The Age of A.I.

Henry Kissinger spent the last years of his life thinking about machines. The collaboration that became The Age of A.I., published in 2021, grew out of that preoccupation, and out of two conversations he had been having for some time. The first was with Eric Schmidt, the former Google chief executive who by 2020 was chairing the U.S. National Security Commission on Artificial Intelligence and had become one of Washington’s most active voices on the strategic implications of the technology. The second was with Daniel Huttenlocher, the inaugural dean of MIT’s Schwarzman College of Computing, founded in 2019 with a billion-dollar gift to anchor the integration of computing into every discipline. Kissinger himself dated his interest to a 2016 Bilderberg meeting where a demonstration of AlphaGo’s play against Lee Sedol unsettled him; he then published a 2018 essay in The Atlantic titled “How the Enlightenment Ends” arguing that the West was wandering into a philosophical revolution it had not prepared for. The book extends that essay, and inserts a former Secretary of State, a Silicon Valley executive who once ran the world’s most influential information company, and a computer scientist running a billion-dollar academic experiment into the same argument. The conversation the book joins is the one that had been gathering since roughly 2016, when AlphaGo, deep learning, and the early large language models pushed AI out of the lab and into headlines about strategy, jobs, and warfare.

The central argument is that artificial intelligence represents a new mode of knowing, and that humanity has not yet built the philosophical, political, or strategic apparatus to live with it. The authors argue that since the Enlightenment, Western civilization has rested on a tacit agreement that knowledge is what human reason can produce and verify. Faith preceded reason; reason then displaced faith; what remained was a world in which truth claims had to be defensible by a human mind to a human mind. AI breaks that compact. A system like AlphaZero, trained without human games, plays chess in styles no grandmaster recognizes; a deep-learning model trained on molecular structures suggests an antibiotic that human chemists would never have isolated; a transformer-based language model produces fluent prose that captures regularities its programmers cannot fully describe. In each case, the authors argue, humans can use the output and cannot, in any complete sense, explain why it is what it is. That gap — between what we can deploy and what we can justify — is the book’s recurring subject. It is presented not as a technical bug to be patched but as a category change comparable to the printing press, the scientific method, or nuclear weapons.

The book proceeds in seven chapters. An opening chapter, “Where We Are,” sets the scene with anecdotes — including the discovery of the antibiotic halicin by researchers at MIT, who in 2020 used a neural network to identify a molecule effective against drug-resistant bacteria after screening over a hundred million compounds in silico. A second chapter, “How We Got Here,” traces the intellectual lineage from medieval philosophy through the Enlightenment to the post-war computing revolution, framing AI as the latest in a long sequence of shifts in what counts as authoritative knowledge. The third, “From Turing to Today — and Beyond,” walks readers through the technical genealogy of machine learning: symbolic AI, expert systems, the connectionist turn, the rise of deep neural networks, reinforcement learning, and the transformer architectures that powered GPT-3, which the authors discuss at length. The fourth chapter, “Global Network Platforms,” examines the companies and services — search engines, social media, cloud providers — that have become the substrate on which AI is built and through which it reaches billions of people. The fifth, “Security and World Order,” is the most explicitly geopolitical, and is where Kissinger’s voice is most audible. The sixth, “AI and Human Identity,” is the most philosophical, asking what becomes of selfhood, agency, and creativity when machine intelligences participate in everyday life. A final chapter, “AI and the Future,” returns to policy and proposes principles for living with the technology.

The concrete examples are the heart of the book. AlphaZero appears repeatedly: trained only on the rules of chess, it produced styles of play that grandmasters described as alien — sacrificing pieces in positions that human theory would call losing, then converting them. The point the authors press is not that AlphaZero is strong but that it is opaque; the moves it makes are not derivable from any human theory of chess, and yet they work. Halicin is offered as a parallel case in medicine: a model trained to predict antibacterial activity flagged a molecule whose structure had no obvious relationship to existing antibiotics, which then proved effective in animal trials. GPT-3, then the largest publicly known language model, anchors the discussion of generative AI; the authors note that its outputs are at once fluent and unreliable, and that its facility with language outpaces any settled account of why it works. Image recognition, recommendation systems, and ad-targeting models illustrate the more ambient kind of AI, embedded in everyday products. On warfare, the book points to autonomous and semi-autonomous systems: loitering munitions, swarming drones, automated cyber-defence, and the gradual compression of the decision loop in which a human operator approves or vetoes a machine’s recommendation. The cases are not encyclopaedic — the book is not a survey of weapons programmes — but the cumulative impression is of a technology already at work in domains that matter, and outrunning the institutions meant to govern it.

The strategic chapter draws on Kissinger’s long preoccupation with nuclear deterrence. He had spent the 1950s and 1960s thinking publicly about how rival powers stabilize a relationship under conditions of catastrophic risk; the book asks whether anything like that stabilization is possible with AI. The authors argue that AI complicates deterrence in several ways. First, the technology is dual-use and diffuse; unlike enriched uranium, the code and the models can be copied, shared, and reproduced cheaply, which makes verification regimes hard to imagine. Second, autonomous systems compress decision time below the threshold at which human deliberation is feasible, raising the risk of inadvertent escalation. Third, the inscrutability of advanced systems means commanders cannot always know what their own weapons will do in novel circumstances, let alone what an adversary’s weapons will do. The authors do not propose a treaty. They propose that the great powers, particularly the United States and China, recognize the strategic novelty of the technology and begin a sustained dialogue about restraints — much as the Cold War rivals eventually built channels around nuclear weapons through SALT, ABM, and the Open Skies arrangement. The book acknowledges that the political conditions for such a dialogue are weaker now than they were in the 1960s, and offers no confidence that one will be built in time.

The philosophical chapter is where the book’s distinctive voice is clearest. The authors argue that AI changes the texture of human experience in ways that go beyond economics or warfare. When a doctor’s diagnostic recommendation comes partly from a system whose reasoning the doctor cannot inspect, the locus of medical judgment shifts. When a judge consults a recidivism model, the sources of justice change. When a child grows up speaking to a conversational agent before learning to write, the developmental sequence of literacy and selfhood is rearranged. The authors are careful not to forecast specific outcomes; they argue that the cumulative shift in who, and what, takes part in human reasoning will reshape categories — knowledge, creativity, identity — that have been stable for centuries. They invoke Kant, Descartes, and the Enlightenment philosophes not to claim continuity but to mark a break. The book’s tone in these passages is melancholy: there is no triumphalism about progress and no Luddism about loss. It is the voice of a statesman who has watched several civilizational shifts and is registering another.

Reception was mixed, and split along predictable lines. Reviewers in policy publications welcomed the book as the first serious attempt by a former senior official to think about AI at the strategic level; reviewers in technical publications complained that the authors waved at the technology rather than engaging with it, and that the philosophical claims rested on a thin reading of recent machine-learning research. Several reviewers noted that the most concrete arguments — about deterrence, about platform power, about opacity — had already been made by specialists, and that the book’s value was in the assembly rather than in original analysis. Others pointed out that Eric Schmidt’s commercial and policy roles raised questions about the book’s framing of platforms and competition with China, and that some of the proposed responses to AI risk were notably congenial to large American technology firms. The book sold strongly and entered the policy reading lists, and it is now routinely cited in National Defense University discussions and in European Parliament debates on the AI Act, more often as a reference point than as a source of operational detail. Within the academic AI safety community, the book is treated as a generalist’s entry to a conversation those researchers had been having for a decade — useful for senior officials, but not where the technical arguments live.

For someone reading widely on AI in war and on AI in society, the book functions as a philosophical bookend rather than a manual. It pairs naturally with Paul Scharre’s Army of None, which appeared in 2018 and covers the autonomous-weapons debate in much greater technical and operational detail; with Stuart Russell’s Human Compatible, which makes the technical case for redesigning AI objectives; and with Kai-Fu Lee’s AI Superpowers, which sets out the US–China competitive frame more explicitly. Against those, The Age of A.I. is shorter on specifics and longer on register; readers looking for weapons system descriptions, model architectures, or policy proposals will find more depth elsewhere. What this book offers that the others do not is the voice of a Cold War strategist applying his habit of mind to a new technology, and the implicit claim that the people running governments will, sooner or later, have to think this way. The book does not cover the labour-market effects of AI in any depth, says little about bias or fairness, and treats the open-source ecosystem briefly. Its lens is high politics and high philosophy.

Some of the book is already dated. The discussion of GPT-3 has been overtaken by GPT-4 and successor models, and by the explosion of open-weight models from Meta, Mistral, and Chinese labs; the platform analysis predates the regulatory turn embodied by the EU AI Act and the U.S. executive orders that followed; the Sino-American strategic framing has hardened in directions the book sketches but does not fully anticipate. What looks likely to age well is the central frame: the argument that AI represents a new kind of knowing, that the institutions humanity built to govern itself were designed for a world in which knowing was human, and that the work of building new institutions has barely begun. That argument does not depend on which model is current, on which company leads, or on which administration is in office. It is the kind of claim that becomes more, not less, relevant as the underlying technology accelerates, and it is the contribution that justifies the book’s place on the shelf alongside the more technical works it does not try to replace.