What Computers Still Can't Do

Hubert L. Dreyfus applies the phenomenological traditions of Heidegger and Merleau-Ponty to challenge the fundamental assumptions of symbolic artificial intelligence. He argues that human intelligence is not a matter of manipulating internal physical symbols according to formal rules, but is instead rooted in biology and social context. The book identifies four specific hurdles—biological, psychological, epistemological, and ontological—proposing that because machines lack bodies and a world of shared practices, they cannot achieve true common sense or contextual understanding. This 1992 edition updates these critiques to address the emergence of neural networks and connectionism.

Computer scientists and philosophy students read this book to understand the limitations of computationalism and the representational theory of mind. It provides a framework for recognizing why software often fails at tasks involving ambiguity or physical intuition. Readers gain a perspective on why human expertise relies on unconscious skill acquisition rather than conscious rule-following. By finishing this text, the reader understands the difference between calculating data and the situated experience of human intelligence.

Deep and outward-looking; a demanding read.

Best for readers like The Deep Diver. Which reader are you?