Preface

The historian seeks to understand the boundary between heaven and humanity, the changes from past to present, and to make of it a coherent voice. — Sima Qian

Chicago, 1943. A neurophysiologist in his forties and a homeless youth of twenty — Warren McCulloch and Walter Pitts — sat in a cramped office and wrote down the first mathematical model of a neuron, translating the brain into logic. Thirteen years later, a small group of mathematicians, engineers, and psychologists gathered at an Ivy League college in New Hampshire and gave their shared pursuit a name: Artificial Intelligence. Eighty years on, in 2026, the offspring of that discipline has found its way into every pocket — writing code, reading papers, assisting research, even listening through the late hours of the night.

From a logical calculus on a sheet of paper to the global tide of large models, AI has traveled more than eighty years. Along the way there has been feverish optimism, painful winter, accidental breakthrough, and one more attempt, again and again, to ask what intelligence really is. This is not only a history of technology. It is an epic of humanity trying to understand its own mind, and to extend that mind into a force that serves civilization.

Why this book

AI is changing our world at a speed without precedent — helping scientists discover new drugs, letting the visually impaired "see" images, offering instant translation to billions of people. And yet, for most of the public, knowledge of AI begins and ends with ChatGPT and a handful of large-model products. Few realize that today's wave stands on more than eighty years of exploration — every breakthrough on the shoulders of those who came before, every winter carrying within it the seed of the next spring.

This book records the history of AI in the five-fold structure of Sima Qian's Records of the Grand Historian. It contains ninety chapters, more than seventy core figures, and over one hundred and thirty timeline entries strung across eighty years:

• Annals: ten chronicles, organized around ideological conflict, recording the central debates and pivots of each era
• Houses: thirty-one institutional histories, covering the five major regions — United States, Canada, China, Japan, Europe — from the MIT AI Lab, OpenAI, NVIDIA, TSMC, Apple, and Amazon to Mistral, DeepSeek, and Kimi, recording the organizations that shaped the field
• Biographies: twenty-three lives, recording the fates, choices, and contributions of key figures
• Treatises: twenty-five thematic and national histories, tracing the road-fights of neural networks, NLP, and other domains, alongside the AI paths of individual nations
• Timeline: a chronology of the key inflection points

Every chapter closes with a "Historian's Note" — a brief judgment in the spirit of Sima Qian — and an "Eyewitness Accounts" section, which invites the community to contribute first-hand memory.

Why the Records of the Grand Historian form

There is no shortage of AI chronologies and technical surveys. Most of them, however, organize around either time or technique, and so they struggle to convey the many faces of history — the way institutions maneuver, individuals choose, and technical paths split and reconverge, all within a single decade.

The five-fold structure that Sima Qian invented solves precisely this problem. Annals provide the panorama of the age; Houses look inside institutions; Biographies focus on the fate of individuals; Treatises trace the lineage of ideas; the Timeline supplies a quick-reference index. Five dimensions intertwine, and the same event can be revisited from several angles —

Take 1969. Minsky and Papert publish Perceptrons, proving mathematically the limitations of single-layer neural networks. In the Annals, this is the watershed that ends the golden age and ushers in the first winter. In the Biographies, it is the most controversial decision of Minsky's life, and one of the last blows Rosenblatt absorbs before drowning in Chesapeake Bay two years later. In the Houses, it is the coronation of the symbolist line at the MIT AI Lab. In the Treatises, it is the start of a twenty-year freeze on neural networks — until the backpropagation paper of 1986 brings them back. In the Timeline, it is a single line — but behind that line lie nearly two decades of silence.

Or take Geoffrey Hinton's long defense of backpropagation in Toronto. From 1986 to 2006 — a full twenty years — almost no one in the mainstream answered. You will read of him in the Annals on the eve of the deep-learning explosion, in the Biographies for the cost of his choices, in the Houses for how Google Brain raced to bring him in.

This "one event, many tellings" is the essence of the Records, and it is also the form a history as tangled as AI's needs.

Who this is for

Whichever page you open, may it give you something.

• If you work in AI: here is the family tree of the tools you use every day. Behind every line of PyTorch you write, a chain of names is standing.
• If you are a student or researcher: there is no textbook coldness here, only a series of real choices — see how those before you bet their direction with incomplete information.
• If you are a curious reader: no programming background is required. Every chapter enters through story and character; the technical concepts will arrive naturally in context.
• If you love history and the humanities: the history of AI is itself a history of ideas — about what intelligence means, about the philosophy of human-machine relations, about the rise and fall of scientific communities.

Why open source

A book about AI ought, by its nature, to be written in the open. History should not be written by one hand alone. The development of AI is the work of countless researchers, engineers, founders, and thinkers; no single author can see all of it whole. Through open collaboration, we hope to:

1. Reduce bias — diverse perspectives can correct an individual's blind spots
2. Stay current — the field moves fast; an open model allows continuous update
3. Invite participation — bring more people into the work of recording AI's history
4. Ensure accuracy — the community's many eyes can find and fix factual errors

How to read this book

You can approach the book in any of these ways:

• Chronologically: start with the first Annal, The Dawn, and read in order
• By interest: jump straight to a biography or treatise that draws you in
• By institution: trace the rise and evolution of organizations like OpenAI and DeepMind through the Houses
• At a glance: skim the Timeline to take in the whole arc

Not sure where to begin? Try one of these:

If you are interested in…	Suggested entry point
The origin story of AI	Annals · The Dawn
Why deep learning won	Annals · The Vision Revolution
One person who bent history	Biographies · Hinton
The roots of the LLM era	Annals · Generative AI
The drama inside OpenAI	Houses · OpenAI

A note on the writing

This book was written with the help of an AI tool (Claude) for research, drafting, and revision. But the choice of subjects, the narrative shape, the historical judgments, and the values throughout are decided by human authors, and every factual claim has been checked and cross-verified by hand. AI is a pen, not an author — what to tell, what to emphasize, and what to leave out remain human choices. A book about the history of AI, written with the help of AI, is itself a footnote of this age.

On contributions

If you find an error, or wish to add something, you are welcome to contribute on GitHub. See How to Contribute.

Between Heaven and Humanity

The story of AI is still being written. We record the past so as to better understand the present, and to face the future with clearer eyes. You are welcome to join the record.