Australian, Gambetta has a PhD in physics uppermost in his mind. Of far more concern, he says, was whether
from Griffith University, on Australia’s Gold Coast. He chose to a system reflecting years of behind-the-scenes development work
go there in part because it allowed him to feed his surfing addic- would survive being hooked up to the real world. “We watched
tion. But in July 2004, he wrenched himself away and skipped off the first jobs come in. We could see them pinging on the quantum
to the Northern Hemisphere to do research at Yale University computer,” he says. “When it didn’t break, we started to relax.”
on the quantum properties of light. Three years later (by which Cloud-based quantum computing was an instant hit. Seven
time he was an ex-surfer thanks to the chilly waters around New thousand people signed up in the first week, and there were 22,000
Haven), Gambetta moved even further north, to the University of registered users by the end of the month. Their ventures made
Waterloo in Ontario, Canada. Then he learned that IBM wanted it clear, however, that quantum computing had a big problem.
to get a little more hands-on with quantum computing. In 2011, The field’s eventual aim is to have hundreds of thousands, if
Gambetta became one of the company’s new hires. not millions, of qubits working together. But when it became pos-
IBM’s quantum engineers had been busy building quantum sible for researchers to test out quantum computers with just a few
versions of the classical computer’s binary digit, or bit. In classi-qubits working together, many theory-based assumptions about
cal computers, the bit is an electronic switch, with two states to how much noise they would generate turned out to be seriously off.
represent 0 and 1. In quantum computers, things are less black Some noise was always in the cards. Because they operate
and white. If isolated from noise, a quantum bit, or “qubit,” can at temperatures above absolute zero, where thermal radiation
exist in a probabilistic combination of those two possible states, is always present, everyone expected some random knocks to
a bit like a coin in mid-toss. This property
of qubits, along with their potential to be
“entangled” with other qubits, is the key “We watched the first jobs come in.
to the revolutionary possibilities of quan- We could see them pinging on the quantum computer.
tum computing. When it didn’t break, we started to relax.”
A year after joining the company,
Gambetta spotted a problem with IBM’s
qubits: everyone could see that they were
getting pretty good. Whenever he met up
with his fellow physicists at conferences, they would ask him to the qubits. But there were nonrandom knocks too. Changing
test out their latest ideas on IBM’s qubits. Within a couple of temperatures in the control electronics created noise. Applying
years, Gambetta had begun to balk at the volume of requests. “I pulses of energy to put the qubits in the right states created
started thinking that this was insane—why should we just run noise. And worst of all, it turned out that sending a control sig-
experiments for physicists?” he recalls. nal to one qubit created noise in other, nearby qubits. “You’re
It occurred to him that his life might be easier if he could find manipulating a qubit and another one over there feels it,” says
a way for physicists to operate IBM’s qubits for themselves— Michael Biercuk, director of the Quantum Control Laboratory
maybe via cloud computing. He mentioned it to his boss, and at the University of Sydney in Australia.
then he found himself with five minutes to pitch the idea to By the time quantum algorithms were running on a dozen
IBM’s executives at a gathering in late 2014. The only question or so qubits, the performance was consistently shocking. In a

IBM’s Quantum System One,
a commercially available
quantum computer, uses
this chandelier-like
structure to cool qubits.
2022 assessment, Biercuk and others calculated the probability
that an algorithm would run successfully before noise destroyed
the information held in the qubits and forced the computation
off track. If an algorithm with a known correct answer was run
30,000 times, say, the correct answer might be returned only
three times.
Though disappointing, it was also educational. “People learned
a lot about these machines by actually using them,” Biercuk says.
“We found a lot of stuff that more or less nobody knew about—
or they knew and had no idea what to do about it.”
49
Fixing the errors
Once they had recovered from this noisy slap,
researchers began to rally. And they have now
come up with a set of solutions that can work
together to bring the noise under control.
Broadly speaking, solutions can be classed
into three categories. The base layer is error
suppression. This works through classical
software and machine-learning algorithms,
which continually analyze the behavior of the
circuits and the qubits and then reconfigure
the circuit design and the way instructions
are given so that the information held in the
qubits is better protected. This is one of the
things that Biercuk’s company, Q-CTRL, works
on; suppression, the company says, can make
quantum algorithms 1,000 times more likely
to produce a correct answer.
The next layer, error mitigation, uses the
fact that not all errors cause a computation to
fail; many of them will just steer the compu-
tation off track. By looking at the errors that
noise creates in a particular system running a
particular algorithm, researchers can apply a
kind of “anti-noise” to the quantum circuit to
reduce the chances of errors during the compu-
tation and in the output. This technique, some-
thing akin to the operation of noise-canceling
headphones, is not a perfect fix. It relies, for
instance, on running the algorithm multiple
times, which increases the cost of operation,
and the algorithm only estimates the noise.
Nonetheless, it does a decent job of reducing
errors in the final output, Gambetta says.
Helsinki-based Algorithmiq, where
Maniscalco is CEO, has its own way of cleaning
up noise after the computation is done. “It basi-
cally eliminates the noise in post-processing,
like cleaning up the mess from the quantum
computer,” Maniscalco says. So far, it seems
to work at reasonably large scales.
On top of all that, there has been a grow-
ing roster of achievements in “quantum error correction,” or
QEC. Instead of holding a qubit’s worth of information in one
qubit, QEC encodes it in the quantum states of a set of qubits.
A noise-induced error in any one of those is not as catastrophic
as it would be if the information were held by a single qubit: by
monitoring each of the additional qubits, it’s possible to detect any
change and correct it before the information becomes unusable.
Implementing QEC has long been considered one of the
essential steps on the path to large-scale, noise-tolerant quan-
tum computing—to machines that can achieve all the promise
50

of the technology, such as the ability to crack popular encryp- A quantum computer doesn’t have to be a better computer than
tion schemes. The trouble is, QEC uses a lot of overhead. The any other kind of machine to attract paying customers, Tan says.
gold-standard error correction architecture, known as a surface It just has to be comparable in performance and cheaper to run.
code, requires at least 13 physical qubits to protect a single use- He expects we’ll achieve that quantum energy advantage in the
ful “logical” qubit. As you connect logical qubits together, that next three to five years.
number balloons: a useful processor might require 1,000 phys-
ical qubits for every logical qubit. Finding utility
There are now multiple reasons to be optimistic even about A debate has long raged about what target quantum computing
this, however. In July 2022, for instance, Google’s researchers researchers should be aiming for in their bid to compete with
published a demonstration of a surface code in action where classical computers. Quantum supremacy, the goal Google has
performance got better—not worse—when more qubits were pursued—a demonstration that a quantum computer can solve a
connected together. problem no classical computer can crack in a reasonable amount
There have also been promising demonstrations of theoret- of time? Or quantum advantage—superior performance when it
ical alternatives to surface codes. In August 2023, an IBM team comes to a useful problem—as IBM has preferred? Or quantum
that included Gambetta showed an error correction technique utility, IBM’s newest buzzword? The semantics reflect differing
that could control the errors in a 12-qubit memory circuit using views of what near-term objectives are important.
an extra 276 qubits, a big improvement over the thousands of In June, IBM announced that it would begin retiring its entry-
extra qubits required by surface codes. level processors from the cloud, so that its 127-qubit Eagle pro-
In September, two other teams demonstrated similar improve- cessor would be the smallest one that the company would make
ments with a fault-tolerant circuit called a CCZ gate, using super- available. The move is aimed at pushing researchers to priori-
conducting circuitry and ion-trap processors. tize truly useful tasks. Eagle is a “utility-scale” processor, IBM
That so many noise-handling techniques are flourishing is a says—when correctly handled, it can “provide useful results to
huge deal—especially at a time when the notion that we might problems that challenge the best scalable classical methods.”
get something useful out of small-scale, noisy processors has It’s a controversial claim—many doubt that Eagle really is
turned out to be a bust. capable of outperforming suitably prepared classical machines.
Actual error correction is not yet happening on commercially But classical computers are already struggling to keep up with
available quantum processors (and is not generally implementable it, and IBM has even larger systems: the 433-qubit Osprey pro-
as a real-time process during computations). But Biercuk sees quan- cessor, which is also cloud-accessible, and a 1,121-qubit Condor
tum computing as finally hitting its stride. “I think we’re well on processor set to debut by the end of 2023, after this issue goes
the way now,” he says. “I don’t see any fundamental issues at all.” to press. (Gambetta has a simple rationale for the way he names
And these innovations are happening alongside general IBM’s quantum processors: “I like birds.”) The company has a new
improvements in hardware performance—meaning that there are modular design, called Heron, and Flamingo is slated to appear in
ever fewer baseline errors in the function-
ing qubits—and an increase in the number
That so many noise-handling techniques are
of qubits on each processor, making big-
ger and more useful calculations possible.
flourishing is a huge deal—especially at a time when the
Biercuk says he is starting to see places notion that we might get something useful out of
where he might soon choose a quantum small-scale, noisy processors has turned out to be a bust.
computer over the best-performing clas-
sical machines. Neither a classical nor a
quantum computer can fully solve large-
scale tasks like finding the optimal routes
for a nationwide fleet of delivery trucks. But, Biercuk points out, 2025—with fully quantum connections between chips that allow
accessing and running the best classical supercomputers costs the quantum information to flow between different processors
a great deal of money—potentially more than accessing and unhindered, enabling truly large-scale quantum computation.
running a quantum computer that might even give a slightly That will make 2025 the first year that quantum computing will
better solution. be provably scalable, Gambetta says: “I’m aiming for 2025 to be
“Look at what high-performance computing centers are an important year for demonstrating key technologies that allow
doing on a daily basis,” says Kuan Tan, CTO and cofounder of us to scale to hundreds of thousands of qubits.”
the Finland-based quantum computer provider IQM. “They’re IQM’s Tan is astonished at the pace of development. “It’s
running power-hungry scientific calculations that are reachable mind-boggling how fast this field is progressing,” he says. “When I
[by] quantum computers that will consume much less power.” was working in this field 10 years ago, I would never have expected
 51
IBM’s modular Heron chip is
designed to enable multi-chip
quantum computers that are linked
up with classical communication
connections. The company aims for
future chips to support quantum
communication.

creating qubits in silicon devices that the com-

to have a 10-qubit chip at this point. Now we’re talking about
hundreds already, and potentially thousands in the coming years.”
It’s not just IBM. Campbell has been impressed by Google’s
quiet but emphatic progress, for instance. “They operate differ-
ently, but they have hit the milestones on their public road map,”
he says. “They seem to be doing what they say they will do.” Other
household-name companies are embracing quantum computing
too. “We’re seeing Intel using their top-line machines, the ones
that they use for making chips, to make quantum devices,” Tan
says. Intel is following a technology path very different from IBM’s:
pany knows how to manufacture at scale, with
minimal noise-inducing defects.
As quantum computing hits its stride and
quantum computers begin to process real-world
data, technological and geographical diversity
will be important to avoid geopolitical issues
and problems with data-sharing regulations.
There are restrictions, for instance, aimed
at maintaining national security—which will
perhaps limit the market opportunities of mul-
tinational giants such as IBM and Google. At
the beginning of 2022, France’s defense min-
ister declared quantum technologies to be of
“strategic interest” while announcing a new
national program of research. In July 2023,
Deutsche Telekom announced a new part-
nership with IQM for cloud-based access to
quantum computing, calling it a way for DT
customers to access a “truly sovereign quan-
tum environment, built and managed from
within Europe.”
This is not just nationalistic bluster: sov-
ereignty matters. DT is leading the European
Commission’s development of a quantum-
based, EU-wide high-security communications
infrastructure; as the era approaches when
large-scale quantum computers pose a seri-
ous threat to standard encryption protocols,
governments and commercial organizations
will want to be able to test “post-quantum”
encryption algorithms—ones that withstand
attack by any quantum computer, irrespective
of its size—within their own borders.
Not that this is a problem yet. Few people
think that a security-destroying large-scale
quantum processor is just around the corner.
But there is certainly a growing belief in the
field’s potential to be transformative—and use-
ful—in other ways within just a few years. And
these days, that belief is based on real-world
achievements. “At Algorithmiq, we believe in
a future where quantum utility will happen soon, but I can trace
this optimism back to patents and publications,” Maniscalco says.
The only downside for her is that not everybody has come around
the way she has. Quantum computing is here now, she insists—
but the old objections die hard, and many people refuse to see it.
“There is still a lot of misunderstanding: I get very upset when
I see or hear certain conversations,” she says. “Sometimes I wish
I had a magic wand that could open people’s eyes.”
Michael Brooks is a freelance science journalist based
in the UK.
Cured Sublime Cement
shown in a variety
of molds.

Opposite, top right:

Non-carbonate
minerals.

Opposite, bottom
left: Sublime
Silicates (top)
and Sublime Lime
(bottom), the two main
reactive components
for Sublime Cement.
 53

Cement is a climate nightmare. This startup is using electricity to change that.

Climate’s hardest problem

By CASEY CROWNHART Q Q Q Photographs by Bob O’Connor

54

Cement
construction, Sublime will need to persuade
builders to use its material in the first place.

he cement industry pumps 2.6 billion

hides in plain sight—it’s used to build everything from

T metric tons of carbon dioxide into the
atmosphere every year. Cleaning that
up will require addressing the two distinct
roads and buildings to dams and basement floors. sources of cement’s greenhouse-gas emis-
sions: heat and chemistry.
But there’s a climate threat lurking in those ubiq- To make cement today, a mixture often
uitous gray slabs. Cement production accounts for containing limestone, sand, and clay is
more than 7% of global carbon dioxide emissions— ground up and baked in kilns at tempera-
more than sectors like aviation, shipping, or landfills. tures of up to 1,500 °C (just over 2,700 °F).
The heat kicks off reactions, transforming
limestone to lime and combining that with
silicon dioxide in the sand and clay. Those
Humans have been making cement, in kilns, Sublime’s technology zaps them in reactions are complex, but in most cases the
one form or another, for thousands of years. water with electricity, kicking off chemical crucial final product is a mixture of a few
Ancient Romans used volcanic ash, crushed reactions that form the main ingredients in silicon, calcium, and oxygen compounds that
lime, and seawater to build the aqueducts its cement. allow the cement, after it’s mixed with water
and iconic structures like the Pantheon. The Over the course of the past several years, (and sand and gravel are added), to harden
modern version of hydraulic cement—the the startup has gone from making batches into concrete, a sturdy building material.
sort that hardens when mixed with water of cement that could fit in the palm of your Humans use more concrete, by weight,
and allowed to dry—dates back to the early hand to starting up a pilot facility that can than any other material except water.
19th century. Derived from widely available produce around 100 tons each year. While Cement is the glue that holds that mate-
materials, it’s cheap and easy to make. Today, it’s still tiny compared with traditional rial together, accounting for about 10% of
cement is one of the most-used materials cement plants, which can churn out a mil- its volume.
on the planet, with about 4 billion metric lion tons or more annually, the pilot line Roughly 40% of the greenhouse-gas
tons produced annually. represents the first crucial step to proving emissions associated with cement produc-
Industrial-scale cement is a multifaceted that electrochemistry can stand up to the tion come from fossil fuels that generate
climate conundrum. Making it is energy challenge of producing one of the world’s the heat required. This is a common prob-
intensive: the inside of a traditional cement most important building materials. lem across heavy industry: fossil fuels have
kiln is hotter than lava in an erupting vol- By the end of the decade, Sublime plans become ingrained in production processes
cano. Reaching those temperatures typi- to have a full-scale manufacturing facility up because of their low cost, says Rebecca Dell,
cally requires burning fossil fuels like coal. and running that’s capable of producing a head of the industry group at the nonprofit
There’s also a specific set of chemical reac- million tons of material each year. But tra- research group ClimateWorks. However,
tions needed to turn crushed-up minerals ditional large-scale cement plants can cost lower-cost renewables are coming to the
into cement—and those reactions release over a billion dollars to build and outfit. grid, opening the door for more industrial
carbon dioxide, the most common green- Competing with established industry play- sectors to switch to electricity.
house gas in the atmosphere. ers will require Sublime to scale fast while It’s possible to use electric kilns to make
One solution to this climate catastro- raising the additional funding it will need to cement, and some major industry players,
phe might be coursing through the pipes support that growth. The end of 0% interest like Cemex, are working to pilot the tech-
at Sublime Systems. Founded by two MIT rates makes such a task increasingly diffi- nology to cut emissions from heat demand.
battery scientists, the startup is developing cult for any business, but especially for one These electric kilns, if powered by renewable
an entirely new way to make cement. Instead producing a commodity like cement. And electricity, could provide a potential path-
of heating crushed-up rocks in lava-hot in a high-stakes, low-margin industry like way to cutting cement’s climate impacts.

MIT professor Yet-Ming Chiang and

Leah Ellis, CEO of Sublime Systems,
cofounded the company in 2020.
56

But that does nothing to remedy the

other sources of emissions from cement
production: the roughly 60% that come not
from heat but from the chemical reactions
required to transform the starting materials
into the building material.
Most cement starts with limestone, a sed-
imentary rock that contains calcium, oxygen,
and carbon. In cement kilns, limestone is
turned into lime by means of reactions that
peel away carbon dioxide, which is usually
released into the atmosphere. Limestone is
about half carbon dioxide by mass, so those
emissions can add up. That means fully
decarbonizing the cement industry might
require a more radical reinvention. This is
where Sublime comes in.
“Cement doesn’t really get the pizzazz
that I think it deserves,” says Leah Ellis,
Sublime’s cofounder and CEO. She speaks
A Sublime
quickly, as if running short on time, thoughts employee removes
pouring out and conveyed with a slight the company’s
reactive silica
Canadian accent. As she explains the intri- product from its
cacies of cement chemistry, her eyes shine equipment as
behind bubblegum-pink glasses. part of an R&D-
scale production
Ellis didn’t always plan on getting into the process.
building materials business. After growing
up in Canada, she completed her graduate
work under the famed battery pioneer Jeff
Dahn. She then struck out for MIT to work
with Yet-Ming Chiang, another major figure
in battery research and a serial entrepreneur.
Chiang has founded several energy storage
ventures, including A123 Systems, 24M,
and Form Energy, and is now a cofounder
of Sublime Systems with Ellis.
When Ellis first arrived at MIT, it was
Chiang who floated a different idea for her
potential research there. She recalls him ush-
ering her into his office and, to her surprise,
asking if she was bored with batteries. “And
I thought it was a trick question,” she says,
“because he’s obviously a famous battery
scientist.” But Chiang had an idea for making
cement in a new way, using a tool from the
battery-science tool kit: electrochemistry.
 57

Instead of burning fossil fuels to heat up

cement kilns, Chiang suggested, there might
be a way to spark the necessary chemical
reactions to make cement using electricity.
The team later discovered that a device
called an electrolyzer might be useful in
such a process. Electrolyzers are devices
that use electricity to kick off chemical reac-
tions. They’re usually used to split water into
its components, producing hydrogen and
oxygen. But electrolyzers can work with
other chemical reactions too, like generat-
ing acids and bases—the potential key to a
new cement production process.
By 2019, Ellis and her fellow researchers
had discovered a method with potential
to fulfill Chiang’s initial idea. They found
that an electrolyzer could be used to form
a pH gradient in a tank, with dissolved
limestone on the acidic end and hydrated
A Sublime lime on the other. That lime could then
employee tests be combined with reactive silica to form
the compressive
strength of the
the same compounds formed by tradi-
company’s cement tional cement.
in concrete After some additional technical devel-
cylinders.
opment and industry analysis, the pair
decided to spin out the research into a
company with Ellis at the helm. Chiang
says, “It didn’t take very long before I rec-
ognized that there was both a starting idea
and also a person to carry it.”
Four years later, Sublime is running a
pilot manufacturing line. It’s a major step
up from the early days at MIT, where Ellis
and her lab mates would produce about
enough material to make a single die. The
line started up in late 2022, says Mike
Corbett, Sublime’s director of engineering,
and is 20 times larger than anything the
company had run in the lab.
Inside a cavernous room at the start-
up’s headquarters, the stainless-steel
tanks lining the walls of the pilot space
are arranged roughly from left to right.
Into tanks on one side of the room go
ground-up rocks, and out of those on the

At full capacity, the pilot line would take
a week to make enough cement to supply
a single concrete truck.
other comes hydrated lime, one of the main
ingredients in Sublime’s cement.
The results of various experiments and
trial runs sit nearby, in five-gallon buck-
ets stacked on metal racks. When Corbett
and a colleague crack the lid on one, the
hydrated lime inside is an unassuming
white powder, chalky and slightly clumped
together like baking soda in a box that’s
been left open for too long.
It would be difficult to use what’s in these
buckets to build anything: at full capacity, the
pilot line would take about a week to make
enough cement to supply a single concrete
truck. And it takes three to four truckloads
of cement to pour the foundation for just
one average single-family home in the US.
Instead, the startup is making material
to send to potential partners, running tests
on cement blocks, and, crucially, helping
design the next facilities. Those, which will
be significantly larger, could be the final
step in proving that Sublime’s process can
work in the cement industry.
S
ublime Systems isn’t the only player
trying to green up the gray glue.
Early efforts to cut emissions from
cement have largely focused on efficiency,
says Radhika Lalit, the initiative director
for industry at the Climate Imperative
Foundation, who formerly worked in the
industry group at the Rocky Mountain
Institute, a nonprofit research agency.
For example, adding fillers called supple-
mentary cementitious materials, which
can react with the active ingredients in
cement, can help cut down on total emis-
sions without affecting the properties of
the concrete.
Equipment inside Sublime Systems
headquarters. The company is
working on designing its next
facilities, which will have even
greater capacity.
59
This strategy only works up to a certain
point, Lalit points out: after that, these fill-
ers will begin to degrade the strength and
lifetime of cement.
There are other ways to add materials to
cement to cut climate impacts. The Canadian
company CarbonCure Technologies, for
example, has developed technology to
inject carbon dioxide into cement mixtures.
According to CarbonCure, the gas can then
react with the mixture and mineralize, lock-
ing it away from the atmosphere and adding
to the material’s strength. CarbonCure is
working with carbon removal companies like
Heirloom Carbon Technologies to demon-
strate the possibility of using concrete for
long-term carbon dioxide storage.
But methods like cutting down the total
material used or adding in fillers are limited
in how much they can cut emissions: it’s
not possible to make unlimited efficiency
tweaks to get to zero. So many established
players in the cement world are looking to
add carbon capture and sequestration plants
to existing facilities, Lalit says. By snaring
carbon dioxide from plant exhaust before
it can be released into the atmosphere, this
add-on helps active facilities cut their emis-
sions so that making the industry cleaner
doesn’t require replacing them entirely.
Adding carbon capture technology to
existing infrastructure could mean a lon-
ger lifetime for conventional equipment, a
major benefit for recently built plants that
are often designed to run for 30 to 50 years.
But carbon capture is still largely unproven
in heavy industry: existing methods often
aren’t able to capture all emissions, and the
few existing large-scale units have faced
delays, Lalit says.
60
“A cement startup is probably one of
the most difficult ones you can imagine.
Not only is it technically difficult, but it’s
very capital intensive ... And it’s not sexy.”
To be on track for net-zero goals, the
cement industry would need to add car-
bon capture facilities to between 33 and
45 existing cement plants by 2030, accord-
ing to RMI. The first such industrial-scale
facility at a cement plant is scheduled to
come online this year.
This could be expensive: a carbon cap-
ture system could cost as much as $120
per metric ton of carbon dioxide removed,
which Ellis says would roughly double the
final cost of cement at current prices.
The cost is one reason Sublime and a
number of other startups are eschewing car-
bon capture and aiming to cut cement’s car-
bon footprint using more radical formulas.
One of the other best-funded cement
startups is Brimstone, a California-based
group with what it calls a carbon-negative
process for making cement. This means
that the product ends up pulling more car-
bon dioxide out of the atmosphere than it
emits, says Cody Finke, Brimstone’s CEO.
This works in two major ways. First,
rather than limestone, Brimstone uses other
minerals called silicates that don’t contain
carbon dioxide, so there are no process
emissions from the company’s cement, says
Finke. In addition, a waste product of the
silicates is a magnesium-containing material
that acts like a sponge, soaking up carbon
dioxide from the air and mineralizing it.
There are still emissions associated
with making Brimstone’s cement, largely
related to the heat required. But they can
be effectively canceled out with the mineral
process, Finke explains, especially since
the company plans to use electric kilns. If
those are powered by electricity from the
average US power grid, the startup claims,
its material would wind up removing more
carbon dioxide than it produces—about
130 kilograms more per ton of cement.
The reason Brimstone is sticking to this
high-temperature process, despite its asso-
ciated energy needs and emissions, is that
it’s producing a cement with a chemical
formula that dominates the industry today.
Portland cement, which has been around
since the early 1800s, is likely what you
think of when you think of cement. It is
known for producing predictable and strong
concrete. However, there’s a catch: making
it necessitates ultra-high heat, because a
key ingredient of the cement—alite—can
only form at temperatures over 1,250 °C.
Originally, Sublime planned to make
Portland cement as well. “For something
as low-cost as cement, we thought you’d
need to produce what the market is used
to using,” Chiang says. But the temperature
requirements pushed Sublime to reconsider
as it began making its product at larger
scales: “There was this 200-year-old inven-
tion that we were inventing around.”
So the team went hunting and dis-
covered that there are other pathways to
the chemical bonds that give concrete
made with Portland cement its strength.
Sublime’s material takes one of these alter-
native routes: rather than alite, it uses lime
and reactive silicates to react with water
to form the final material.
The choice of cement chemistry might
seem like a minor quibble. But in a high-
stakes industry like construction, it could
be a key deciding factor in which startups
win large contracts and partnerships, and
which fizzle out. “People are, I think, right-
fully skeptical of a new cement,” Ellis says,
before stopping herself: “Well, it is and it
isn’t a new cement.”
Ellis maintains that Sublime’s material is
as strong and durable as Portland cement,
if not more so. But there’s a chance that
developers will be hesitant, at least initially,
to move away from the material they already
know, says Climate Imperative’s Lalit.
ne of the major challenges ahead
O for Sublime is making its material
at the massive scales required for
large building projects. Modern cement
facilities can typically produce over a million
tons of cement each year, while Sublime’s
pilot can make about 100 tons per year. It’s
basically “a cement plant for ants,” Ellis says.
Sublime has its sights set on scaling,
and quickly. The next stage for the startup
is a demonstration commercial facility pro-
ducing tens of thousands of tons of material
each year, which should come online in early
2026. “That’s the size where you’re no longer
invisible to the cement world,” Ellis says.
After that would come a full-scale com-
mercial plant capable of the million-ton
annual production capacity that’s typical of
the industry. The company is still working
out where to build that, though the hope
is to get it running around 2028.
Sublime’s founding duo isn’t naïve
about the challenges ahead. “A cement
startup is probably one of the most diffi-
cult ones you can imagine,” Ellis says. “Not
only is it technically difficult, but it’s very
capital intensive. It’s massive. And it’s not
sexy … everybody uses and owns cement,
but they don’t see it.”
But fixing this invisible problem could
transform the world around us. Cement
scaffolds our society, and efforts to change
it—despite the tremendous challenges
they face—are critical to cleaning up the
future of construction.
Casey Crownhart is a climate
reporter at MIT Technology Review.
A Sublime employee sets up a cement
bag, ready to fill it with the
company’s cement before shipping it
out to customers.
62

This Japanese woodblock print from 1855 depicts

the namazu, a giant mythical catfish said to be
responsible for earthquakes. In this scene, crafts-
men who will profit from the reconstruction of the city
hurry to rescue it from attacking peasants.

Before the ground


I
n September 2017, about two
minutes before a magnitude 8.2
earthquake struck Mexico City,
blaring sirens alerted residents
that a quake was coming. Such
alerts, which are now available in
the United States, Japan, Turkey,
Italy, and Romania, among other
countries, have changed the way
we think about the threat of earth-
quakes. They no longer have to take
us entirely by surprise.
Earthquake early warning systems
can send alarms through phones or
transmit a loud signal to affected
regions three to five seconds after
a potentially damaging earthquake
begins. First, seismometers close to
the fault pick up the beginnings of
the quake, and finely programmed
algorithms determine its probable
size. If it is moderate or large, the
resulting alert then travels faster
than the earthquake itself, giving
seconds to minutes of warning. This
window of time is crucial: in these
brief moments, people can shut off
electricity and gas lines, move fire
trucks into the streets, and find safe
places to go.
But these systems have limita-
tions. There are false positives and
false negatives. What’s more, they
react only to an earthquake that
has already begun—we can’t pre-
dict an earthquake the way we can
forecast the weather. And so many
earthquake-prone regions are left in a
PUBLIC DOMAIN VIA WIKIMEDIA COMMONS
state of constant suspense. A proper
forecast could let us do a lot more to
manage risk, from shutting down the
power grid to evacuating residents.
When I started my PhD in seis-
mology in 2013, the very topic of
earthquake prediction was deemed
unserious, as outside the realm of
shakes Researchers are applying machine learning and other
techniques to the age-old problem of earthquake prediction.
By Allie Hutchison
63
mainstream research as the hunt for
the Loch Ness Monster.
But just seven years later, a lot had
changed. When I began my second
postdoc in 2020, I observed that
scientists in the field had become
much more open to earthquake pre-
diction. The project I was a part of,
Tectonic, was using machine learning
to advance earthquake prediction.
The European Research Council was
sufficiently convinced of its potential
to award it a four-year, €3.4 million
grant that same year.
Today, a number of well-respected
scientists are getting serious about
the prospect of prediction and are
making progress in their respective
subdisciplines. Some are studying a
different kind of slow-motion behav-
ior along fault lines, which could
turn out to be a useful indicator that
the devastating kind of earthquake
we all know and fear is on the way.
Others are hoping to tease out hints
from other data—signals in seismic
noise, animal behavior, and electro-
magnetism—to push earthquake sci-
ence toward the possibility of issuing
warnings before the shaking begins.
In the dark
Earthquake physics can seem espe-
cially opaque. Astronomers can view
the stars; biologists can observe an
animal. But those of us who study
earthquakes cannot see into the
ground—at least not directly. Instead,
we use proxies to understand what
happens inside the Earth when its
crust shakes: seismology, the study of
the sound waves generated by move-
ment within the interior; geodesy,
the application of tools like GPS to
measure how Earth’s surface changes
over time; and paleoseismology, the
64
study of relics of past earthquakes con-
cealed in geologic layers of the landscape.
There is much we still don’t know.
Decades after the theory of plate tecton-
ics was widely accepted in the 1960s, our
understanding of earthquake genesis hasn’t
progressed far beyond the idea that stress
builds to a critical threshold, at which point
it is released through a quake. Different
factors can make a fault more suscepti-
ble to reaching that point. The presence
of fluids, for instance, is significant: the
injection of wastewater fluid from oil and
gas production has caused huge increases
in tectonic activity across the central US
in the last decade. But when it comes to
knowing what is happening along a given
fault line, we’re largely in the dark. We
can construct an approximate map of a
fault by using seismic waves and mapping
earthquake locations, but we can’t directly
measure the stress it is experiencing, nor
can we quantify the threshold beyond
which the ground will move.
For a long time, the best we could do
regarding prediction was to get a sense
of how often earthquakes happen in a
particular region. For example, the last
earthquake to rupture the entire length
of the southern San Andreas Fault in
California was in 1857. The average time
period between big quakes there is esti-
mated to be somewhere between 100
and 180 years. According to a back-of-
the-envelope calculation, we could be
“overdue.” But as the wide range suggests,
recurrence intervals can vary wildly and
may be misleading. The sample size is
limited to the scope of human history and
what we can still observe in the geologic
record, which represents a small fraction
of the earthquakes that have occurred
over Earth’s history.
In 1985, scientists began installing
seismometers and other earthquake mon-
itoring equipment along the Parkfield
section of the San Andreas Fault, in cen-
tral California. Six earthquakes in that
section had occurred at unusually reg-
ular intervals compared to earthquakes
along other faults, so scientists from the
US Geological Survey (USGS) forecasted
with a high degree of confidence that the
next earthquake of a similar magnitude
would occur before 1993. The experiment
is largely considered a failure—the earth-
quake didn’t come until 2004.
Instances of regular intervals between
earthquakes of similar magnitudes have
been noted in other places, including
Hawaii, but these are the exception, not the
rule. Far more often, recurrence intervals
are given as averages with large margins
of error. For areas prone to large earth-
quakes, these intervals can be on the scale
of hundreds of years, with uncertainty bars
that also span hundreds of years. Clearly,
this method of forecasting is far from an
exact science.
Tom Heaton, a geophysicist at Caltech
and a former senior scientist at the USGS,
is skeptical that we will ever be able to pre-
dict earthquakes. He treats them largely
as stochastic processes, meaning we can
attach probabilities to events, but we can’t
forecast them with any accuracy.
“In terms of physics, it’s a chaotic sys-
tem,” Heaton says. Underlying it all is
significant evidence that Earth’s behavior
is ordered and deterministic. But with-
out good knowledge of what’s happening
under the ground, it’s impossible to intuit
any sense of that order. “Sometimes when
you say the word ‘chaos,’ people think
[you] mean it’s a random system,” he says.
“Chaotic means that it’s so complicated
you cannot make predictions.”
But as scientists’ understanding of what’s
happening inside Earth’s crust evolves and
their tools become more advanced, it’s not
unreasonable to expect that their ability to
make predictions will improve.
Slow shakes
Given how little we can quantify about
what’s going on in the planet’s interior, it
makes sense that earthquake prediction
has long seemed out of the question. But
in the early 2000s, two discoveries began
to open up the possibility.
First, seismologists discovered a strange,
low-amplitude seismic signal in a tectonic
region of southwest Japan. It would last from
hours up to several weeks and occurred at
somewhat regular intervals; it wasn’t like
anything they’d seen before. They called it
tectonic tremor.
Meanwhile, geodesists studying the
Cascadia subduction zone, a massive stretch
off the coast of the US Pacific Northwest
where one plate is diving under another,
found evidence of times when part of the
crust slowly moved in the opposite of its
usual direction. This phenomenon, dubbed
a slow slip event, happened in a thin sec-
tion of Earth’s crust located beneath the
zone that produces regular earthquakes,
where higher temperatures and pressures
have more impact on the behavior of the
rocks and the way they interact.
The scientists studying Cascadia also
observed the same sort of signal that had
been found in Japan and determined that
it was occurring at the same time and in
the same place as these slow slip events.
A new type of earthquake had been dis-
covered. Like regular earthquakes, these
transient events—slow earthquakes—
redistribute stress in the crust, but they
can take place over all kinds of time scales,
from seconds to years. In some cases, as
in Cascadia, they occur regularly, but in
other areas they are isolated incidents.
Scientists subsequently found that
during a slow earthquake, the risk of reg-
ular earthquakes can increase, particularly
in subduction zones. The locked part of the
fault that produces earthquakes is basically
being stressed both by regular plate motion
and by the irregular periodic backward
motion produced by slow earthquakes,
at depths greater than where earthquakes
begin. These elusive slow events became
the subject of my PhD research, but (as is
often the case with graduate work) I cer-
tainly didn’t resolve the problem. To this
day, it is unclear what exact mechanisms
drive this kind of activity.
Could we nevertheless use slow earth-
quakes to predict regular earthquakes?
Since their discovery, almost every big
earthquake has been followed by several
papers showing that it was preceded by a
slow earthquake. The magnitude 9 Tohoku-
AP IMAGES
Oki earthquake, which occurred in Japan in
2011, was preceded by not one but two slow
 65

The magnitude 9 Tohoku-Oki

earthquake of 2011 was preceded
by two slow earthquakes.

ones. There are exceptions: for example,

despite attempts to prove otherwise, there
is still no evidence that a slow earthquake
preceded the 2004 earthquake in Sumatra,
Indonesia, which created a devastating
tsunami that killed more than 200,000
people. What’s more, a slow earthquake
is not always followed by a regular earth-
quake. It’s not known whether something
distinguishes those that could be precur-
sors from those that aren’t.
It may be that some kind of distinc-
tive process occurs along the fault in the
hours leading up to a big quake. Last sum-
mer a former colleague of mine, Quentin
Bletery, and his colleague Jean-Mathieu
Nocquet, both at Géoazur, a multidisci-
plinary research lab in the south of France,
published the results of an analysis of
data on crustal deformation in the hours
leading up to 90 larger earthquakes. They
found that in the two hours or so preced-
ing an earthquake, the crust along the fault
begins to deform at a faster rate in the
direction of the earthquake rupture until
the instant the quake begins. What this
tells us, Bletery says, is that an accelera-
tion process occurs along the fault ahead
of the motion of the earthquake—some-
thing that resembles a slow earthquake.
“This does support the assumption
that there’s something happening before.
So we do have that,” he says. “But most
likely, it’s not physically possible to play
with the topic of prediction. We just don’t
have the instruments.” In other words, the
precursors may be there, but we’re cur-
rently unable to measure their presence
well enough to single them out before an
earthquake strikes.
Bletery and Nocquet conducted their
study using traditional statistical analy-
sis of GPS data; such data might contain
Without good knowledge information that’s beyond the reach of
of what’s happening under our traditional models and frames of ref-
erence. Seismologists are now applying
the ground, it’s impossible to intuit machine learning in ways they haven’t
any sense of order. before. Though it is early days yet, the
machine-learning approach could reveal
hidden structures and causal links in what
would otherwise look like a jumble of data.
66

In Italy, increased agitation among

animals was linked to strong earth-
quakes, including the deadly Norcia
quake in 2016.

Finding signals in the noise

Earthquake researchers have applied
machine learning in a variety of ways. Some,
like Mostafa Mousavi and Gregory Beroza
of Stanford, have studied how to use it on
seismic data from a single seismic station
to predict the magnitude of an earthquake,
which can be tremendously useful for early
warning systems and may also help clarify
what factors determine an earthquake’s size.
Brendan Meade, a professor of earth
and planetary science at Harvard, is able to
predict the locations of aftershocks using
neural networks. Zachary Ross at Caltech
and others are using deep learning to pick
seismic waves out of data even with high
levels of background noise, which could
lead to the detection of more earthquakes.
Paul Johnson of the Los Alamos
National Laboratory in New Mexico, who
became something between a mentor
and a friend after we met during my first
postdoc, is applying machine learning to
help make sense of data from earthquakes
generated in the lab.
There are a number of ways to create
laboratory earthquakes. One relatively
common method involves placing a rock
sample, cut down the center to simulate a
fault, inside a metal framework that puts
it under a confining pressure. Localized
sensors measure what happens as the
sample undergoes deformation.
In 2017, a study out of Johnson’s lab
showed that machine learning could help
predict with remarkable accuracy how long
it would take for the fault the researchers
created to start quaking. Unlike many
methods humans use to forecast earth-
quakes, this one uses no historical data—it
relies only on the vibrations coming from
the fault. Crucially, what human research-
ers had discounted as low-amplitude noise
turned out to be the signal that allowed The closer the animals were
machine learning to make its predictions.
In the field, Johnson’s team applied these
to the earthquake’s source, the more
findings to seismic data from Cascadia, advance warning their seemingly
where they identified a continuous acoustic
signal coming from the subduction zone
panicked behavior could provide.
that corresponds to the rate at which that
fault is moving through the slow earthquake
cycle—a new source of data for models

of the region. “[Machine learning] allows
you to make these correlations you didn’t
know existed. And in fact, some of them
are remarkably surprising,” Johnson says.
Machine learning could also help us
create more data to study. By identifying
perhaps as many as 10 times more earth-
quakes in seismic data than we are aware
of, Beroza, Mousavi, and Margarita Segou, a
researcher at the British Geological Survey,
determined that machine learning is use-
ful for creating more robust databases of
earthquakes that have occurred; they pub-
lished their findings in a 2021 paper for
Nature Communications. These improved
data sets can help us—and machines—
understand earthquakes better.
“You know, there’s tremendous skep-
ticism in our community, with good rea-
son,” Johnson says. “But I think this is
allowing us to see and analyze data and
realize what those data contain in ways
we never could have imagined.”
Animal senses
While some researchers are relying on the
most current technology, others are looking
back at history to formulate some pretty
radical studies based on animals. One
of the shirts I collected over 10 years of
attending geophysics conferences features
the namazu, a giant mythical catfish that
in Japan was believed to generate earth-
quakes by swimming beneath Earth’s crust.
The creature is seismology’s unoffi-
cial mascot. Prior to the 1855 Edo earth-
quake in Japan, a fisherman recorded
some atypical catfish activity in a river.
In a 1933 paper published in Nature, two
Japanese seismologists reported that cat-
fish in enclosed glass chambers behaved
with increasing agitation before earth-
quakes—a phenomenon said to predict
them with 80% accuracy.
Catfish are not the only ones. Records
dating back as early as 373 BCE show
that many species, including rats and
snakes, left a Greek city days before it
was destroyed by an earthquake. Reports
SIPA USA VIA AP
noted that horses cried and some fled
San Francisco in the early morning hours
before the 1906 earthquake.
Martin Wikelski, a research director
at the Max Planck Institute of Animal
Behavior, and his colleagues have been
studying the possibility of using the behav-
ior of domesticated animals to help predict
earthquakes. In 2016 and 2017 in central
Italy, the team attached motion detectors
to dogs, cows, and sheep. They determined
a baseline level of movement and set a
threshold for what would indicate agitated
behavior: a 140% increase in motion relative
to the baseline for periods lasting longer
than 45 minutes. They found that the ani-
mals became agitated before eight of nine
earthquakes greater than a magnitude 4,
including the deadly magnitude 6.6 Norcia
earthquake of 2016. And there were no false
positives—no times when the animals were
agitated and an earthquake did not occur.
They also found that the closer the animals
were to the earthquake’s source, the more
advance warning their seemingly panicked
behavior could provide.
Wikelski has a hypothesis about this
phenomenon: “My take on the whole thing
would be that it could be something that’s
airborne, and the only thing that I can think
of is really the ionized [electrically charged]
particles in the air.”
Electromagnetism isn’t an outlandish
theory. Earthquake lights—glowing emis-
sions from a fault that resemble the aurora
borealis—have been observed during or
before numerous earthquakes, including
the 2008 Sichuan earthquake in China,
the 2009 L’Aquila earthquake in Italy, the
2017 Mexico City earthquake, and even the
September 2023 earthquake in Morocco.
Friedemann Freund, a scientist at
NASA’s Ames Research Center, has been
studying these lights for decades and attri-
butes them to electrical charges that are
activated by motion along the fault in cer-
tain types of rocks, such as gabbros and
basalts. It is akin to rubbing your sock on
the carpet and freeing up electrons that
allow you to shock someone.
Some researchers have proposed dif-
ferent mechanisms, while others discount
the idea that earthquake lights are in any
way related to earthquakes. Unfortunately,
measuring electromagnetic fields in Earth’s
67
crust or surface is not straightforward. We
don’t have instruments that can sample
large areas of an electromagnetic field.
Without knowing in advance where an
earthquake will be, it is challenging, if not
impossible, to know where to install instru-
ments to make measurements.
At present, the most effective way to
measure such fields in the ground is to
set up probes where there is consistent
groundwater flow. Some work has been
done to look for electromagnetic and iono-
spheric disturbances caused by seismic and
pre-seismic activity in satellite data, though
the research is still at a very early stage.
Small movements
Some of science’s biggest paradigm shifts
started without any understanding of an
underlying mechanism. The idea that
continents move, for example—the basic
phenomenon at the heart of plate tecton-
ics—was proposed by Alfred Wegener in
1912. His theory was based primarily on the
observation that the coastlines of Africa and
South America match, as if they would fit
together like puzzle pieces. But it was hotly
contested. He was missing an essential
ingredient that is baked into the ethos of
modern science—the why. It wasn’t until
the 1960s that the theory of plate tectonics
was formalized, after evidence was found of
Earth’s crust being created and destroyed,
and at last the mechanics of the phenome-
non were understood.
In all those years in between, a growing
number of people looked at the problem
from different angles. The paradigm was
shifting. Wegener had set the wheels of
change in motion.
Perhaps that same sort of shift is hap-
pening now with earthquake prediction. It
may be decades before we can look back
on this period in earthquake research with
certainty and understand its role in advanc-
ing the field. But some, like Johnson, are
hopeful. “I do think it could be the begin-
ning of something like the plate tectonics
revolution,” he says. “We might be seeing
something similar.”
Allie Hutchison is a writer based in
Porto, Portugal.
68

US-based Noveon
Magnetics extracts
materials from
discarded commercial
magnets like this one
to make new magnets
for wind turbines and
electric vehicles.

GUTTER CREDIT HERE


Abandoning fossil fuels and adopting
lower-carbon technologies are our best
options for warding off the accelerating
threat of climate change. Access to rare
earth elements, key ingredients in many
of these technologies, will partly deter-
mine which countries will meet their goals
for lowering emissions or increasing the
proportion of electricity generated from
non-fossil-fuel sources. But some nations,
including the US, are increasingly worried
about whether the supply of those elements
will remain stable.
According to the International Energy
Agency, demand for rare earth elements
is expected to reach three to seven times
current levels by 2040; demand for other
critical minerals such as lithium may mul-
tiply 40-fold. Delivering on the 2016 Paris
Agreement, under which signatory nations
are obligated to reduce emissions to cap the
global temperature increase, would require
the global mineral supply to quadruple
within the same time frame. At the current
rate, supply is on track to merely double.
GUTTER CREDIT HERE
69
Obtaining rare earth elements begins
with obtaining source materials, which
can happen, broadly, in three ways:
primary extraction, or mining directly
from the earth; recovery from second-
ary sources, such as end-of-life electron-
ics; and extraction from unconventional
sources, including industrial wastes like
coal ash and waste products from mines.
But China so dominates the market—it
controlled 60% of global production in
2021—that other countries are at a dis-
advantage. After China announced export
restrictions in 2023 on gallium, germa-
nium, and graphite, nations scrambled to
find alternative sources in anticipation of
future restrictions.
Primary extraction in the US is limited;
only one active mine, the Mountain Pass
Rare Earth Mine and Processing Facility
in California, produces rare earth elements
domestically. Opening new mines can take
decades. As a result, scientists and compa-
nies alike are intent on increasing access
and improving sustainability by exploring
secondary or unconventional sources.
Inside the
race to make
critical
materials
China has dominated the market
for rare earth elements. US scientists
and companies are now scrambling to catch up.
By Mureji Fatunde
 70
Finding critical materials
All but one of the 17 rare earth elements
appear on a 2022 list of 50 designated “crit-
ical materials”—meaning they are econom-
ically important yet vulnerable to supply
disruption. The 17, such as praseodymium
(used in aircraft engines), gadolinium (used
in MRI imaging), and neodymium (used in
computer hard drives), include the “lan-
“I want to be one player in a big
ecosystem where there’s a lot of folks
producing rare earths. That’s the best
outcome for everyone.”
thanide series”—the 15 elements with
atomic numbers 57 to 71 near the bottom
of the periodic table—as well as two chem-
ically similar elements. The “rare” in “rare
earth elements” refers not to the quantity
available but rather to their wide disper-
sion—it’s hard to find an economically
meaningful quantity in a single location.
One unconventional source of rare earth
elements is coal ash, the residual solid
waste from burning coal at power plants.
Historically, coal ash has often been mixed
with water to form a slurry that is stored in
ponds (also called surface impoundments).
This ash, which contains elevated concen-
trations of rare earth elements, could be a
significant domestic source of the mate-
rials in former US coal towns, which face
challenges due to plant closures. There are
more than 1,000 coal ash ponds across the
US, mostly spread across the eastern part
of the country. One of the largest facilities,
Plant Barry in Mobile County, Alabama,
contains more than 21 million tons of ash
spread over 600 acres.
These ponds are not harmless; accord-
ing to the US Environmental Protection
Agency, improper management of them
can compromise waterways, groundwater,
drinking water, and air via contaminants
such as mercury, cadmium, and arsenic.
A document submitted by Earthjustice, a
nonprofit environmental law organization,
and Earthworks, a nonprofit focused on
preventing the destructive impacts of oil,
gas, and mineral extraction, responding to
a 2023 request for information from the
US Department of Energy, noted that “91%
of power plants storing coal combustion
residuals (CCRs) are polluting the under-
lying groundwater to levels that exceed
federal drinking water standards.” Ponds
can also be destabilized during extreme
weather events, and the resulting flood of
contaminated material can destroy wildlife,
damage property, and threaten community
health and safety.
A startup, Rivalia Chemical, believes
the health hazard posed by ash ponds
can be addressed by repurposing ash to
create a domestic supply of rare earth
elements. Laura Stoy, the environmental
engineer who founded Rivalia in 2021, says
she is motivated by both environmental
concerns and the potential for economic
revitalization.
Stoy began developing Rivalia’s flag-
ship technology during graduate school
at the Georgia Institute of Technology
and is now working to scale it within the
Chain Reaction Innovations program at
the DOE’s Argonne National Laboratory.
In 2019, Georgia Tech supported the bud-
ding company in filing a patent (currently
pending) for its technology, for which
Rivalia holds an exclusive license.
That technology extracts rare earth
elements from coal ash, leaving behind a
solution rich in those elements and a resid-
ual solid containing iron and other metals.
Through sequential steps of heating and
cooling, rare earths are transferred into
an ionic liquid—a salt in liquid state—
via a proton-exchange mechanism. Acid-
based reduction techniques and salt-based
leaching can reduce the amount of iron in
the final solution, after which rare earths
must be further separated to produce pure
metals or oxides. Rivalia can sell primary
outputs to companies that handle subse-
quent processing steps, manufacturers
using rare earths, and sell residual solids
to concrete producers. Stoy says Rivalia’s
efforts will produce materials that could
be used for cleaner products and alterna-
tive energy sources. Furthermore, they
could help reduce the carbon footprint
of concrete production by repurposing
the solid residue as a replacement for
emission-heavy Portland cement—a major
ingredient in concrete. (For more on this,
see page 52.)
Rivalia prefers to work with existing
waste products as opposed to coal that
has not yet been burned. This approach
is risky; extraction from unconventional
sources can cost more than mining, given
the low concentrations of rare earth ele-
ments and the greater initial concentration
of toxic contaminants.
Still, Stoy says, this is a strategic move
in light of the need to diversify supply.
It’s also an opportunity to make use of a
widely available material with few alterna-
tive uses and significant economic value;
the value of rare earth elements in US coal
ash reserves was previously estimated at
$4.3 billion (based on 2013 prices) and has
likely grown since then. As a fairly new
startup, the company is still in the R&D
stage and is currently focused on reducing
extraction costs.
The race to produce rare earth ele-
ments domestically in the US is, at least
partially, an attempt to figure out how to
do so economically; however, companies
PREVIOUS SPREAD: COURTESY OF NOVEON MAGNETICS
are unlikely to get production costs low
enough to be able to compete on price
alone. Experts hope consumers will be
willing to pay a premium, partly absorbing
the increased costs.
“Hopefully there is a market for a
domestically produced material that’s
produced in an environmentally con-
scious manner and an ethical manner
that’s respectful of the workers producing
the material,” says Evan Granite, program
 71

manager for the carbon ore program at the
DOE’s Office of Fossil Energy and Carbon
Management.
Regulators have started addressing the
coal ash problem, so startups hoping to use
the material will need to watch ongoing
developments closely. The EPA began reg-
ulating the management of coal ash ponds
in 2015 following destructive spills in 2008
and 2014. A recently proposed update to
the 2015 rule mandates that older, inac-
tive ponds that were previously exempt
be covered or excavated.
Following the 2015 regulation,
Earthjustice said that closing ponds by
capping them in place is insufficient if
they are within five feet of groundwater,
and that in such cases only full excava-
tion will prevent future damage. Either
option—capping or excavation—would
make coal ash harder to access for com-
panies like Rivalia. Stoy says she considers
this a reason to move decisively.
Stoy says she is wary of inadvertently
creating new markets for coal by-products,
which could jeopardize the country’s
clean-energy ambitions. Ironically, if
utilities stopped using coal, Rivalia’s
source materials would eventually dry
up. However, she isn’t worried just yet—
even in the absence of new production,
the US now has 2 billion metric tons of
ash, and many other countries seem likely
to continue burning coal for the foresee-
able future.
Handling all that ash will have to be
done with care, says Lisa Evans, senior
counsel in the clean-energy program at
Earthjustice. Evans says that even for
companies motivated by cleanup hopes,
additional regulatory oversight is needed to
ensure they dispose of by-products appro-
priately. “What I’ve experienced in so many
years of looking at how industries behave
is that they don’t do anything they’re not
required to do,” she says, adding that the
government should also ensure that com-
munities receive adequate notice of nearby
extraction activities.
Modernizing extraction
Another unconventional source of critical
materials is tailings—the waste products of
mines themselves. The EPA does not yet
regulate mine tailings, even though they
are similar to coal ash in the environmental
risks they pose, says Evans of Earthjustice.
Phoenix Tailings is a Massachusetts-
based startup extracting rare earth ele-
ments from mining sites. Two of Phoenix’s
founders, who grew up in communities
affected by mining, say they are motivated
by personal experience in addition to the
growing demand for rare earth elements.
Besides the four rare earths used most
commonly in magnets (neodymium, pra-
seodymium, dysprosium, and terbium),
Phoenix recovers battery metals, plati-
num group metals, low-carbon irons, and
other materials in what it calls a “portfolio
approach” that improves economic via-
bility. Like Rivalia, Phoenix repurposes
residual materials into concrete and other
aggregates. This, the company says, pro-
vides long-term storage for carbonaceous
materials, reducing environmental impact
by trapping them and preventing them
from ending up in the water supply.
Phoenix works to modernize extraction,
reducing the amount of energy, equip-
ment, and funding required, says

But wait, You’re already a subscriber.

Activate your account
and start enjoying:

there’s • Unlimited web access

• Exclusive digital stories
• Access to 120+ years

more.
of publication archives

technologyreview.com/subonly
72
cofounder Anthony Balladon. “You develop
chemistries that are tuned for the rare
earths, as opposed to trying to brute-force
your way through them,” he says.
After obtaining an oxide concentrate
containing the rare earths, Phoenix uses
separation techniques to draw out the
desired end products. This is followed
by reduction into final metal and alloy
products using mixed-halide molten-salt
electrolysis, resulting in 35% to 45% lower
energy requirements. Chief technology
officer Tomás Villalón says Phoenix’s pro-
cess reduces the amount of material inad-
vertently lost between processing steps and
improves the purity of the final product.
Phoenix’s founders also highlight the sus-
tainability of the company’s process, which
they say uses no hazardous materials and
creates zero direct carbon emissions. The
company is currently producing rare earth
metals for commercial clients and expects
to be producing over 3,000 tons per year
of finished rare earth metals by 2026.
Villalón estimates that Phoenix will
be busy for a long time: at least 10 billion
tons of mine tailings are created each year
from new activity.
Increased demand for magnets
Some companies target recycled mate-
rials rather than coal wastes as a source
of recoverable rare earths. Noveon
Magnetics—formerly Urban Mining—
extracts critical materials from discarded
commercial magnets (from motors or
medical devices, for example, or from
storage drives used by data centers) or
those withdrawn from the supply chain
because of manufacturing defects or obso-
lescence. From these materials, Noveon
manufactures new sintered neodymium
boron magnets, critical components of
generators in wind turbines and motors
in electric vehicles.
According to DOE projections, US
demand for these rare earth magnets is
set to more than quadruple by 2050. This
is partly because of improved industrial
technologies, says Noveon’s chief com-
mercial officer, Peter Afiuny. “Industrial
pumps, compressors, HVAC systems …
50% of our electric consumption is being
driven by those motors. If you’re talking
about getting to carbon neutral, you need
to upgrade those systems and make them
more efficient,” he says.
There are fewer than 10 active magnet
manufacturers outside China; Noveon is the
only one in the US. Afiuny says it acquires
all its materials domestically.
The company produces a new type of
high-performance magnet, which it calls
“EcoFlux,” using less material than con-
ventional versions, says Afiuny. While it’s
hard for recycled magnets to perform as well
as nonrecycled products, Afiuny says that
Noveon has managed the feat by combin-
ing a proprietary technology that improves
the composition and properties of mag-
netic materials with its patented Magnet-
to-Magnet technology that can recycle
up to 99.5% of input materials. He adds
that Noveon has multiple customers and
produces at commercial scale in its Texas
facility. He says the company plans to pro-
duce 10,000 tons a year within five years.
These new magnets serve the same
types of customers from which the materials
were collected—such as companies using
motors to power consumer electronics and
medical or automotive products. The result
is a loop of reuse.
Can these alternative sources replace
existing imports? In a recent paper
published in the National Academy of
Engineering’s magazine, The Bridge, DOE
researchers estimate that for some critical
materials such as germanium, coal ash can
meet US demand for nearly 4,000 years, but
for most materials, the supply will last for
less than 20 years (and for nickel, for just
a little more than one year).
Additional new sources are needed, says
Granite: “You’re going to need many dif-
ferent waste materials and nontraditional
sources to meet the long-term demand,
because we project growing demands for
many of these critical metals.”
The researchers suggest that a much
broader range of waste sources could be
considered, including “red mud,” cre-
ated during aluminum production, and
“produced waters,” which result from oil
production, as well as materials sourced
from the ocean floor or even outer space.
A universal policy priority
Between 2015 and 2021, the DOE awarded
at least $27 million to projects related to
extracting rare earth elements from both
conventional and unconventional resources.
In 2022 and 2023, the government
announced at least $1 billion of funding
available to support related work, includ-
ing significant amounts from the Bipartisan
Infrastructure Law. Other agencies have also
announced support for companies working
to help boost the nation’s supply of critical
materials, signaling a renewed sense of
urgency for a longtime item on the policy
agenda. Rivalia, Phoenix, and Noveon have
all benefited from government support,
suggesting that the government is willing
to place bets on companies at varied sizes
and stages of progress.
These funding allocations often reveal
the priorities of the issuing administration;
the focus under former president Donald
Trump, for example, was independence
from China, while the Biden administra-
tion’s support for domestic production of
rare earths seems more tied to its push
for wider adoption of electric vehicles.
Regardless of motivation, all parties seem
aligned on the importance of rare earth
elements.
“It’s something that’s broadly supported
in a bipartisan way,” says Rivalia’s Stoy. “It’s
something that I think is very safe from a
research funding perspective. The govern-
ment is interested in this and is going to be
funding it for a long time.”
As the race to achieve self-sufficiency in
rare earth elements and critical materials
intensifies, the US is likely to further expand
both the number of organizations involved
and the diversity of potential sources.
Despite growing competition, Stoy says
there’s room for everyone. “I want to be one
player in a big ecosystem where there’s a
lot of folks producing rare earths,” she says.
“That is the best outcome for everyone.”
Mureji Fatunde is an academic and
writer who explores how companies
and consumers make decisions.
Expand your
knowledge
beyond
the classroom.
Invest in your future with a student subscription
and save 50% on year-long access to
MIT Technology Review’s trusted reporting,
in-depth stories, and expert insights.

Scan this code to access your 50% student savings

or learn more at TechnologyReview.com/StudentOffer
74
Displays are
G
oogle Glass, a prototype augmented-reality headset
released in April 2013, had the makings of a hit. It
promised intuitive, hands-free access to a smart-
phone’s most important features—video recording,
navigation, and even email. Forget touch screens
and buttons: the future of computing was on your face.
It was a disaster.
Though beautiful in concept, Glass was awkward to wear and
struggled to deliver a sharp, bright image outdoors. Then came
the “glasshole” backlash. The size of the display made wearers
easy to spot in a crowd and, on at least two occasions, led to
physical altercations.
The implications were clear. Hands-free augmented reality
(AR) was fun on paper, but with tensions over Big Tech’s influ-
ence mounting, it couldn’t overcome the stigma of making peo-
ple look like extras in a cyberpunk flick.
Now, more than a decade later, the future Google envisioned—
and much more—is on the brink of becoming reality. Tiny new
displays, some small enough to fit on the tip of your finger, will
contain micro-LEDs and micro-OLEDs (organic LEDs). They
are set to deliver a wave of headsets that may convert even the
most ardent AR skeptics.
Apple’s Vision Pro, slated for release in 2024, will lead this
change—though it might not shake the cyberpunk aesthetic.
The fully enclosed headset, vaguely reminiscent of ski goggles,
is intended for a mixture of AR and virtual reality (VR) that Apple
calls “spatial computing.”
The Vision Pro avoids some of the problems Google Glass
faced by narrowing the product’s scope. Apple hopes the headset
might replace a computer, tablet, and TV—though only within
the confines of your home or office.
The real innovation is inside: a pair of micro-OLED displays
no larger than a postage stamp that pack 4K resolution into a
screen just 1.3 inches square. Each display contains more than
11 million pixels spaced only 6.3 micrometers apart—less than
the diameter of a human red blood cell.
COURTESY OF APPLE
It’s a spectacular upgrade. Apple’s Vision Pro, like the Meta
Quest 3 and the HTC Vive XR Elite, uses cameras to replicate
the outside world on internal displays, a technique known as
pass-through mixed reality. But its competitors use liquid crystal
 75

LEDs and
their organic
counterparts
are getting
truly tiny.

shrinking before our eyes

This could
be the
moment
augmented-
reality
companies
have been
looking for.

By
Matthew
Smith

displays that lack the sharpness to faithfully reproduce the world
around you, so tasks that should be simple, like glancing at a
handwritten note, can prove difficult.
“I think overall they’ve achieved something impressive,”
says Anshel Sag, principal analyst at Moor Insights & Strategy.
“This is the headset that you build if you want people to really,
fully understand what the maximum potential of AR and VR is.”
Sag believes the individual pixels on Vision Pro displays will be
invisible to most people, “unless you have extremely impressive
visual acuity, like 20/10.”
The Vision Pro’s pixel-dense displays are widely believed to
be the culmination of years of work from Sony’s Semiconductor
Solutions Group. The division’s micro-OLED adventures were
originally focused on colorful high-resolution digital viewfinders
for cameras like the Sony SLT-A77. The group also built them
for a head-mounted device, the HMZ-T1 Personal 3D Viewer,
that Sony launched in 2011, pitching it as a movie-theater-like
experience for watching video.
The HMZ-T1 headset performed best with 3D films, which
proved to be a fad. But Sony didn’t give up on micro-OLEDs,
and in 2018 it announced a 0.5-inch micro-OLED display that
reduced the distance between pixels from 7.8 to 6.3 micro-
meters (the same as the larger displays found in the Vision Pro),
an innovation made possible by a breakthrough that placed the
color filter closer to the OLED’s light-emitting organic material.
With a display this small, any subtle change in the angle of light
emitted from red, blue, and green subpixels can hurt color per-
formance. Moving the color filter improves the viewing angle
of each pixel, which makes a smaller display possible without
compromising image quality.
Micro-OLEDs benefit from some of the traditional strengths
of light-emitting diodes made with organic films. Each pixel is
self-emissive, which means its brightness is zero when it’s “off.”
The LCDs in most headsets can’t achieve this, and as a result,
darker scenes have a hazy, gray glow. And when micro-OLEDs
are on, they’re on. The Vision Pro’s displays are quoted at a peak
brightness of 5,000 nits, the industry’s go-to measure of bright-
ness. It’s a 50-fold improvement over Meta’s Quest 2, which hits
just 100 nits. (Meta hasn’t revealed the Quest 3’s brightness, but
it’s likely similar.)
76

The Vision Pro is likely to quicken the adoption of micro-OLED Most monolithic micro-LED displays are currently monochrome,
technology. But despite its many strengths, those miniature OLEDs meaning they display a single color (usually red, blue, or green).
still have some shortcomings. Michael Murray, CEO of Kopin, a But full-color micro-LED displays are right around the corner.
display company in Westborough, Massachusetts, notes that micro- Mojo Vision hopes to have a color micro-LED prototype ready in
OLED displays are excellent for moving images, such as movies, early 2024, and one of its competitors, Shanghai-based Jade Bird
but sometimes less so for static text—a reason, he says, why Meta’s Displays (often referred to by its initials, JBD), has demonstrated
Quest headsets have stuck with LCD. While micro-OLED displays a functional color micro-LED prototype with a pixel pitch of five
can be bright, the organic molecules inside them can degrade over micrometers—larger than what Mojo Vision hopes to achieve, but
time, a phenomenon known as burn-in. Micro-OLED also fails to smaller than Apple’s Vision Pro.
entirely resolve the design issues of Google Glass: the display is The key benefit of smaller, denser pixels is the reduction of dis-
improved, but the headset is even more conspicuous. play size at any given resolution, which in turn reduces the size and
Fortunately, micro-LEDs offer a solution. weight of an AR headset. JBD’s monochromatic AmuLED series,
for example, achieves 640 x 480 resolution
Truly microscopic on a display a carpenter ant could carry on
Micro-OLED and micro-LED displays its back—with room to spare.
differ in the details, but their production Micro-LEDs also score a massive win
shares broad similarities. Both pair a sili- in brightness. The range is from 1.8 million
con “backplane,” which provides structure up to 3 million nits, Murray says: “It will
and power, with a display “frontplane” that literally tear the retina out of your eye and
creates visible light. Each is named for the blind you for life.” The brightest OLED
type of frontplane used: a layer of organic displays, by comparison, currently peak
material that emits light in response to at around 15,000 nits.
an electric current in the case of micro- The possibility of permanent eye dam-
OLED displays, and a very small array of age might seem an odd perk, but not to
electronic diodes made from semiconduc- worry—no one will be looking at the micro-
tors in the case of micro-LEDs. LEDs directly. Placing a display directly
Micro-LED display technology is not in front would block the wearer’s view of
Mojo Vision hopes to have
as mature as micro-OLED, but the possi- a color micro-LED prototype the real world, so many AR devices place
bilities are alluring. “Micro-LED happens ready in early 2024. the display to the side. Waveguides then
to be the best of all worlds,” says Murray. redirect the light from the offset display
“It has the best display quality, it has lon- to make it visible. This process can prove
gevity, doesn’t have burn-in issues, has high brightness that you tremendously inefficient, especially for modern AR glasses like the
can control … that’s where the future is going.” Magic Leap 2 and Vuzix Blade 2, which focus and redirect light
Mojo Vision, a display technology company based in Saratoga, through multiple waveguides arranged like mirrors in a fun house.
California, was among the first companies to realize the LED’s “[The efficiency] is something like 5% to 10%,” says Michael
potential in tiny devices. It made waves in 2020 with a contact Miller, augmented-reality hardware lead at Niantic. “If you have
lens with a flexible, transparent AR display. The company has a display of 3,000 nits, you will get 300 nits out. You can put a
since abandoned the contact lens to focus just on the display, dark lens on top of it so you can maybe use it outdoors, but it’s
and in 2023, Mojo Vision demonstrated micro-LED displays not good enough.”
with an astounding 28,000 pixels per inch. That works out to a Displays built from micro-LEDs should be able to make it
pixel pitch—the distance between the centers of two adjacent through a gauntlet of waveguides and still be bright enough to be
pixels—of just 1.87 micrometers, smaller than some bacteria viewed on transparent lenses that look just like prescription ones.
and a third the size of what you’ll find in the Apple Vision Pro.
Such extreme pixel density is the result of a fundamental shift in Awesome performance, awesome cost
micro-LED design. The first micro-LED displays were built with a Headsets with cutting-edge displays, like the upcoming Vision
technique called “mass transfer.” Red, blue, and green LEDs were Pro, thrash the performance of mass-market VR headsets. They’re
produced on wafers and transferred one by one to a display sub- also more expensive: the Vision Pro will retail for $3,499.
COURTESY OF MOJO VISION

strate (a technique that is still used to make larger displays). But The displays deserve some of the blame.
small micro-LED arrays, like those produced by Mojo Vision, take Each micro-OLED display can cost $400 to manufacture,
a monolithic approach: the micro-LEDs and the silicon backplane says Murray. “If you’re building a Meta Quest, or something like
are bonded in a production pipeline like that used to manufacturer it, you need two of them,” he says, “and your bottom-line cost is
cutting-edge computer chips. already $800.”

HMDmd’s Model CR3 headset (top) is designed
specifically for surgeons. Like other AR glasses,
the Vuzix Blade 2 (bottom) redirects light from the
side to the front through multiple waveguides.
The big price tag attached to such small displays might seem
strange. After all, OLED displays are a mature technology found
in hundreds of millions of smartphones, tablets, and televisions
worldwide. LEDs are even more ubiquitous: just flip a nearby light
switch to see one in action. These are well-understood technolo-
gies found in many affordable devices.
On this diminutive scale, however, building a display is no
longer a job for a factory. It requires a foundry—a specialized
chip-manufacturing facility.
Costs could come down as manufacturers shift to building the
displays on larger silicon wafers. Larger wafers are more expen-
sive, but each one can pack more displays, which lowers the cost
COURTESY OF HMDMD; COURTESY OF VUZIX
of each display. Micro-OLED makers are in the midst of a shift
from eight-inch wafers to 12-inch wafers, which is the standard
in high-volume, cutting-edge silicon manufacturing. Micro-LED
production is less mature, with some companies relying on inef-
ficient four-inch wafers.
Producing usable displays with the extreme pixel densities
that micro-OLEDs and micro-LEDs can achieve is a challenge.
The fundamental problem is a defect you’ve likely witnessed more
than once: the “dead” pixel. A dead pixel displays one color—often
77
perfect black or a bright, blinding white—and refuses to respond
to display signals. Avoiding this defect is already difficult for smart-
phone displays, where pixels might be separated by 500 microm-
eters. With monolithic micro-LEDs, the smallest, most densely
packed displays ever produced, the slightest flaw in the silicon, or
the slimmest sliver of debris, can render a display useless.
“Here’s the scary math,” says Murray. “The amount of usable
displays after you’re done is probably a tenth of the usable silicon
you started with.” In other words, more than 90% of a silicon wafer
could be wasted. Yet the company producing the micro-LED displays
still pays for the entire wafer—adding huge costs to each display.
Micro-LED pioneers are investing in tools and processes that
reduce the steps involved in production. That’s critical, because
the more complex the production process, the higher the risk of
introducing a defect.
Soeren Steudel, CTO of the Belgian display developer Micledi,
is hyper-focused on this problem. The company has partnered
with the semiconductor manufacturer GlobalFoundries and plans
to move production there to reduce costs. “Micro-LED is not yet
a mature product. It was a wild dream 10 years ago, and now the
first companies have demos,” says Steudel. “The question now is,
how can you manufacture that in volume without defects?”
Augmented reality finally goes mainstream
The difficulty of producing micro-OLED and micro-LED dis-
plays is high, but the problems are worth solving. These displays
could make AR a virtual space people can easily and quickly
access in their daily lives—not only because the displays appear
more lifelike, but also because small, thin, high-quality displays
give engineers more freedom to tailor a headset’s look and feel.
The impact of micro-OLEDs is already apparent. Kopin pro-
duces displays for HMDmd’s Model CR3, a headset designed for
surgeons, and defense projects, such as an AR weapon sight for the
M1 Abrams main battle tank. XReal, another AR pioneer, recently
released its Air 2 headset, which packs Sony micro-OLED displays.
The possibilities for the future could be even more dramatic.
The displays’ extreme brightness, diminutive size, and low power
consumption could unlock the dream of light, attractive, fully
transparent AR glasses that don’t immediately stand out from
conventional eyewear.
“People want to go to consumer augmented reality. And consumer
AR means that you have lightweight glasses, maybe 50 grams, and
that you don’t look like Darth Vader,” says Steudel. Vuzix, a leader
in lightweight VR headsets, has achieved this with the Ultralite,
a prototype platform revealed in 2023 that uses micro-LED tech-
nology to provide sleek, slim spectacles that weigh just 38 grams.
Augmented reality still needs its “iPhone moment”—the debut
of an easy-to-use device that offers irresistible benefits. Better
displays will make AR—if it ever gets widely adopted—bright,
sharp, convincing, and—most important of all—pleasant to use.
Matthew Smith is a freelance technology journalist based
in Portland, Oregon.
Help
shape the
future
of tech.
Take part in original research and gain valuable business
insights into today’s most important technology trends.
Join the Global Insights Panel.

• Participate in • Receive the • Share experiences, • Receive special

research programs latest tech news weigh in on ideas promotions and
and surveys and updates in from Insights, and discounts for MIT
(each takes about our exclusive form collaborations Technology Review
5 to 10 minutes to newsletters. via our Global conferences and
complete). Insights Panel subscriptions.
LinkedIn group.

Scan this code or visit us at the link below to join the

Global Insights Panel today for free:
technologyreview.com/GlobalPanel

Recapturing
the whimsy of the
early internet
W
e b s i t e s w e r e n’ t
always slick digital
experiences.
There was a time
when surfing the
web involved open-
ing tabs that played music against your will
and sifting through walls of Times New
Roman text on a colored background. In
the 2000s, before Squarespace and social
media, websites were manifestations of
individuality—built entirely from scratch
using HTML, by users who had some
knowledge of code and a desire to be on
the internet.
Scattered across the web are communi-
ties of programmers working to revive this
seemingly outdated approach. Anchored
in the concept of “HTML Energy,” a term
coined by artists Laurel Schwulst and
Elliott Cost, the movement is anything but
a superficial appeal to retro aesthetics. It
focuses on the tactile process of coding
in HTML, exploring how the language
invites self-expression and empowers
individuals to claim their share of the
web. Taking shape in small Discord chan-
nels and digital magazines, among other
spaces, the HTML Energy movement is
about celebrating the human touch in
digital experiences.
oday, the majority of the internet is
T optimized for social engagement,
e-commerce, and streaming. Most
internet traffic is concentrated in a small
number of sites, all of which are owned
by the same handful of companies. From
lengthy ads to aggressive cookie set-
tings, minor obstacles and nuisances are
baked in. Users are constantly reminded
that their access to the internet is condi-
tional on the monetary interests of a few.
The situation with X (formerly known as
Twitter) perfectly encapsulates this state
of internet ownership: it only took one
executive to spark a mass exodus from
the platform and to fragment its long-
lived communities.
However, despite the monopolistic
landscape of Big Tech, one fundamental
reality continues to justify the internet’s
democratic reputation: anyone can pub-
lish a website for free with HTML. With
an abundance of real estate, the web tech-
nically has space for everyone. It’s just a
matter of traffic.
When I spoke to different members
of the HTML Energy community, all
consistently returned to one basic mes-
sage: Everything on the web boils down
to HTML. HTML is the backbone of any
website. It’s the only thing needed for a
website to run. While popular web devel-
opment languages today use abridged
commands that hide technical complexity
through what’s known as data abstraction,
HTML is granular, and previous coding
knowledge is not a prerequisite.
As Cost explains, it is precisely how
forgiving HTML is that gives eager indi-
viduals an opportunity to self-publish
on the web. With HTML, a site will still
load even if a line of code is missing. The
79
Modern forms of coding make
most websites feel like commercial
transactions. The HTML Energy
movement aims to bring back the
joys of the early days.
By Tiffany Ng
HTML Energy movement embraces these
possibilities: learning via trial and error is
welcomed, and creative experimentation
is encouraged.
ince the rise of site-building
S tools like Wix, the intricate and
sometimes clunky experience of
hard-coding fonts or pixel spacing into a
site has been replaced by templates and
conventions of user experience design.
As mainstream digital experiences trend
toward a homogeneous visual language,
the human touch gets lost in the many
layers of abstraction. Site creators grow
more distant from their sites, and the web
becomes more transactional.
But the HTML Energy movement calls
on people to reexamine our relationship
with technology. Crafting a site using
HTML allows programmers to explore
what a website can be. Unlike their cor-
porate counterparts, people creating sites
on their own don’t answer to shareholders.
They don’t have the pressure to create
profitable experiences, so their creations
can take an endless variety of forms.
Common types of HTML Energy sites
include digital gardens, where elements
change with the seasons; interactive poetry
generators, where inputs from the user
give rise to new meaning; and personal
sites that share intimate details about
their creators’ lives. In an internet that is
increasingly consumerist, HTML Energy
sites offer a gentle reminder that websites
can be meditative experiences.
80

The HTML Energy community

advocates understanding HTML for
what it quite literally is: a language. And
it celebrates the way the rudimentary
character of that language demands
intention from the user. As an amal-
gamation of minuscule and intricate
creative decisions, a site constructed
using only HTML is a form of self-
expression. Viewing a site’s source
code is as important as navigating its
interface. There are often Easter eggs
hidden in that code, such as cheeky
messages or citations taken from other
HTML sites. In many ways, an HTML
site captures something of the cre-
ator’s identity: what did that individual
choose to build, and how?
This fascination with different appli-
cations of HTML is also seen in phys-
ical community gatherings sometimes
called “freewrites,” where members of
the movement get together to write
code. Sunday Sites and Fruitful School
PROJECT 2:
are among the websites that organize
these gatherings, often integrating A Room with a Window
educational elements into their ses- Above: Shelby Wilson’s A Room with a
sions to empower more people to join Window is a site that allows for only one
interaction: opening and closing a set of
the movement. Meanwhile, sites like
window shades. The site intentionally con-
HTML Review showcase some of its flates physical and digital spaces: Wilson
products in the format of a literary plays with the idea of a browser as a por-
magazine. tal to a place with physical boundaries
and edges, but also maintains surrealist
components (the room doesn’t get darker

T
here is no centralized source for
when the blinds close) and randomized
HTML Energy sites: serendipity elements (the color of the room changes
makes them finding them feel on each visit) to highlight the digital form.
special, like happening upon a piece of
street art behind a parking lot. They’re PROJECT 1: PROJECT 3:

not designed for discovery, nor are they
optimized for any particular action.
They simply engage with a visitor on
the visitor’s terms, offering a portrait
of their creator. If sites like Google or
Facebook are the supermarkets and
shopping malls where you buy your
necessities, HTML Energy sites are
like the hidden gardens you happen
upon, unmarked on any map.
Tiffany Ng is a freelance writer
exploring the relationship
between art, tech, and culture.
Terrarium of Many HTML Garden
Sceneries Right: Spencer Chang’s site imagines what
Above: Ji Kim’s Terrarium of Many a garden might look like on the internet.
Sceneries collages snippets of footage Several “plants” made of native HTML
from an old iPhone. As visitors scroll elements grow, and the passage of time is
through the site, images overlap and acknowledged and noticeable upon each
embedded audio clips play. When users visit—seasons change, plants sprout and
click any image, a small description of bloom. There’s no explicit action called
when and where it was taken appears, for—just observation.
alongside more accompanying media.
Kim’s site is designed to mimic the
sporadic, layered nature of memory. It
is a digital experience that is intention-
ally fragmented and overwhelming—like
trying to remember a family trip taken
years ago.
 81

PROJECT 4:

Prose Play
Left: Katherine Yang’s Prose Play
is an interactive poem that encour-
ages users to input different words
into a pre-set sentence structure.
Framing words as variables, the
site explores the interactivity of the
internet. It puts the literary theory of
the “Death of the Author”—the idea
that the meaning of a text is not
determined by the author’s intention
but by the reader’s interpretation—
in the context of code.

PROJECT 5: PROJECT 6:

Erich Friedman Museum of Screens

Below: Erich Friedman’s site is a per-
sonal encyclopedia of his life, with
archives of everything from movie rat-
ings to reviews of mini-golf courses
across central Florida. Organized into
the categories of Math Stuff, Puzzle
Stuff, Personal Stuff, and Professional
Stuff, the site is simple in structure.
It uses basic HTML to showcase
Friedman’s eclectic interests over
the past decade, including a list of
fun facts for every number from 0 to
9,999 and collections of math and
trivia problems.
The site does not drive any spe-
cific action. It merely stands as an
exhaustive, candid portrait of Erich
Friedman, occupying a small piece
of the web.
Above: Toulou TouMou’s
Museum of Screens is a site that
houses browser games created
by game enthusiasts. In order
to interact with the games on
display, users have to navigate
the digital space like a physi-
cal museum visualized in ASCII
graphics. There are actual visit-
ing hours, with a “rest day” cho-
sen at random.
Created to give due credit to
amateur developers during the
era of Flash games, TouMou’s
museum aims to highlight the
importance of acknowledging
authorship and the rich history
of independent games.
82
A new history of scientific
inquiry invites us to marvel at the
elegance of experimentation.
By Philip Ball
Above: Evangelista Torricelli’s studies of air pres-
sure in the 1640s revealed the extraordinary pres-
sure that pushes down on everything at Earth’s
surface owing to the weight of the atmosphere—
what he called an “ocean of air”—above it.
Opposite: Torricelli’s studies led to the invention
of the barometer for measuring atmospheric
pressure. His ideas were verified in 1648 by the
French philosopher Blaise Pascal, who had his
brother-in-law carry a primitive barometer (an
inverted tube filled with mercury) up a mountain
in the French region known as the Massif Central
and observe that the air pressure dropped at the
higher altitude.
What is a beautiful
experiment?
Some philosophers have argued that
“beauty” in science stands as a proxy for
truth. Many scientists have agreed. The
British physicist Paul Dirac, for example,
claimed that it is more important for a the-
ory to be beautiful than for it to conform
with experimental tests. And Einstein
stated that “the only physical theories
we are willing to accept are the beautiful
ones.” Such aesthetic judgments seem
a little shallow, however, and also peril-
ous: we might be tempted to place undue
trust in an idea simply because we deem
it beautiful. Indeed, other scientists are
skeptical that perceptions of beauty are
any guide to validity; what matters in the
end is whether a theory fits what we see
in the real world. The 19th-century British
zoologist Thomas Henry Huxley said that
the great tragedy of science is “the slaying
of a beautiful hypothesis by an ugly fact.”
But hypotheses and theories are not the
only sources of scientific beauty.
Experiments are at the core of science.
It is typically by experimentation that
scientific progress is made: say, from the
discovery of the first virus in 1892 to the
creation of covid-19 vaccines in 2020. We
might be tempted to assume, then, that the
process by which experiments lead to reli-
able and useful knowledge is well under-
stood. But that’s not really so. In putting
together my book Beautiful Experiments, a
(highly selective) history of experimental
science, one of my aims was to show that
scientific knowledge has not been steadily
churned out from the well-oiled machinery
of experimental methodology; the emer-
gence of robust theories and concepts from
empirical investigations is altogether more
haphazard, and more interesting.
We can, as I do here, talk about some
experiments as being “beautiful,” though
what that means isn’t easily stated. There
are many potential aesthetic virtues in an
experiment: beauty of concept, economy
of instrumental design, the aptness with
which the two are aligned, and elegance of
reasoning in interpreting the results. These
are qualities that require creativity and imag-
ination—there is no prescription for them.
There are scientists who seem to have a
talent for aesthetically pleasing experimen-
tal design—none more so than the New
Zealand–born physicist Ernest Rutherford,
who in 1908 discovered the dense atomic
nucleus by scattering alpha particles from
gold foil and subsequently proposed the
“solar system” model of the atom, in which
subatomic particles orbit one another in
mostly empty space. A beautiful experi-
ment marshals the available resources to
disclose what casual inspection will not.
For example, many biologists consider the
1958 experiment by Matthew Meselson and
Franklin Stahl on how DNA replicates to be
the most beautiful in their discipline. Their
trick was turning a seemingly impossible
puzzle—how to distinguish between pos-
sibilities whose outcomes look identical—
into a soluble one by using atomic isotopes
to create DNA strands that are chemically
identical but physically separable.
An elegant experiment can look like
a collaboration with nature to uncover
“something deeply hidden,” as Einstein put
it. Physics Nobel laureate Frank Wilczek
has suggested that beauty in a scientific
idea becomes manifest when you get out
more than you put in: the idea delivers
something new and unexpected.
When an experiment does that, we see
science at its most magical: we ask a ques-
tion of the universe, and it tells us some-
thing more. Every scientist longs for such
moments, and treasures them if they come.
This story was excerpted from
Beautiful Experiments: An
Illustrated History of Experimental
Science (2023).
LIBRARY OF CONGRESS
GUTTER CREDIT HERE
 85

Below: Known for his theory of evolution by nat-
ural selection—the central organizing framework
for understanding life—Charles Darwin was also
an astute observer of nature and an avid exper-
imenter, driven by a deep curiosity about the
living world. His experiments on earthworms,
conducted in his garden from the 1870s to the
1880s, showed how efficiently they manipulated
leaves to plug the holes they make in the soil as
protection against predators or rain. Even the
humble worms, said Darwin, “show some degree
of intelligence.”

Opposite: In the mid-18th century, the Italian

anatomist Luigi Galvani suspected that
electricity held the secret of life after he
observed that a current discharged from
a storage device called a Leyden jar could
make the dissected legs of frogs twitch as
though they were reanimated. He also saw
such movements during a thunderstorm,
as if the atmospheric electricity found its
way into the limbs. He thought that ani-
mals themselves might generate their own
electricity, and his ideas inspired Mary
Shelley when she described the reanima-
tion of a reconstructed corpse in her book
Frankenstein (1818).
WELLCOME COLLECTION ( (GALVANI, DARWIN); CALTECH

Above: Electrons were the first subatomic par-
ticles—“pieces of atoms”—to be discovered.
Though they were known to carry an electrical
charge, it was extremely challenging to measure
that charge accurately. The American scientist
Robert Millikan did it in experiments conducted
in 1909–’13 in which he observed the movements
of tiny oil droplets in an electric field, assuming
that the smallest difference in their charge corre-
sponded to a difference of just a single electron.
His work showed how important it was for exper-
imenters to develop an instinctive feeling for their
apparatus.
86
Below: Physicist Chien-Shiung Wu’s demon-
stration of the violation of parity (1956), which
showed that nature distinguishes right from left,
shattered one of the long-standing assumptions
of fundamental physics. The experiment, said
her friend Wolfgang Pauli, the Austrian physicist,
Opposite: The 2015 discovery of gravitational
waves by the Laser Interferometer Gravitational-
Wave Observatory (LIGO) in the US verified the
prediction of general relativity: that violent events
involving extremely massive astrophysical bod-
ies, such as the merging of two black holes, pro-
duce ripples in space-time called gravitational
showed that “God is left-handed.” The discovery
of parity violation won the 1957 Nobel Prize for
the two theorists who suggested it—Tsung-Dao
Lee and Chen Ning Yang. Wu should have shared
that award but did not.
SMITHSONIAN INSTITUTION (WU); CALTECH/MIT/LIGO
waves. The experiment used an immense
instrument in which light beams were sent and
GUTTER CREDIT HERE
reflected back along two tubes that were each
four kilometers long. The distortions of space-
time caused by a passing gravitational wave
change the way the light beams interfere when
they return and cross one another.
GUTTER CREDIT HERE
88

Happy anniversary
For 125 years, Technology
Review has aimed to explain
the “always new and ever more
complex problems” we face.

“To a community groaning

under an ever increasing
weight of periodical literature,
a new magazine is forced to
present itself in an attitude
of apology ... The Technology
Review must make plain its
purposes, its capacity, its
determination to be useful,
before it can expect to receive
recognition from a public too
busy to be indulgent. Realizing
this, and mindful, too, of
the spirit and traditions of
the Massachusetts Institute
of Technology, The Review
neither throws itself upon
the charity of its friends nor
prays them to be blind to
its shortcomings. Doubtless
it will need indulgence,
doubtless its attainment will
fall much below its aspiration;
but if it does not so far
succeed in its attempt as to
gain support through feelings
other than those of simple
friendliness, the existence
of The Review cannot be too
quickly ended.”
—Vol. 1, No. 1, January 1899

MIT Technology Review (ISSN 1099-274X), January/February 2024 issue, Reg. US Patent Office, is published bimonthly by MIT Technology Review, 196 Broadway, 3rd floor, Cambridge, MA 02139. Entire contents ©2024. The
editors seek diverse views, and authors’ opinions do not represent the official policies of their institutions or those of MIT. Periodicals postage paid at Boston, MA, and additional mailing offices. Postmaster: Send address changes
to MIT Technology Review, Subscriber Services, MIT Technology Review, PO Box 1518, Lincolnshire, IL. 60069, or via the internet at www.technologyreview.com/customerservice. Basic subscription rates: $120 per year within the
United States; in all other countries, US$140. Publication Mail Agreement Number 40621028. Send undeliverable Canadian copies to PO Box 1051, Fort Erie, ON L2A 6C7. Printed in USA. Audited by the Alliance for Audited Media.
Be a part of the
biggest conversation
happening now.

Cut through the AI hype with our

weekly newsletter, The Algorithm.

Get access to exclusive AI insights

and in-depth stories delivered
right to your inbox.

Scan this code to sign up

for free or learn more at
technologyreview.com/thealgorithm
      
          
       
            
            
Velocity.BoozAllen.com