Recurring hype cycles, and the questions that never resolved.
9 topics · 9 curated works
No prior grounding assumed.
Computing Machinery and Intelligence
Alan Turing · 1950
Replaces 'can machines think' with an operational imitation game, and answers the standard objections in advance.
+3 more at this level
Assumes you know the vocabulary.
Minds, Brains, and Programs
John R. Searle · 1980
Argues via the Chinese Room thought experiment that manipulating symbols according to a program, however successfully it passes a behavioural test,…
+4 more at this level
9 works
Alan Turing
Replaces 'can machines think' with an operational imitation game, and answers the standard objections in advance.
Assumes no mathematics and no philosophy training; Turing wrote it to be argued with
28 pageslink checked 17 Sept 2026John McCarthy, Marvin L. Minsky, Nathaniel Rochester & Claude E. Shannon
Proposes that every aspect of learning and intelligence can in principle be described precisely enough to simulate on a machine, and asks funding for a summer workshop to find out — the document that named the field.
link checked 17 Sept 2026James Lighthill
Commissioned by the UK Science Research Council, concludes that AI research had failed to deliver on its grandiose promises and that most of its techniques did not scale past toy problems — the report credited with triggering the first AI winter's funding cuts.
link checked 17 Sept 2026Nils J. Nilsson
A biographical memoir tracing McCarthy's path from a politically radical Los Angeles childhood through Caltech, Dartmouth and Stanford to his founding role in AI — coining the term, inventing Lisp, and formalising common-sense reasoning.
17 pageslink checked 17 Sept 2026John R. Searle
Argues via the Chinese Room thought experiment that manipulating symbols according to a program, however successfully it passes a behavioural test, cannot by itself constitute understanding, because syntax alone never adds up to semantics.
link checked 17 Sept 2026Nils J. Nilsson
A participant's history of the field from its logical origins to the statistical turn, organised around which ideas actually worked and how long each detour took.
link checked 17 Sept 2026Gary Marcus
Lists ten open problems — data hunger, shallow transfer, opacity, weak integration with prior knowledge among them — that deep learning had not solved as of 2018, and argues progress will need hybrid architectures rather than scale alone.
link checked 17 Sept 2026François Chollet
Argues skill at any fixed task is a poor definition of intelligence because it rewards memorised priors over generalisation, and proposes measuring intelligence instead as skill-acquisition efficiency relative to the priors and experience a system is given.
link checked 17 Sept 2026David J. Chalmers
Works through the standard markers of consciousness — self-report, global workspace, embodiment — against what is actually known about large language models, and concludes current systems are unlikely but not obviously impossible candidates.
link checked 17 Sept 2026This subject genuinely sits in more than one domain. These fields approach the same ground with different methods.