Machines Who Think
Source: https://archive.org/details/machineswhothink0000mcco ↗
The first narrative history of artificial intelligence, written by someone who personally knew the founders — McCarthy, Minsky, Newell, Simon, Samuel.
McCorduck traces the dream of intelligent machines from antiquity through the Dartmouth conference to the expert systems era, with first-hand interviews that capture the ambitions, rivalries, and blind spots of the field's creators.
The revised 2004 edition extends the story through neural networks and the internet.
As a historical document it is irreplaceable: no other account gives you the texture of what it felt like to be inventing AI in the 1950s and 1960s.
For anyone trying to understand why AI's current moment echoes so many earlier ones, McCorduck provides the long view.
Central argument
McCorduck's central argument is that the dream of thinking machines is not a product of the computer age but a persistent human aspiration stretching from antiquity through medieval automata to the formal logic of the nineteenth century — and that the specific shape AI took in the twentieth century was determined less by technological inevitability than by the particular personalities, feuds, and philosophical commitments of a small founding community. By documenting the Dartmouth circle through first-hand interviews, she shows that the field's core assumptions — symbolic reasoning, the brain-as-computer metaphor, optimistic timelines — were choices made by specific people in specific rooms, not neutral conclusions forced by evidence. The implication is that AI has always been shaped as much by its creators' blind spots as by its successes.
Critique
Because McCorduck was personally close to the founding generation — McCarthy, Minsky, Newell, Simon — the account risks hagiography: the insiders get texture, interiority, and sympathy, while critics of the symbolic AI paradigm (early connectionists, skeptics like Dreyfus) tend to appear as obstacles in someone else's story rather than as serious intellectual alternatives with their own merits. This proximity that makes the book irreplaceable as primary source material is also what limits its capacity for structural critique of the field's founding assumptions. A reader relying on McCorduck alone would underestimate how contested those assumptions were from the very beginning.
Why it matters for product
For a CPO, the book's core lesson is that technological paradigms are set early and by small groups, and that the assumptions baked in at founding — what counts as intelligence, what counts as success, which problems are worth solving — tend to outlast their original justifications and constrain every subsequent product decision built on top of them. When evaluating AI capabilities to integrate into a product roadmap, McCorduck's history is a useful corrective against vendor narratives of inevitability: the current moment has precise historical echoes in the expert systems enthusiasm of the 1980s, including the same pattern of impressive demos, overstated generalizability, and eventual disillusionment. Understanding that cycle helps a product leader set realistic capability bets, avoid over-indexing on a single paradigm, and ask harder questions about what the current generation of AI builders might be systematically unable to see about their own work.