Parmy Olson, a Bloomberg technology columnist, tells the story of modern artificial intelligence as a double biography: Sam Altman of OpenAI and Demis Hassabis of DeepMind, two founders who both promised to build machines smarter than people and to do it for humanity rather than for shareholders. The book won the Financial Times Business Book of the Year in 2024, and it earns that on reporting rather than prophecy. It is much less interested in whether AI will become conscious than in who currently owns it.
The core argument
Olson’s thesis is that idealism about artificial general intelligence was never structurally protected, and so it lost. Hassabis, a former chess prodigy and games designer turned neuroscientist, founded DeepMind in London with the aim of “solving intelligence” and then using it to solve disease and climate change. Altman, a Y Combinator operator with an unusual talent for getting powerful people to say yes, co-founded OpenAI as a non-profit explicitly positioned as a counterweight to Google’s growing dominance, with early backing from Elon Musk and Peter Thiel among others.
Both then ran into the same wall: training frontier models costs more money than any independent lab can raise. DeepMind sold to Google in 2014 and spent years trying to claw back autonomy — pushing for an ethics board with real teeth, and later for a spin-out structure with independent oversight so that control of AGI would not sit inside one advertising company. Google declined. OpenAI took a different route to the same destination, bolting a capped-profit subsidiary onto the non-profit and accepting billions from Microsoft. Olson calls this pull a vortex. Neither founder was corrupted in any cartoonish sense; both simply discovered that the only available fuel came with an owner attached.
The final act is the ChatGPT release in late 2022, which turned a slow research rivalry into a public arms race, panicked Google into shipping faster than it wanted to, and made safety talk commercially inconvenient at exactly the moment it mattered most. The 2023 board crisis that briefly ousted Altman reads, in Olson’s telling, as the last time the original governance structure tried to assert itself — and lost within a week.
Key ideas
- Capital captures mission. Founding charters, ethics boards and capped-profit wrappers are weak against whoever is paying the compute bill.
- Two temperaments, one outcome. Hassabis the scientist and Altman the dealmaker took opposite routes and both ended up inside a hyperscaler’s orbit.
- AGI as a belief system. Olson is sharp on the quasi-religious register of the field, where abundance and salvation are discussed as scheduled events rather than hypotheses.
- The wrong risk got the airtime. Existential doom is cheap for incumbents to talk about; bias, misinformation and labour displacement are expensive, so they get less.
- The critics who were pushed out. Researchers such as Timnit Gebru and Margaret Mitchell raised documented harms in large language models and found their positions untenable.
- Regulation lags dependency. Governments increasingly need these firms to deliver on public problems, which weakens the leverage they would need to constrain them.
Who it’s for
Best if you have followed AI headlines without ever getting a coherent account of how the industry actually assembled itself, and you want the corporate and governance history rather than another explainer on transformers. Olson is a reporter first; the book is briskly paced, well sourced and largely free of futurology. The limitations are the ones any book on this subject has. It moves fast enough that technical detail is thin, some scenes clearly rest on partisan recollections, and the pace of the field means specific events already read as historical. It is also, by design, a story about two men and their funders — if you want the labour, energy or Global South dimensions of AI, look elsewhere. As an account of how a stated commitment to safety got out-competed by a commercial race, it is the best single volume available.