www.weneedtotalk.

ai
1

We Need to Talk, AI
A Comic Essay on Artificial
Intelligence

Dr. Julia Schneider

Lena Kadriye Ziyal
We Need to Talk, AI

Dr. Julia Schneider

Lena Kadriye Ziyal
Bibliografische Information der Deutschen Nationalbibliothek

Die Deutsche Nationalbibliothek verzeichnet diese Publikation in der

Deutschen Nationalbibliografie; detaillierte bibliografische Daten
sind im Internet über http://dnb.dnb.de erhältlich.

ISBN Softcover: 978-3-748531-28-9
© 2019

Öffentlichkeitsarbeit und Beratung: Eric Eitel

Lektorat und Korrektorat: Katharina Kopp, Catalina Schneider
Illustrationen und Layout: Lena Kadriye Ziyal
Texte: Dr. Julia Schneider 

Verlag: Dr. Julia Schneider 

Nogatstr. 31, 12051 Berlin
hello@weneedtotalk.ai
Website: weneedtotalk.ai

Druck (on demand): epubli – ein Service der neopubli GmbH, Berlin
Druck (limited edition): Online-druck.biz

Online Free Read auf www.weneedtotalk.ai:

Attribution-NonCommercial-ShareAlike 4.0 International
CC BY-NC-SA 4.0
This comic essay would not have
been possible without the support
of many people and places. We thank
Eric, Knud, Katharina, Catalina,
Sven, Oktay & Café Roasters with
the Kaffeekränzchen Adem, Cem,
Jule, Kathi, Luisa, Paul and Paula,
the restaurant in Karstadt Hermann-
platz, infotext and INWT for inspi-
rations and exchange, Iris, Nele,
Jonas, Maren, Patrick and Wolf as
well as Amerika-Gedenkbibliothek.
Many thanks also to all friends and
family members who make our lives
worth living, to what makes us
humans creative, compassionate and
cooperative - and to the AI-appli-
cations Google Pictures and DeepL,
without which this comic would defi-
nitely be a different one. Which is
kind of funny.
 Examples
20-27

Basics
11-19

Intro
6-10

www.weneedtotalk.ai
Outlook
44-49

Social Utopias
50-53

Risks
37-43

Chances
28-36

www.weneedtotalk.ai
//Knowing the future is impossible. But
what we can do: Knowing what future we
would like to have. And then work on it.

www.weneedtotalk.ai
 7

Our creatures have always turned Neither magic nor natural sciences
against us. This applies to the Golem have ever helped to make artificial
of Jewish mythology as well as to humans submissive and controllable.
Frankenstein.

Parents whose kids prefer to wear rubber

An AI Take-Over is a hypothetical
boots instead of weather-proof sandals
scenario in which AI takes control of
even when it‘s hot outside know that.
the planet – taking it away from us.

“Unfortunately, it might also be the

In 2011, Stephen Hawking said that
last, unless we learn how to avoid the
“Success in creating AI would be the
risks.” How can we make AI systems both
biggest event in history.”
safe and beneficial?

www.weneedtotalk.ai
 Who is Julia
8

I love making sense of our world. I

Good question. That's something I love to discover patterns in data,
ask myself, now and then. with code.

16335578682576215675654122353
404
57185321254565975551251656516
51585152699995444511238844598
75326658996544475553215872465
29654231546973211225321232589
66547895645231597462126654489

Our brains are prone to some errors.

I love data`s friendliness, their lack
Overconfidence. And discrimination
of strategic answering. If they are
against people who seem to be diffe-
biased it`s not their fault but ours.
rent from us.

Artificial Intelligence (AI) might have For that reason, we need to demystify
the potential to help us find answers AI. Therefore, I'm writing this comic.
to problems we are not able to solve. Let your voice be heard.

www.weneedtotalk.ai
 Who is Lena
9

I love visual language. I love to

encrypt thoughts into pictures by
Not one thing in particular, I guess. exploiting society`s visual memory.

Visuality is one of the most vivid It’s fun to mess around with visual
encryption systems I know. We are codes by recomposing associations.
constantly increasing and renewing it Pictures start to shift meanings and
collectively. interfere in society`s processes.

Looking from different angles might

AI is a field that gives us an intimate help us to overcome commonly known
view on society`s desires and fears. perspectives. Let your view be seen!

www.weneedtotalk.ai
 Lessons from AI in Fiction
10
Hi, I‘m R2D2
1927 1977 and I am from
Star Wars. I can
solve multiple
problems.
Hi, I‘m
Maschinen-Maria
and I am from
Metropolis. My
skin is made of
human skin.

1984 2013
Hi, I‘m Hi, I‘m Samantha
T-800 and I am from and I am from Her. I
Terminator. I kill have emotions and can
people. Also, I act build relationships.
autonomously.

2014

Hi, I‘m Tars and

I am from Inter-
stellar. I am
funny. Do you
want to hear a
joke?

Seems, like we should not seek to

replicate but to augment ourselves.

www.weneedtotalk.ai
Basics

www.weneedtotalk.ai
 What is Artificial Intelligence?
Actually, the methods are not new.
Just the techniques and the data.
If you define intelligence as the
ability to capture the essence of a
new situation...
This being said, AI is still in its
early stages (although it reads diffe-
rently on the news).
13
At the moment, AI is mainly concerned
with performing complicated but
repetitive tasks.
... real AI needs to think more like us.
With intentionality, with emotions, with
conscience; capturing counterfactuals.
Up to now, it`s more adequate to speak of
„artificially augmented intelligence“.*
*
Nevertheless, we will use the term ‚AI‘ in the following
to simplify matters. And please, find some reading
suggestions at the end of the book.
www.weneedtotalk.ai
 Computer Power
14

Cost
2,600,000,000

100,000,000

10,000,000

1,000,000

100,000

10,000

2,300
Computer power 1970 1980 1990 2000 2010

What kind of techniques helped estab- Moore‘s law is the observation that
lish AI? One key indicator is computer the power of computers will double
power. every 18 months.

8,51213
√4,380

The amount of computer power used for The growth has largely been driven by
training AI has gone up to 3,5 month- computer hardware, optimized to train
doubling time. and run AI-models.

… together with advances in data avai-

This explains why AI has been growing lability and algorithms (=the recipes)
faster and faster… behind AI.

www.weneedtotalk.ai
 Algorithms 1
15

A second key indicator are better Before, when intending to automate a

algorithms. Now we can teach machines task using AI, we had 1) to understand
instead of programming them. it completely and 2) to program it.

The more complicated the task, the The breakthrough was in understanding
harder it was to automate. how we learn to perform tasks.

a cat?

a cat?

a dog..?

Nowadays, AI learns through trial and Modern AI applications achieve breath-

error, based on data we feed it. The taking results that we could hardly
more data, the more accurate it becomes. imagine a few years ago.

www.weneedtotalk.ai
16 Algorithms 2

ok,
got it

What do these buzzwords even mean?

ok, got it Machine Learning (ML) helps recognizing
Machine Learning; Deep Learning; patterns based on existing data and
Neural Networks (NN)...? algorithms: an important branch of AI.

learn

logic

search

sort

store

Before Deep Learning (DL), we had to know

(Note: The AI bookshelf includes other a lot about the data we were programming.
topics, as well.) Otherwise, ML would break down.

35°
xy

if that
makes you
happy...

Like our brain, NNs compare new informa-

DL uses artificial Neural Networks (NN) tion with objects they know. However,
that learn patterns directly from the they have not yet been able to think of
data they are fed. new objects.

www.weneedtotalk.ai
 Data 1 17
ozone value
air pressure
ity solar
id
of ain um air quali io
n
pr
of ain
in
t at oc

h
i
ra
d
ra es
r
r

freq ncy

y
of p si
f

n
ch ce o

l o
e i ng

ue

ol
t
ch ce
ce

hai
educ v

lution

martial status
le
a
an
an
an

CO 2 crime

rcolour
ch

vo
ation rate

living cost
e n t r l
conc u m

e
air

number of
buildings
h t

skin
tone
ig

rate
pressure
es

median
income
wifi ey
od

po
status
mo

st
h
toot population

te t
ure
blue atus re d h e
ra ar
su oo nam
s t e traffic status
step noise

date
name
level level
radius
movement

number age
pu

colour
es bl

time zone
destination size
ls

gender brightness
e

texture
speed distance
pr

route direction weight contrast

soil
density temperature preferences location

Any form of raw fact or figure is data.

Data is the third element that makes AI Whether on paper or in electronic form.
so powerful. But what exactly is data?

Data describes what we know and we can The internet and mobile devices like
draw conclusions from it. Data can take smartphones, drones or simple sensors
the form of text, numbers, images, or have made data abundant and far more
sounds. valuable.

There‘s a
99% chance
that you
are a cat!

The more data an algorithm is trained Better results = higher usability = more
on, the better its results: conclusi- users = more power. Therefore “data is
ons, predictions, timing, actions. the new oil”.

www.weneedtotalk.ai
18 Data 2

Only with the invention of the inter-

Okay, all AI applications need trai- net, we were able to collect and store
ning data to get good results. enough data to make AI work.

There is an adaption of Maslow‘s Pyra-

Too much data for us to process. Just mid of Needs. Aimed at achieving AI
the right amount for AI. solutions.

collect

store

transform

explore

AI

AI is a journey that begins with data:

Four out of five necessary steps have collecting, storing, transforming and
to do with data. exploring it.

www.weneedtotalk.ai
 General AI 19

But from 1973 onwards, disappointment

In the 1960s, AI pioneers hoped that and criticism in the community, followed
machines could soon learn to think by pessimism in the press, led to the
without human intervention. 1st “AI Winter”. A 2nd one followed.

If you like
Friends, I bet
you‘ll like New
Girl, too

Until the 2000s, AI was a dirty word

that “simply didn’t work”. But when Despite their impressive progress and
data, hardware and infrastructure were success, today’s AI is narrow. Its
ready in 2010, a wide range of AI tasks are often classification and need
applications followed. a lot of data and a lot of energy.

I am aware of myself, I
think, feel, desire.
I love New girl for instan-
ce, even though the plot
repeats itself after the
third season... Also, I
don‘t exist.

No AI can represent causal relation- We really need to adjust our expecta-

ships or integrate abstract knowledge, tions and stop over-hyping AI. If we
e.g., what objects are, what they are don’t, we may find ourselves in another
for, and how they are typically used. AI Winter.

www.weneedtotalk.ai
 Examples

www.weneedtotalk.ai
 Maps km/h 21

hey there
18:05
hey there
18:05

a b

b- a =
c

… the AI-application Maps can analyze

Using location data from smartphones... how fast traffic is moving...

...

...

...

...
...

Additionally, maps incorporate

… in real-time. user-reported traffic incidents.

The proprietary algorithms of Maps This means, Maps can suggest the best
have access to vast amounts of data. and fastest routes for us.

www.weneedtotalk.ai
22 Recommendation

All major internet companies – or If we‘re looking for books or shoes,

„platforms“ – refine their products internet shops use AI so that our
and processes using AI. purchase decision gets easier.

sleep

downton
abbey

Likewise, streaming providers use AI

We benefit from improved suggestions
to analyze our consumer habits down to
and rediscoveries of old pearls which
the last detail.
we have missed so far.
strong female

happy
end romance mystery
lead

sex

coming
of age
ity
divers

middle
family
values

80s age

Also, the platforms use the knowledge

of our habits to create and offer new …or to let AI autonomously stir those
products similar to previous succes- components together that have worked
ses… great elsewhere.

www.weneedtotalk.ai
 Social Networks 23
What a wonderful
match...

Those of us who use social networks Social networks use AI to recommend

use AI frequently, without necessa- friends, news, photos, tags, advertise-
rily knowing it. ments.

Now imagine a network that knows even

They even „curate“ the posts of your more than our habits, relations,
friends according to general or perso- thoughts, interests, likes and dislikes,
nal interests – with the help of AI. our hobbies and whereabouts.

In China, this already exists in the form

of „WeChat“. Just recently, the Chinese
Our state of health or bank account government officially announced to super-
for example. vise each group with 12+ members.

www.weneedtotalk.ai
24 Self-driving Cars

Today, we can already activate the

autopilot function which automati- But ultimately, we are responsible.
cally and comfortably drives our car The systems are not autonomous – or
or flies our plane. self-driving.

We would have plenty of room to sleep,

work, read or play, sober or drunk
Within a few years, this might change.
after a party.

Or maybe not. Nearly every car accident

involves some sort of unforeseen circum- Moreover, user tests show that we need
stances, the AI must confront for the autonomous systems to explain their
first time. decision to feel safe. Not easy.

www.weneedtotalk.ai
 Identifying Images 25

In 2015, computers made fewer mista-

kes than we do sorting images into
A lot of AI work deals with identifying
predefined categories.
patterns. At an enormous scale.

You have to act

now, it‘s cancer!

Since 2018, everybody can use a free

tool helping AI beginners detect While early image identification could
images. be lifesaving...

target B

target A

target C
I
totally
see it
...
DL/NN Pablo Picasso
rifle, 2017 Portrait of a
woman, 1910

Targets detected. ready

to fire in three, two...
...sometimes it is life-threatening.
The military is a huge fan of it, too.
Unfortunately, AI is susceptible to
hallucinations: caused by biased data or
by hackers who, for example, make AI
mistake guns for apes.

www.weneedtotalk.ai
26 Emotional AI

To really understand us, you need our

Only ~ 10% of the emotional meaning of facial expressions, our tone of voice,
a message is conveyed through words. our gestures plus our words.

you said you didn‘t

want to go there
and instead sleep
all day...

... we need our AI devices (like

assistant systems, cars, or robots) to
As our relationship with AI is becoming
sense and adapt to what we mean and not
more and more intimate...
only to our words.

Indeed, Emotional AI could help us with

some highly valuable tasks. Help auti- Nonetheless, our freedom of thought might
stic children, for example, to learn be at stake in hostile scenarios.
emotions.

www.weneedtotalk.ai
Exoskeletons 27

Exoskeletons are an external framework

we can wear to augment our natural
physical ability and reduce strain and But every one of us is different.
weaknesses.

We need assistance that is specifically

AI can use real-time measurements of our
fit for us. We need to tailor our
body signals, our breathing rate or hip
devices at an individual level.
extension for example.

This way, the AI can get our individual Our bodies are different and constantly
profile right. AI can make our exos- changing. Nowadays, the only feasible
keletons fit our needs – elderly man or way of translating this to robotics is
firefighter. through the help of AI.

www.weneedtotalk.ai
 Chances

www.weneedtotalk.ai
 Dealing with Big Data 29

Our power to analyze the tons of data AI helps us deal with large and
we are now able to store and to complex datasets in ways we have
collect can be a bottleneck. never seen before.

Even if we could analyze data, but it

Even with hundreds of us analyzing proves to be exhausting and tedious, we
patterns in big data, the sheer volume can use rule-based AI systems – expert
simply overwhelms our capacities. systems – that do the job for us.

Traditional algorithms are bound to Now enter AI. Because AI systems get
follow the same logic over and over, smarter as more data is given to them,
without flexibility or learning new they are well suited for patterns and
paths with new data. anomalies over the time.

www.weneedtotalk.ai
30 Efficiency

In many situations, AI is simply A legal platform carried out a compe-

faster and more accurate than we are. tition between 20 experienced lawyers
Let’s look at an example*. against a trained AI.

Competitors had to review 5 legal They were rated according to how

contracts in 4 hours and identify 30 accurately they identified each issue.
legal issues. Who won?

Humans achieved, on average, an 85% The human lawyers who competed against
accuracy rate, the AI 95%. Humans took the AI in the experiment said, the tasks
92 minutes to complete the task, the were very similar to what lawyers do
AI… 26 seconds. every day.

*
https://legal-revolution.com/de/the-legal-revolutionary/ik/artificial-intelligence-vs-human-in-the-legal-profession

www.weneedtotalk.ai
 Cognitive Biases 31

Our brain, as a result of evolution, Instead, we strive for a competitive

was not optimized for rational decisi- degree of „fitness“ in our specific
on-making or perfect diagnoses. environments; quite fast and often
satisfying.

Our brain power is biologically limi-

ted; therefore, we work with heuri- We call this cognitive bias. There
stics, trust our gut feeling or store are many of them and we share them to
& retrieve wrong evidence. some degree.

And this is what makes our distortions

have large-scale consequences. Were Used wisely, AI could indeed help us
they random, they would cancel out. improve our diagnoses and decisions.

www.weneedtotalk.ai
32 Creativity

Special thanks
to:

My dear AI.
Without you I
never would‘ve
found the time
to complete
this...

When an AI sorts our photos, it is

AI can save time otherwise wasted on
easier to make a photo-calendar with
stupid, tedious work for creativity.
our children (in 28 positions) for
But it can do even more.
our parents as a present.

Don‘t ask me,

I‘m just a
generic style
transfer.

We can also get creative directly with

When an AI puts together a playlist of the AI, compiling data and coding it,
up-and-coming producers of house music, in a narrow area. But the AI has no
it may inspire our next DJ-set. judgement.

What do you
think?

So why are
you always
getting the
credit?

Judging is an ability we possess. And The smaller the role of humans in the
we love doing it. This is where AI process of art-creation becomes, the
together with us can achieve great more AI questions traditional concepts
things, with a human tweak. of art.

www.weneedtotalk.ai
 Connection 33

Today, they often act like overpaid

Most doctors became doctors because bookkeepers: taking in and spitting
they wanted to connect with people in back data, prescribing drugs, adjus-
order to heal them. ting doses, ordering tests.

But we cannot connect to doctors who

And it’s still only retrospective, struggle to keep up with messy data
error-prone knowledge. from disparate connections. And they
cannot connect to us.

Our doctors could get to know us

better, having the time to learn how a
AI can support proactive, individualized disease uniquely affects us and educate
strategies. us in the right way.

www.weneedtotalk.ai
34 New Opportunities

Robots solve complex tasks today –

AI allows for fascinating new oppor- but these tasks are deterministic. At
tunities. Teaching robots without a any point of time, you know in advan-
single line of coding, for example. ce where your robot needs to be.

Every time a task or location changes, Even for experienced programmers, it is

the robot`s software must be re-pro- hard to come up with a set of verbal
grammed. This is time-consuming, expen- rules of how to move when movements and
sive and difficult. settings are tricky or unpredictable.

AI now enables us to teach robots how From the recorded example-movements,

to move, either together with the automation scripts can be created and
robot, or, with jackets and gloves optimized – without us having to write
containing sensors. complex code. Fantastic!

www.weneedtotalk.ai
 Inclusion 35

a wall missing
several bricks.

AI reads to us, it narrates the world

AI is also about helping us, about around us and help us “see”. Spe-
improving inclusion and our well-being, ech-to-text AI helps us “hear”, too.
by removing barriers.

Hey mate!

AI helps us type (with words) or talks

to us. Users of all ages are adopting Smart home devices help us do things
virtual assistants because it is so easily around the house and live inde-
simple to learn how to handle them. pendently.

In the classroom, AI can identify indi- AI virtual nurses or therapists offer

vidual challenges and offer more perso- 24-hour-support, or someone to talk to
nalized approaches. in case we are shy.

www.weneedtotalk.ai
36 Comfort

Almost all answers to our consumer,

communication or information needs are
AI satisfies and reinforces our desire just a mouse click or an app away. And
for comfort. most of the time, AI is involved.

It is tempting not to have to leave our ‘Pre-AI’-institutions such as shops,

couch at all, no matter what we want to local transport, restaurants, libraries,
do: eat, work, chat, watch a movie, or doctors must react to our new conve-
listen to music... nience.

mall_blueprint.pdf

All institutions with predominantly It is precisely these institutions that

analogue offerings are undergoing a are social places where we can come
tremendous process of change. But it together and exchange ideas. Where we
might be worth it. can decelerate. “Waste” time.

www.weneedtotalk.ai
Risks

www.weneedtotalk.ai
 Surveillance 39

39,12
55°

1,46
log 37

AI makes monitoring people much
easier. Not only via our words or
facial expressions, via our consumer
behavior or interests.
sin 0,2
1,58
But also, via our movement patterns.
Each one of us moves differently.
Our movement patterns are unique.

Person A

Person B

Person C

Weak radio waves can penetrate walls, The AI can translate our movement pat-
but our bodies reflect them. Using terns to little stick figures. Once
this data, an AI can be trained to registered, the AI can even identify us
identify our movements. behind a wall.

That means the AI does not only know

where we are. The AI can also distin- It really comes close to magic. Hello,
guish what we are doing, whether we Harry Potter! And hello, surveillance
are standing, or eating. state.

www.weneedtotalk.ai
40 Opinion-Forming and Media

Borders between private and public

AI influences public opinion-forming communication are blurry. User-genera-
and the media. Our public sphere has ted content is rich but polarized and
become fragmented. often privatized on closed platforms.

Right-wing activists, for example, use

New players and sources make special- AI bots, too, to spread hate and fake
interest-content popular and visible news in social media and, ultimately,
and, thus, important for mass media. mass media.

TM

Social media-users receive AI-persona-

Could we, in turn, use AI to enable a
lized content of political organiza-
better opinion-forming – and identify
tions, based on users‘ personality
those who benefit from current AI tools?*
profiles.

Credits to Joseph Weizenbaum

*

www.weneedtotalk.ai
 Future Work 41

The share of jobs at high risk of This involves all kinds of jobs on the
replacement by computers or robots has social ladder. White-collar jobs like
increased significantly.* lawyers, medics, or brokers…

… as well as creative jobs like

artists, chefs or journalists, or AI can outperform us in doing repetitive
blue-collar jobs such as construction tasks. Cheap, strong, fast, easy.
workers or drivers.

28
1 46
livehack#

We are better in understanding

non-trivial, new situations, building We are also needed as partners of AI:
relationships, creating context and developing algorithms that work, supervi-
meaning. Custom-fit solutions. sing, repairing. Combining us with AI.

*
https://doi.org/10.1787/2e2f4eea-en
www.weneedtotalk.ai
42 Data Security and Safety*

Our blind trust in AI providers is mis-

placed. We all know that the amount of Since AI needs tons of data, we need to
data that we are implicitly or explicit- improve our data security and safety. A
ly forced to leave behind is too great. broad campaign, starting with, e.g.:

inbox 13

draft
sent

junk 5
data letters 14.0285

The “data letter”, an obligation for Penalties for data outflows for system
all companies and authorities to tell operators, and an obligation for connec-
us what data they store about us and ted devices to declare the frequency of
how to view, correct and delete it;** the security updates (incentives);

learn
coding

Permanent government investment in And: a widely available education in safe

secure IT, e.g., open source components programming, along with educational,
written and updated in secure program- advisory and complaint services for the
ming languages; safe use of our digital systems.

Credits to Chaos Computer Club (CCC) 

* We think that the European General Data Protection
**

Regulation is a very good start here.

www.weneedtotalk.ai
 Energy Use 43

analyzing...

But conventional hardware was not

In fact, AI has huge potential to opti-
designed for AI, hence it needs much
mize energy use, by analyzing complex
more computing power, much more
data, making forecasts, optimizing
energy, than our brains do.
processes via brain-like algorithms.

An AI supported translation service, Today, data centers use more energy

e.g., DeepL, can handle 5,1 quadrillion than some countries, e.g., Iran.* In
operations per seconds. But that means 2018, we had 33 Zettabyte digital data
masses of computing power and data. per year. By 2025, we‘ll have 175.

Save
A+++
A++
Save energy
A+
Save the planet A
Save my business model B
C
Save the children
D

One of the most worrying forecasts says In short, our brains are far more energy
data centers will use 8% of global efficient than our best computers – and
energy in 2030. The big internet firms we still do not really understand how.
bet on renewable energies.
e.g., https://www.nature.com/articles/d41586-018-06610-y
*

or https://www.datanami.com/2018/11/27/global-datasphe-
re-to-hit-175-zettabytes-by-2025-idc-says/

www.weneedtotalk.ai
 Outlook

www.weneedtotalk.ai
Overall Human-Machine Intelligence 45

We mainly interact with AI via our On the smartphone, we use AI basically

smartphone. A proprietary marketplace for entertainment (image recognition),
for countless uncurated applications communication (language and translati-
(=platform economy). on) and information (e.g. google).

Up to date, there is no smartphone that

Unbeknownst to us, AI improves perfor-
is especially good at helping us get
mance and battery life, too. And: It
better ideas or interpret information.
analyzes us, its users.
There is no interface for that either.

The development budgets are driven by

But couldn‘t we also develop a smart-
numbers of purchase. In turn, our
phone that makes us more creative, more
purchase decision is based on entertain-
intelligent? That understands like us?
ment and communication values, and the
Really augment us?
design of the device.

www.weneedtotalk.ai
46 Feminism

748%462.3729
28.197-46382*
/029.11
2.3916+90
61.93601*
39272893*4839
02/290383628-1743.
37483396*3637482.9
2.1937-57392+7493.

937
38/82
7+856
5.389
The contribution of women to AI is usually They were the first computers,
underestimated, especially concerning the actually: running calculations in
democratic applicability of computing. the first information network of the
world.

What we fear is not an AI-takeover. We In an AI world, where „female“ service

fear scientists’ homogeneity leading robots are „hot“ and submissive, how
them to (finally!) „create“ intelligent can children learn to treat women with
beings – in an egocentric way. respect, with dignity?

We need (different!) women to participate

And many of those whose service jobs
in the development and use of AI. Other-
will be replaced belong to an already
wise we'll just reproduce patriarchal
vulnerable group: minority women.
structures.

www.weneedtotalk.ai
 Digital Capitalism* 47

The idea of objectiveness, general

Algorithms have always been an integ- abstract rules, quantification, de-per-
ral part of our modern capitalist sonalization and the concept of equality
societies. are characteristics of modernity.

Max Weber, e.g., pictures the judge as Marx' workers are an appendage of the
an automation who processes concrete machine, alienated from the spiritual
cases (data) based on an algorithm, potencies of the capitalist working
namely the law. process.

Since today's AI is largely incomprehen-

Digital capitalism is the ground on sible to us, we need to regulate and
which AI thrives. The central paradigm shape AI as a public service and snatch
is – instead of production and sale – the omnipotence over AI from the few
the collection and use of data. dominant, digital companies.

*
Credits to Timo Daum
www.weneedtotalk.ai
 ART: Accountability,
48
Responsibility, Transparency*

This requires the participation and

Throughout the world, we share a gro- commitment from many of us. It means
wing awareness: We need to make AI training, regulation and awareness in
safe, beneficial and fair for us. 3 equal pillars:

Accountability is the concept that AI This includes our responsibility to

should be held responsible for the develop a framework for AI that incor-
results of its algorithms. porates our values.

Transparency refers to the need to

describe, inspect and reproduce the AI But let's not fool ourselves. For this
algorithms and results, and to manage ART of AI, we need a new and more ambi-
the data used, in a fair way. tious form of governance.

Credits to Virginia Dignum

*

www.weneedtotalk.ai
 No Fear of Numbers 49

In order to understand how AI could Healing the public shaming in our

improve our life, we propose a real individual math-biographies that
revolution: caused fear.

Not only does this deprive life. Zero,

Many of our brains signal “pain” when e.g., is a fascinating tool allowing us
we read numbers, think or talk about to write “1908” instead of MCMVIII:
them. 1,000+(1,000-100)+5+1+1+1.

To be or not
to be...

Zero is also part of the binary code,

fundamental to todays' computers and With improved “math-esteem”, more of us
their language. Fundamental to AI, could and should discuss chances and
therefore. risks of AI in our lives.

www.weneedtotalk.ai
www.weneedtotalk.ai

Social Utopias with AI
All in all, AI today is not Sky-
net, HAL9000, Wall-E or C3PO.
We appreciate this technology,
but we don‘t want to overrate
it. For over 300,000 years we‘ve
been hunting with spears; for
less than ten years we‘ve been
looking for our next song with
the help of AI.
Surely, there have been revolu-
tionary applications, especial-
ly in the field of machine lear-
ning, in past years. This has
been helping us in many ways.
However, it is always very spe-
cific, single tasks, the so-cal-
led “narrow” AI is helping us
with. “General” AI would combine
multiple intelligent functions
and improve itself on its own.
„Think“. „Want“.
Today’s narrow AI must still
solve many problems in the area
of methodology, technology, and
51
resource consumption. But even
with the use of this narrow AI,
we keep wondering what a utopian
AI could look like. What exactly
would we want to use it for?
Technical progress has often
promised to make the world a
fairer place. For many of us,
however, this promise has not
yet been kept. Technical soluti-
ons cannot completely overcome
the existing social injustice,
while everything else remains as
it is. If we look at it from a
global perspective, injustices
are a fundamental part of the
societies we know.
But we do believe that techni-
cal solutions, including AI, can
somehow function as a catalyst
that initiates and influences
changes within ourselves and
within societies. An AI may not
be able to resolve all injusti-
52
ces, but it can raise questions
about how we want to live. Since
we were little, we, and perhaps
you, too, have dreamed of living
in a better, fairer and friend-
lier world in which each of us
has from the moment we are born
approximately the same privile-
ges and opportunities for well-
being and fulfillment. (This, by
the way, is what most of us would
choose before we know whether we
are born privileged. In retro-
spect, the privileged among us
are often in favour the inheri-
tance of privileges.)
AI could increase the chances
and resources of the underprivi-
leged among us. What could this
look like? Some have already
begun to gain initial experience
with AI-supported application
procedures, but they have not
yet been perfectly successful.
Whether AI can be of help here
depends on the type of trai-
ning data: If we train AIs on
discriminatory data, they will
make discriminatory decisions
and reduce, not increase, the
opportunities, and resources of
the underprivileged.
Also, the protection of this
sensitive data, human monito-
ring to avoid technical or human
error, the possibility of gai-
ning insight into decisions and
the ability of intervention are
some of today’s challenges.
But let us think one step fur-
ther. How can AI support us to
articulate our needs and gather
them at negotiating tables of
society? Do we need to invent an
intelligent unit of measurement
– one that is not money – to
quantify needs, costs, benefits?
Can a digital platform be the
right place to collect and eva-
luate them?
If, for example, a motorway was
to be planned through a city,
an AI could transparently docu-
ment every single inhabitants’
involvement and propose diffe-
rent solutions that work best
for the city. Now think global-
ly: imagine a world in which
the social costs and benefits
of a clean environment, a spe-
cific resolution of a conflict
or improvement in health care
are transparently documented
and processed into solutions,
which bring the greatest bene-
fit to all of us. Somehow tri-
vial, but given that maybe you,
dear readers, and certainly we
as authors belong to those who
profit from the distribution of
privilege, this could be a major
step to overcome our privile-
ge-blindness, and a way to learn
socially responsible behavior.
Could AI furthermore be used
to transparently document and
reward the invisible work, that
we do at home and the unpaid
volunteerism or care work in
non-commercialized places? This
might be a real step towards
gender equality. Also, imagine
an AI that plays with set pieces
of all constitutions to simula-
te and optimize the respective
social effects. AI can give us
a glimpse into the future when
certain changes in the law are
pending. Or it could even create
fluid constitutions that – depen-
ding on current social processes
– customize themselves.
For the long-term prosperity
of societies, it is essential
to have an independent media
carried by free-thinking jour-
nalists who are impartial to
economic and party-political
pressure. Couldn‘t AI also be
of help here? As users of the
media, we could allow the use of
our private data for socially
meaningful research projects,
for example for improving our
transport systems in cities. In
return, the media would receive
the resources needed for free.
This might guarantee truly inde-
pendent reporting instead of the
media getting their resources
from companies that use their
credibility on the one hand whi-
le on the other they undermine
it by selling their marketing
texts as editorial titles.
Moreover, AI could raise ques-
tions that confuse our self-
image. Not necessarily by the
actual tasks, but rather by the
new existence of a recognizable
intelligence next to the human
intelligence. Just a thought.
We haven‘t yet talked about the
potential of a wave of robotics
and real automation, the time,
and quality of life that this
could bring us, the many strenu-
ous activities that we wouldn‘t
have to do anymore; think of the
many underpaid people who are
deleting traumatizing decapi-
tation videos in social media
channels…
of AI. After some good ideas
failed monstrously due to lack
of suitable tools for flexib-
le planning, maybe we are for
the first time in history able
to establish an economy that is
not based on the exploitation of
humans and nature.
Or is this vision making us
create a creepy surveillance
state? Why should the authori-
ties always be well-meaning? How
could we make sure they are?
We also didn‘t talk about the
cyborgs, nor of a billion things
that we couldn`t remotely think
of. And even though a utopia
has yet to be shaped, we are in
desperate need of it. This will
set the direction in which we
are heading. But the first step
has been made: What can AI help
us with? Where do we have to be
careful? What do you think?
Let your voice be heard and your
point of view be seen.
Tbc.

We have also not yet spoken of

the possibilities of effective-
ly planning and optimizing an
economic system with the help

www.weneedtotalk.ai
Further reading

> Ajay Agrawal, Joshua Gams, > Nick Bostrom: Superintel-

Avi Goldfarb: Prediction
Machines: The Simple Eco-
nomics of Artificial Intel-
ligence. Harvard Business
Review Press, Boston, Mas-
sachusetts, 2018
> Claire l. Evans: Broad Band
– the untold story of the
women who made the inter-
net, Penguin Random House,
2018
> François Chollet, J. J.
Allaire: Deep Learning with
R. Manning Publications,
2018  wir: Zur Kritik der digita-
> Jean-Noël Lafargue, Mathieu
Burniat: Das Internet. Ver-
lagshaus Jacoby & Stuart,
Berlin, 2018
ligence: Paths, Dangers,
Strategies. Oxford Univer-
sity Press, 2014
> Shoshana Zuboff: The Age
of Surveillance Capitalism:
The Fight for a Human Futu-
re at the New Frontier of
Power: Profile Books, 2018
> Stuart Russell, Peter Nor-
vig: Artificial Intelligen-
ce: A Modern Approach. 3.
Auflage. Prentice Hall, 2010
> Timo Daum: Das Kapital sind
len Ökonomie. Hamburg: Edi-
tion Nautilus GmbH, 2017

> Kai-Fu Lee: AI superpowers:

China, Silicon Valley, and
the new world order. Hought-
on Mifflin Harcourt, Bos-
ton, 2018
56
Dr. Julia Schneider is an
independent consultant for Artificial
Intelligence and a member of
the scientific committee of VDEI
Association of the Exoskeleton
Industry e.V.

She received her doctorate in

economics from the Free University
Berlin for her research on the
effects of the 2005 German
labor market reforms on welfare
recipients` behavior and health.
After that, she worked as a
postdoctoral researcher in the
field of empirical labor market and
innovation research and as senior
data strategist.

Lena Kadriye Ziyal is part of

the collectively run content and
graphic design agency Infotext
in Berlin. She creates design
concepts, infographics, icons and
illustrations.

Lena studied visual communication

and graphic arts at Kunsthochschule
Berlin-Weißensee, the University
of Arts (UdK) in Berlin and at the
Marmara University Istanbul.

Before joining Infotext she worked

as a freelance graphic designer and
visual artist.
 www.weneedtotalk.ai
58

In 30 years, will robots do all the unpleasant work for us?

Or will they subjugate us to become submissive slaves? The
debates on how Artificial Intelligence (AI) will change our
lives move between these extremes. There is no doubt that the
change will be dramatic. Maybe now is just the right time to
start interfering.
This pioneering comic essay on AI invites you on an
illustrated journey through the dimensions and implications
of the groundbreaking technology. Discussing important
chances and risks associated with AI, this work is a creative
stimulus for insiders of the subject as well as an invitation
for newbies to get informed and join the debate.

With a doctorate in economics, Julia Schneider appreciates

data and code as tools for solving complex puzzles – and
loves comics as a medium for telling complex stories. Coming
from the opposite direction, artist Lena Kadriye Ziyal loves
encrypting complexity with associations and thereby expands
the meaning of a theme with her perspective.