The Textbook in Disguise

Kai-Fu Lee has said, in public, that he thinks his predictions are probably about 80 per cent accurate. It is a brave number to print on a book. It is also a number you can check, and we've started.

AI 2041 is built like a sandwich. Chen Qiufan, a novelist who lives in Beijing and Shanghai, writes ten short stories set in 2041. After each one, Lee, a venture capitalist and former president of Google China, steps in with an essay explaining the technology the story just dramatised and where it might break. Lee has described the aim as roughly an AI textbook taught through stories. The stories run, in order, from easy ideas to harder ones, and they are scattered across the world: Mumbai, Lagos, Australia and elsewhere. Chen has said he wanted AI not to belong only to the "AI superpower" countries, and the map shows it.

That's a clever design, and a very rare one. Most near-future fiction hides its homework. This one hands it in, with footnotes.

The art: the fiction is the sugar

Be fair first. The best stories here get a lot done in a short space. "The Golden Elephant", set in Mumbai, follows a family whose life is priced by an insurer's apps. It gets at the quiet way a model can turn into a gatekeeper. It doesn't preach, and it lets the product be partly good, which is what makes it believable. "Gods Behind the Masks" puts a Lagos hacker who makes deepfakes in the odd position of being hired to out-fake an anonymous political propagandist, and Chen gives her a real crisis of conscience. One reviewer singled out its Afrofuturist atmosphere, and it earns the praise.

"Twin Sparrows", about identical twins raised with personalised AI companions, turns nature versus nurture into a question about who your education is actually for. And "Quantum Genocide", whose subjects include a Bitcoin theft, quantum computing and autonomous weapons, is the book's deliberately dark entry. Lee said he wanted it darker "so that we can wake up and see how dangerous it is." Its own essay, one reviewer felt, is more unsettling than the plot.

Now the complaint. The fiction exists to illustrate a technology, and you can feel the rails. Characters explain things to each other in ways nobody does at a kitchen table. Plots arrive on schedule, make their point and stop. Several of the stories read as case studies with a pulse, and the book's weaker moments are when you can see the lecture coming a page early. One reviewer put it neatly: the stories are more illustrative than plot-driven, and they lean on Lee's essays to hold the collection together.

The essays are the better half, which is an odd verdict on a story collection. Lee writes clearly about deep learning, deepfakes, autonomous driving, quantum computing and the future of work, and he is unusually willing to set out the risks next to the promise. If you read only the stories, you get ten small pleasures. If you read only the essays, you get a good primer. Together they work, but never as well as the format promises.

The prophecy: small harms beat big bangs

Here the book is more interesting than the usual optimist's brochure, because Chen and Lee mostly chose ordinary harms over apocalypse.

The priced family. "The Golden Elephant" imagines an insurer that tracks a household and sets premiums from the data. We aren't far from that. In 2018 John Hancock announced fitness-tracking add-ons across its life policies, with a discount of up to 15 per cent on premiums, plus a cheap Apple Watch. It was optional, and Engadget corrected an early report that said it was compulsory. But the logic is the one in the story: your behaviour is the price list. The sharper worry is that a model can learn a proxy for a protected trait without anyone telling it to, and here the real world has begun to answer. Colorado passed a law in 2021 forbidding insurers from using external data and algorithms that unfairly discriminate on protected grounds, and told regulators to write the rules line by line. The future in the story is already a legal problem.

The faked election. The Lagos story imagines political deepfakes as a trade. They have become one. Two days before Slovakia's parliamentary election in September 2023, a fabricated recording of a party leader and a journalist discussing vote rigging spread on social media. Nobody can say how much it mattered to the result, and reporting on it is careful about that. In January 2024, a fake Biden voice told New Hampshire voters to skip the primary, and the regulator fined the consultant behind it $6 million. Chen's hacker works for the good side.

The driverless city. In the book, autonomous driving is one of the technologies that arrives and becomes unremarkable. Earlier this year Waymo was giving about half a million paid rides a week across ten US cities. In July it began fully driverless service in Las Vegas, and its target is a million weekly rides by the end of 2026. It is also still being corrected. Late in 2025 Waymo filed a voluntary software recall after its cars passed stopped school buses, and the Austin school district had logged 19 such cases. That's what 2041 will look like too: arrived, and under repair.

The quantum heist. This is the one the clock hasn't caught. Breaking widely used public-key encryption needs a quantum computer far beyond anything built. A 2025 estimate from Google researcher Craig Gidney cut the requirement for cracking 2048-bit RSA to fewer than a million noisy qubits, down from the 20 million estimated in 2019, and NIST published its first post-quantum standards in 2024 so that defenders could move first. The threat is real and the date is open. Twenty years is a plausible horizon for it. It is not a safe one.

What the book misses is mostly what a 2021 publication couldn't see. It was written before chatbots became a mass consumer product, and its stories treat AI as a set of specialised systems placed in separate domains, rather than one general conversational thing that everyone talks to. That's a fair miss, and one that every AI book from 2021 shares.

The reading

AI 2041 is a better teacher than storyteller. Its stories won't be on anyone's shelf for their sentences, and the ones that stay with you do so because a real company or a real law has since wandered into them. But the book's central bet was a good one. Its authors guessed that the near future of AI would be a series of ordinary, local, billable harms, spread across the planet and not stacked up in one catastrophe. Five years on, with insurers, election fixers and robotaxi fleets all in the record, that looks right.

Lee's 80 per cent was a boast. On this evidence, it's a fairer estimate than the sceptics would like. The rest of the score is for craft, and the craft is a competent sandwich.