Machines Who Think and the Old Dream of Artificial Intelligence
Pamela McCorduck's Machines Who Think is not only a history of artificial intelligence. It is a history of the human wish to meet intelligence outside the human body, and of the institutions that repeatedly turn that wish into programs, laboratories, funding cycles, public myths, and machines.
The Book
Machines Who Think: A Personal Inquiry into the History and Prospects of Artificial Intelligence first appeared in 1979 and was reissued in a revised 25th-anniversary edition in 2004. The current publisher listing from A K Peters/CRC Press gives the second edition as 576 pages and describes it as a history of the attempt to duplicate intelligence in an artifact. Google Books lists the Taylor & Francis edition as published March 17, 2004, with ISBN 9781568812052.
McCorduck was unusually close to the field she chronicled. Carnegie Mellon University Libraries notes that she described herself as an eyewitness to the birth and growth of AI, that she knew researchers at Stanford and Carnegie Mellon during formative decades, and that she donated research materials and oral-history interviews for Machines Who Think to the CMU University Archives in 1978. That proximity gives the book its texture. It is not a detached encyclopedia. It is a record of conversations, ambitions, rivalries, hopes, and conceptual habits from the inside edge of a new discipline.
The 2004 reissue matters because it stretches the book across two very different AI moments: the symbolic-AI world of early programs, expert systems, robotics, games, and cognitive speculation; and the early web-era world in which AI had become less exotic because pieces of it were already entering everyday life.
The Witness Position
McCorduck's advantage is that she treats AI as both science and culture. She cares about the technical work, but she also cares about the myths that made the technical work imaginable. The book moves from automata, mechanical reason, formal logic, and cybernetic aspiration into the modern research programs that tried to make machines prove theorems, play games, manipulate symbols, recognize speech, use knowledge, and model thought.
That dual lens keeps the book from becoming a simple origin story. AI does not emerge as one clean invention. It emerges as a series of translations: intelligence into symbol manipulation, learning into procedure, judgment into search, perception into representation, expertise into rules, language into formal operations, and human possibility into institutional projects.
This is why the book still reads as contemporary. Today's systems use different methods, but the translation pressure remains. Institutions still ask how much of thought can be made operational, measured, purchased, delegated, automated, and scaled. The new interface may be a chatbot, agent, recommender, classifier, or multimodal model, but the older question keeps returning: what has been preserved when cognition becomes machinery, and what has been silently thrown away?
The Old Dream
The title is deliberately strange. "Machines who think" gives machines a grammar usually reserved for persons. That grammar is part of the book's subject. People do not merely build tools; they also project mind, agency, intention, danger, and salvation into them. The dream of artificial intelligence is never just an engineering ambition. It is also a story about human exceptionalism under pressure.
McCorduck's history makes clear that AI did not begin with venture capital or product demos. It grew out of older fantasies of artificial life, formal reason, mechanical calculation, and the possibility that thought might have a structure separable from flesh. Nature's 2004 review record frames the book around that long arc: centuries of speculation about artifacts carrying out mental tasks, then Turing's 1950 challenge, then the 1956 Dartmouth conference where John McCarthy's name for the field took hold.
That lineage matters for reading the present. The current AI boom often presents itself as sudden: a model appears, an interface speaks, and public imagination reorganizes around it. McCorduck shows a slower pattern. The shock is prepared by centuries of metaphors and decades of institutional work. When the machine finally answers back, people are ready to hear more than a statistical system. They hear an old possibility returning through a new surface.
Labs, Schools, and Funding Weather
Machines Who Think is also a book about institutions. AI is shown not as pure thought but as a research ecology: universities, laboratories, military money, foundations, conferences, graduate students, personalities, publications, rival paradigms, and public expectations. That ecology matters because each form of support favors certain ideas of intelligence.
A theorem prover, a chess program, a robot, an expert system, and a language model do not merely demonstrate different technical capabilities. They enact different theories of what intelligence is for. Is intelligence search? Is it symbolic reasoning? Is it embodied adaptation? Is it expert knowledge? Is it language performance? Is it prediction? Each answer attracts funding, methods, benchmarks, critics, and institutional incentives.
The book's history of optimism, resistance, disappointment, and renewed ambition is useful because AI keeps moving through cycles of legitimacy. Public claims race ahead of deployed capacity. Critics point to missing commonsense, embodiment, context, meaning, or accountability. A new technique shifts the boundary of what machines can do. The public then forgets that earlier disappointments were not merely failures of hardware. They were also failures of framing.
The AI-Age Reading
Read in 2026, the book's central value is historical sobriety. It makes current AI feel less like an alien rupture and more like a powerful new chapter in a long human argument about mind, mechanism, and authority.
That does not make the present harmless. Large language models have changed the social form of AI by putting machine-generated language into ordinary work, education, search, companionship, software development, media production, and administration. Earlier AI often appeared as a specialized system. Current AI appears as an addressable presence. It can write, summarize, advise, imitate, tutor, flatter, refuse, remember, and act through tools. That social form intensifies the old projection problem McCorduck studied.
The important question is not whether the machine "really thinks" in a single metaphysical sense. The practical question is what people and institutions do once a machine becomes credible enough to receive tasks that used to require judgment, trust, discretion, or care. A system can be philosophically nonconscious and still reorganize classrooms, courts, companies, churches, friendships, newsrooms, and governments around its outputs.
McCorduck helps separate two confusions. One confusion is mystification: treating AI systems as minds because they produce mind-like signs. The other is dismissal: treating mind-like signs as socially irrelevant because the machine lacks inner life. The site-relevant zone is between those errors. Interfaces become powerful when their outputs are taken up by humans, institutions, markets, and feedback loops.
Where the Book Needs Updating
The book necessarily stops before the transformer era, modern foundation models, synthetic media at consumer scale, reinforcement learning from human feedback, prompt injection, data-center politics, AI safety institutes, model cards, tool-using agents, and the current scramble over compute, training data, copyright, and governance.
It also reflects the networks of people and institutions available to McCorduck. Its witness position is a strength, but closeness can privilege founders, labs, and named intellectual lineages over hidden labor, data work, affected communities, extractive supply chains, and the people who experience AI first as an administrative system rather than an intellectual adventure.
For that reason, the book should be read beside newer work on data extraction, labor, classification, surveillance, race, gender, platform power, and governance. Machines Who Think gives the intellectual genealogy. It does not provide the whole political economy of today's machine intelligence.
The Site Reading
The most durable lesson is that AI is never only a machine. It is a belief system, a research program, a funding object, a media story, a product interface, and an institutional permission structure.
Recursive reality begins when the machine's theory of intelligence starts shaping the human environments that will be used to prove the theory. If intelligence is treated as text prediction, institutions produce more text for machines to process. If intelligence is treated as ranking, people learn to live by scores. If intelligence is treated as agentic task completion, organizations rebuild work around delegable actions. The model does not simply enter the world. It helps produce the next version of the world it will be asked to model.
McCorduck's book is valuable because it slows down the enchantment. It reminds readers that today's talking systems stand on older dreams of artificial mind, older conflicts about mechanism and meaning, and older institutions that turned imagination into infrastructure. The right response is neither worship nor contempt. It is historical memory, technical specificity, and a refusal to let the oldest dream become the newest excuse for unaccountable authority.
Sources
- A K Peters/CRC Press, Machines Who Think: A Personal Inquiry into the History and Prospects of Artificial Intelligence, publisher listing, description, contents, edition, and page count.
- Google Books, Machines Who Think, bibliographic record, publication date, publisher, ISBN, and page count.
- Carnegie Mellon University Libraries, "Pamela McCorduck's Contributions to the Birth of AI Continued Through Her Generosity", obituary, archive record, and context on McCorduck's interviews and AI history work.
- Nature, John L. Casti, "Synthetic thought", review record for the 2004 edition, February 19, 2004.
- Legacy.com / New York Times notice, Pamela McCorduck obituary, October 31, 2021.
- Carnegie Mellon ETC Press, This Could Be Important, description noting McCorduck's 1979 AI history and later memoir.
Book links are paid affiliate links. As an Amazon Associate I earn from qualifying purchases.