The Fundamental Limits of Machine Learning

A few months ago, my aunt sent her colleagues an email with the subject, “Math Problem! What is the answer?” It contained a deceptively simple puzzle:

She thought her solution was obvious. Her colleagues, though, were sure their solution was correct—and the two didn’t match. Was the problem with one of their answers, or with the puzzle itself?

My aunt and her colleagues had stumbled across a fundamental problem in machine learning, the study of computers that learn. Almost all of the learning we expect our computers to do—and much of the learning we ourselves do —is about reducing information to underlying patterns, which can then be used to infer the unknown. Her puzzle was no different.

As a human, the challenge is to find any pattern at all. Of course, we have intuitions that limit our guesses. But computers have no such intuitions. From a computer

You're reading a preview, sign up to read more.

More from Nautilus

Nautilus5 min read
The Case for Making Cities Out of Wood
Earlier this year, Dan Doctoroff, the C.E.O. of Sidewalk Labs, Google’s sibling company under Alphabet, answered a question about what his company “actually does” during a Reddit “Ask Me Anything” session, replying, “The short answer is: We want to b
Nautilus3 min readScience
Why Mathematicians Can’t Find the Hay in a Haystack
Reprinted with permission from Quanta Magazine‘s Abstractions blog. The first time I heard a mathematician use the phrase, I was sure he’d misspoken. We were on the phone, talking about the search for shapes with certain properties, and he said, “It’
Nautilus9 min readTech
Why Futurism Has a Cultural Blindspot: We predicted cell phones, but not women in the workplace.
In early 1999, during the halftime of a University of Washington basketball game, a time capsule from 1927 was opened. Among the contents of this portal to the past were some yellowing newspapers, a Mercury dime, a student handbook, and a building pe