What computers still can't do
by Hubert L. Dreyfus
Dreyfus's devastating 1992 critique argues computers lack embodied human intelligence needed for true understanding.
"What makes some things programmable is not itself programmable and moreover it cannot be specified, described, or even named".
Editorial Summary
When it was first published in 1972 and revised in 1992, Berkeley philosopher Hubert L. Dreyfus's manifesto argued that disembodied machines cannot mimic higher mental functions, causing an uproar in the artificial intelligence community. Dreyfus predicted the failure of symbolic artificial intelligence projects like general problem solvers and automatic translation machines, arguing that their conception of mental functioning was naive. Drawing on phenomenological philosophy, particularly Heidegger's work, he contended that human intelligence requires embodied, contextual understanding that cannot be reduced to rule-based computation. The revised edition added analysis of connectionism and neural networks, maintaining his core argument that computers cannot achieve genuine understanding or meaning.
Perspective
"Reading this feels like watching a philosopher systematically dismantle decades of artificial intelligence hubris, forcing you to confront the profound difference between computational processing and genuine understanding. Dreyfus's distinctive contribution is his phenomenological argument that human intelligence is fundamentally embodied and contextual, requiring a kind of situated, holistic engagement with the world that symbolic computation cannot replicate. AI researchers grappling with the limitations of current systems will find his prescient analysis of why rule-based approaches fail, along with his early skepticism about whether even neural networks can bridge the gap to true machine intelligence."
Matched by concept and theme
