No. 403 of 500 No. 77 in Mind, Brain & Philosophy

What Computers Can't Do

A Critique of Artificial Reason

By Hubert L. Dreyfus

Print length
429 pages
Publisher
The MIT Press
Reading level
Advanced
ISBN-13
978-0262540674
Published
October 30, 1992
Written for
Philosophy students, AI researchers and historians interested in the conceptual limits of machine intelligence.

Synopsis

Hubert Dreyfus, a philosopher who began his critique while teaching at MIT, expanded a 1965 RAND Corporation paper comparing AI research to alchemy into this book. First published in 1972, it examines the confident predictions of the field's early years in game playing, language translation, problem solving and pattern recognition, and argues that their repeated shortfalls were not temporary engineering setbacks. The problem, he contends, lay in four philosophical assumptions behind symbolic AI: that the brain processes information in discrete steps, that the mind follows formal rules, that all knowledge can be formalized, and that the world consists of independent facts representable by symbols.

Drawing on Heidegger, Merleau-Ponty and Wittgenstein, Dreyfus argues that human expertise depends on embodiment, situational context and a background of practical know-how that cannot be captured as explicit rules. The AI establishment answered with hostility, yet several of his predictions held as symbolic systems struggled with common sense. The MIT Press edition, retitled What Computers Still Can't Do in 1992, adds a long introduction assessing neural networks and connectionism. Readers gain the clearest early statement of why intelligence may require a body and a world, and a historical lens for judging today's learning-based systems, which sidestep some of his objections while reviving others.

Scorecard

Editorial score

How it scored

Editorial score: 71.1 / 1005 weighted criteria

  • Influence30%

    27.0/30
  • Authority25%

    21.3/25
  • Practical value20%

    8.0/20
  • Currency15%

    9.8/15
  • Readability10%

    5.0/10
  • Editorial score

    71.1/100

How the rank was computed

Rank No. 403 of 500 comes from a composite of 49.9: observable signals 17.8 of 55 points plus editorial review 32.1 of 45 points.

All 11 inputs
InputPoints
Observable signals17.8/55
Reach (Amazon ratings volume)6.2/16.5
Consensus across best-of lists5.5/13.8
Awards and bestseller evidence0.0/8.3
Reader rating, sample-adjusted5.5/8.3
Amazon sales rank0.6/5.5
Recency0.0/2.8
Editorial review32.1/45
Author authority9.6/11.3
Influence12.2/13.5
Practical value3.6/9.0
Readability2.3/4.5
Currency4.4/6.8

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

The landmark philosophical critique of symbolic AI, essential for understanding why rule-based approaches stalled and why intuition resists formalization.

Strengths

  • Foundational argument that shaped AI's philosophy
  • Many core predictions proved prescient
  • 1992 edition addresses neural networks directly

Keep in mind

Its dense continental philosophy and combative tone can frustrate readers, and the examples center on 1960s programs rather than modern machine learning.

Key takeaways

  1. Much human expertise is intuitive know-how rather than explicit rule-following, so it cannot be fully captured in a symbolic program.
  2. Early AI rested on four questionable assumptions about brains, minds, knowledge and the world, each treating intelligence as manipulation of discrete symbols.
  3. Understanding depends on embodiment and context: people grasp relevance through a shared background of practices that formal descriptions cannot exhaust.
  4. Overconfident AI forecasts deserve scrutiny; early claims about translation, chess and general problem solving fell far short of their stated timelines.

Notable facts

  • UC Berkeley's philosophy department notes that What Computers Still Can't Do, the book's revised edition, has been translated into twelve languages. Source (opens in a new tab)
  • The book grew from Alchemy and AI, a 1965 memo Dreyfus wrote at the RAND Corporation; AI researchers including Seymour Papert and Marvin Minsky responded with hostility. Source (opens in a new tab)

About the author

Hubert L. Dreyfus

Hubert L. Dreyfus (1929-2017) was a professor of philosophy at the University of California, Berkeley, from 1968 until 2016, after teaching at MIT from 1960 to 1968. Educated at Harvard, he was a Guggenheim Fellow and a Fellow of the American Academy of Arts and Sciences. His other books include Mind over Machine, with Stuart Dreyfus, and All Things Shining, with Sean Dorrance Kelly.

Questions

What is What Computers Can't Do about?
Hubert Dreyfus argues that early artificial intelligence research was built on mistaken philosophical assumptions, chiefly that intelligence consists of manipulating symbols according to formal rules. Drawing on phenomenology, he contends that human skill depends on bodies, context and tacit know-how that rules cannot capture. He reviewed the overoptimistic predictions of 1950s and 1960s AI and anticipated several of the failures of symbolic, rule-based systems.
Should I read What Computers Can't Do or What Computers Still Can't Do?
Most readers should choose What Computers Still Can't Do, the 1992 MIT Press edition. It contains the original 1972 argument plus a lengthy new introduction in which Dreyfus assesses neural networks and connectionism, the approaches that later grew into modern deep learning. The original edition matters mainly to historians interested in the text exactly as it first confronted the AI community.
Is What Computers Can't Do still relevant in the age of large language models?
Yes, though in a changed way. Modern systems learn from data instead of following hand-written rules, which sidesteps Dreyfus's main target. His deeper questions remain open: whether intelligence without a body or lived context can achieve real understanding, and whether statistical pattern-matching captures common sense. Many current debates about the limits of large language models restate arguments Dreyfus made decades ago.

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.