Books
Ten books on how the field got here
AI publishing splits into three piles: technical texts, futurist argument, and history. This list is almost entirely the third, because the history is what makes the other two legible — the arguments being had now are mostly older arguments with better hardware. Between them these ten cover 1956 to the present from the inside, from the outside, and from the positions that lost.
Number 1: Genius Makers
2021 · Cade Metz
Penguin Random House — Genius Makers
The best single narrative of the 2006–2020 arc.
Metz covers the shift from symbolic AI to data-driven neural networks as a group biography — Hinton, LeCun and Bengio in the wilderness years, then the auction that sent Hinton to Google and the talent race that followed. It reads like a business thriller and is the rare history that is equally good on the ideas and on the rivalries. The single best starting point on this list.
Number 2: Machines Who Think
1979 · Pamela McCorduck · updated 2004
Routledge — Machines Who Think
The original history, written while its subjects were still arguing about it.
McCorduck interviewed McCarthy, Minsky, Newell and Simon when the field was two decades old and its founders were in their prime and in open disagreement. No later history has that access. The 2004 edition adds hindsight, but the value is the 1979 core: AI described before anyone knew how the story turned out.
Number 3: The Quest for Artificial Intelligence
2010 · Nils J. Nilsson
Cambridge University Press — The Quest for Artificial Intelligence
The rigorous academic history of the symbolic era.
Nilsson was there for most of it and writes as a participant-historian: Dartmouth 1956 through the expert-system boom and both winters, with the technical content intact rather than paraphrased away. Dry by design and the most complete account of the period the deep-learning histories skip over. The reference the other books cite.
Number 4: Artificial Intelligence: A Guide for Thinking Humans
2019 · Melanie Mitchell
Macmillan — Artificial Intelligence: A Guide for Thinking Humans
A working researcher on what the systems can and cannot do.
Mitchell takes the field's claims seriously enough to test them, walking through what deep learning actually achieves and where it reliably fails — analogy, abstraction, common sense. Neither hype nor debunking, and it never condescends. The best book for a reader who wants to judge the claims themselves rather than pick a side.
Number 5: Empire of AI
2025 · Karen Hao · NBCC finalist
Penguin Random House — Empire of AI
The investigative account of OpenAI and the commercial turn.
Hao reported on OpenAI for years before this, and the book is the fullest record of how a non-profit research lab became the centre of a capital race — including the labour and resource costs that the founder narratives leave out. The definitive account of 2015–2024, and the necessary companion to Genius Makers, which ends just before this story starts.
Number 6: The Worlds I See
2023 · Fei-Fei Li
The memoir of the person who built the dataset that started it.
Li's ImageNet made the 2012 breakthrough possible, and the book braids that work with an immigrant childhood and a career spent arguing for a human-centred version of the field. It is the best crossover book here: the memoir carries readers who do not care about neural networks, and the science is not diluted for them.
Number 7: The Deep Learning Revolution
2018 · Terrence Sejnowski
MIT Press — The Deep Learning Revolution
Connectionism from inside the camp that spent thirty years being wrong.
Sejnowski co-invented the Boltzmann machine with Hinton and founded the conference that became NeurIPS, so this is a participant's account of why neural networks were dismissed for decades and what changed. Read alongside Nilsson it gives you both sides of the field's central argument, each told by someone who believed it.
Number 8: A Brief History of Intelligence
2023 · Max Bennett
Amazon — A Brief History of Intelligence
Five evolutionary breakthroughs in brains, each mapped onto a current AI limit.
Bennett works through how biological intelligence was assembled — steering, reinforcing, simulating, mentalising, speaking — and at each step asks which of them today's models have and which they do not. The comparison is disciplined rather than decorative, and it produces the clearest available framing of what 'missing common sense' actually means.
Number 9: Superintelligence
2014 · Nick Bostrom
The book that set the terms of the risk debate for a decade.
Bostrom's treatment of intelligence explosion, instrumental convergence and the control problem shaped how the frontier labs talk about their own work — several were founded partly in response to it. The argument is contested and parts have dated, but the vocabulary in use today comes from here, and reading the original is faster than reconstructing it from the arguments about it.
Number 10: Atlas of AI
2021 · Kate Crawford
Yale University Press — Atlas of AI
The counter-history: what the systems cost in minerals, labour and power.
Crawford follows AI down to the lithium mine, the data-labelling contract and the electricity bill, arguing that treating it as immaterial is itself a political act. It is the sharpest available corrective to founder-led accounts, and deliberately unsympathetic. Read after Genius Makers, not instead of it — the disagreement is the point.
On this siteScale AI