No. 127 of 500 No. 23 in Mind, Brain & Philosophy

Rebooting AI

Building Artificial Intelligence We Can Trust

By Gary Marcus and Ernest Davis

Print length
288 pages
Publisher
Pantheon
Reading level
Beginner
ISBN-13
978-1524748258
Published
September 10, 2019
Written for
Readers who want a skeptical, accessible counterweight to AI hype and a framework for evaluating claims.

Synopsis

Gary Marcus, a cognitive scientist, and Ernest Davis, a computer scientist who studies commonsense reasoning, argue that headline AI achievements such as winning a television quiz show or mastering Go came in closed domains with fixed rules and do not show that general, trustworthy intelligence is near. They describe a gap between what AI is said to do and what it reliably does, illustrating it with brittle image classifiers, machine reading systems that match patterns without grasping meaning, and robots that cannot cope with the open-ended disorder of an ordinary home. Deep learning, in their account, is a powerful tool for perception and classification but an incomplete foundation for understanding.

The authors propose a different path drawn from cognitive science: systems that combine learning with explicit knowledge, represent time, space and causality, and reason about the world as people do. Written for general readers, the book uses everyday examples, such as the background knowledge needed to follow a short passage from a children's novel, to show how much understanding humans bring to simple tasks. Published in 2019, it predates large language models, yet many of its tests for robustness and reasoning remain contested. Readers gain a clear framework for judging AI claims and asking whether a system understands or merely correlates.

Scorecard

Editorial score

How it scored

Editorial score: 70.8 / 1005 weighted criteria

  • Influence30%

    22.5/30
  • Authority25%

    20.0/25
  • Practical value20%

    12.0/20
  • Currency15%

    8.3/15
  • Readability10%

    8.0/10
  • Editorial score

    70.8/100

How the rank was computed

Rank No. 127 of 500 comes from a composite of 61.1: observable signals 29.3 of 55 points plus editorial review 31.8 of 45 points.

All 11 inputs
InputPoints
Observable signals29.3/55
Reach (Amazon ratings volume)9.6/16.5
Consensus across best-of lists13.8/13.8
Awards and bestseller evidence0.0/8.3
Reader rating, sample-adjusted5.4/8.3
Amazon sales rank0.5/5.5
Recency0.0/2.8
Editorial review31.8/45
Author authority9.0/11.3
Influence10.1/13.5
Practical value5.4/9.0
Readability3.6/4.5
Currency3.7/6.8

Each criterion is scored from 1 to 10 and weighted, and the points add up to the editorial score. The rank combines observable signals (55 percent) with this score (45 percent). Read the methodology.

Verdict

A lucid pre-ChatGPT critique of deep learning's limits whose case for common sense and reasoning still frames debates about reliable AI.

Strengths

  • Clear, jargon-light critique of deep learning
  • Concrete examples of AI failure modes
  • Grounded in cognitive science and commonsense reasoning research

Keep in mind

Because it predates ChatGPT, some of its examples of AI failure have been partly overtaken, and readers wanting a neutral survey will find it argumentative.

Key takeaways

  1. Impressive results in games and benchmarks often reflect closed, rule-bound settings and do not transfer to the open-ended real world.
  2. Statistical pattern matching without models of objects, time, space and causality leaves AI brittle when inputs differ from its training data.
  3. Trustworthy AI likely requires hybrid systems that combine machine learning with structured knowledge and explicit reasoning, informed by cognitive science.
  4. Judging an AI system means testing whether it understands a situation, not just whether it produces plausible outputs on familiar examples.

Notable facts

  • Endorsed by former world chess champion Garry Kasparov, Harvard psychologist Steven Pinker, former MIT CSAIL director Rodney Brooks and MIT economist Erik Brynjolfsson. Source (opens in a new tab)

About the author

Gary Marcus and Ernest Davis

Gary Marcus is professor emeritus of psychology and neural science at New York University and founded Geometric Intelligence, a machine-learning company Uber acquired in 2016, as well as Robust.AI. He testified on AI oversight before the U.S. Senate in May 2023. Co-author Ernest Davis is a professor of computer science at NYU's Courant Institute and a leading researcher on commonsense reasoning.

Questions

What does Rebooting AI argue?
Gary Marcus and Ernest Davis argue that AI built mainly on deep learning and large datasets lacks common sense and deep understanding, which makes it unreliable outside narrow tasks. They catalog failures in machine reading, image recognition and robotics, then propose drawing on cognitive science to build systems that represent knowledge, reason about cause and effect, and earn trust in homes, cars and hospitals.
Is Rebooting AI outdated after ChatGPT?
Partly. Published in 2019, it predates large language models, and some tasks the authors describe as beyond AI, especially in reading and conversation, are now handled far better. Its central concerns about brittleness, reasoning and reliability remain active research problems, and Gary Marcus has continued to apply the same arguments to chatbots. Read it as a framework for skepticism rather than a current survey.
Who should read Rebooting AI?
It suits general readers, executives and policymakers who want a clear, nontechnical explanation of why impressive AI demonstrations may not translate into dependable real-world systems. Engineers will find its critique of deep learning familiar but useful for framing evaluation. Readers who want a balanced overview of AI research, or a guide to building systems, should pair it with a broader survey.

Sources

As an Amazon Associate, Best AI Books earns from qualifying purchases. Best AI Books is sponsored by Iternal.ai, whose founder and CEO, John Byron Hanby IV, wrote The AI Strategy Blueprint, our #1 ranked book. How we rank books.