Best Books on the History and Future of Artificial Intelligence

Artificial intelligence has been declared dead and reborn several times in its short history. The field began in the 1950s with enormous optimism, hit two prolonged periods of reduced funding and interest known as "AI winters," and then, in the 2010s, broke through into capabilities that surprised even the researchers building the systems. Understanding how that happened, and where it might lead, requires more than following the latest news cycle. It requires history. ## Where AI Actually Started The popular origin story for AI begins with Alan Turing, whose 1950 paper "Computing Machinery and Intelligence" proposed the famous imitation game as a test of machine thinking. Turing was asking whether a machine could respond to questions in a way indistinguishable from a human, and his paper framed the question in a way that shaped the field for decades. But the history is longer and more tangled. The mathematical foundations came from multiple directions: Turing's work on computation, Norbert Wiener's cybernetics, Claude Shannon's information theory, John von Neumann's work on computer architecture. The first explicit AI research program, including the coinage of the term "artificial intelligence," came from a 1956 summer workshop at Dartmouth organized by John McCarthy and Marvin Minsky. What followed was a cycle that repeated itself: a promising technique, extravagant predictions, practical limitations that the predictions ignored, funding cuts, a period of reduced activity, and then a new technique that seemed to solve the problems of the last one. ## Pamela McCorduck, "Machines Who Think" First published in 1979 and updated in 2004, this is the foundational history of the field. McCorduck interviewed many of the founders personally, including Minsky, McCarthy, Herbert Simon, and Allen Newell, and she wrote with both technical understanding and literary skill. The book covers the intellectual history from the automata of the eighteenth century through the neural network revival of the early 2000s. McCorduck is interested not just in what the researchers built but in what they believed they were doing: the philosophical commitments, the personal rivalries, the moments of genuine excitement and crushing disappointment. She captures the culture of AI research in a way that technical histories miss. Reading McCorduck now, knowing what came after, is a strange experience. The problems that seemed intractable in 1979, and again in 2004, have since been at least partially solved, in ways that the field's founders did not anticipate. The techniques that eventually worked, particularly deep neural networks trained on massive datasets, were not the approaches that the leading researchers of the 1970s thought most promising. ## Nils Nilsson, "The Quest for Artificial Intelligence" Nilsson was a researcher at SRI International and later Stanford, and he participated in several of the field's key developments, including early work on robotics and planning. His history, published in 2010, is more technical than McCorduck's and more focused on the research itself rather than the people. It is comprehensive in a way that few histories of any scientific field achieve. Nilsson traces the development of every major subfield: theorem proving, game playing, natural language processing, computer vision, robotics, machine learning. He explains what each approach tried to do, why it seemed promising, and what its limitations turned out to be. The book is not light reading, but it rewards attention. The section on machine learning, which traces neural network research from the perceptrons of the 1950s through the backpropagation algorithm that made multi-layer networks trainable, is particularly valuable for understanding why the current moment in AI is not just another hype cycle. The techniques that are working now were developed over decades by researchers who often had no funding and little institutional support. ## The Neural Network Comeback The modern era of AI began, depending on how you count, either in 2006 when Geoffrey Hinton published a paper on deep belief networks, or in 2012 when a system trained by Hinton's students won an image recognition competition by a margin that made the previous approach look antique. The technology that followed, transformers, large language models, image generation, reinforcement learning from human feedback, all descends from that line of research. Hinton, along with Yann LeCun and Yoshua Bengio, received the 2018 Turing Award, computing's highest honor, for their work on deep learning. The citation noted that they had pursued neural network research during the AI winters when it was professionally risky and institutionally unfashionable. What changed was scale. Neural networks trained on small datasets could learn only limited things. Neural networks trained on the text of the internet, on billions of images, on millions of hours of human interaction, learned things that no one had explicitly programmed. Whether that constitutes intelligence in any meaningful sense remains one of the field's most contested questions. ## What the Next Decade Might Look Like The honest answer is that the researchers most closely involved in building current AI systems disagree sharply about what comes next. Some believe that scaling existing architectures with more data and compute will continue to produce rapid improvements. Others argue that current systems have fundamental limitations that cannot be overcome without new conceptual breakthroughs. Still others believe that some of the systems already deployed are more capable, and potentially more risky, than mainstream discussion acknowledges. These disagreements are not merely academic. The decisions made in the next few years about how to develop, deploy, and govern AI systems will shape outcomes for a very long time. Understanding the history of the field, including its repeated cycles of overconfidence and disappointment, its genuine breakthroughs, and the philosophical debates that have run through it from the beginning, is not optional background. It is essential context. ## Further Reading Browse more technology books at [/category/technology](/category/technology)