A Dispatch From the Last Sane Era of AI

Sebastian Mallaby negotiated deep access to Demis Hassabis in November 2022, just a week before ChatGPT launched. Mallaby began his interviews as the pre-2022 era of artificial intelligence—when the field still possessed protagonists—was rapidly being displaced by faceless monoliths and chat-box interfaces, as detailed in his book The Infinity Machine: Demis Hassabis, DeepMind and the Quest for Superintelligence.

Early AI and the Founding Vision of DeepMind

Demis Hassabis grew up poor in North London, achieved chess master status at 13, shipped a bestselling video game at 17, and earned a neuroscience PhD as a stealth route to artificial intelligence, according to the book’s biographical accounts. Early AI was “narrow,” requiring programmers to write code for one specific, deterministic output where identical inputs always produced identical results. Hassabis established DeepMind around an entirely different founding thesis: true intelligence relies on a flexible, general-purpose learning algorithm that reverse-engineers human brain learning from scratch to build Artificial General Intelligence.

AlphaFold and the Biology Breakthrough

DeepMind tackled a 50-year-old grand challenge in biology by treating protein folding as a data problem rather than simulating raw physical and chemical forces, according to Sebastian Mallaby’s reporting. By leveraging evolutionary patterns and attention networks similar to those powering modern chatbots, AlphaFold predicted the 3D structures of amino acid chains and was released as a free “Google Maps for biology” to accelerate global medicine research. However, scaling AlphaFold required massive capital, driving Hassabis to steer DeepMind into Google’s arms, illustrating how raw models depend heavily on surrounding harnesses of compute, capital, and post-training tools.

Narrative Control and the Rush to Commercialization

Hassabis exercised strong narrative control over his biographer, utilizing persuasion techniques that colleagues described as “Jedi” mind tricks, according to The Infinity Machine. The book largely omits the compute-industrial base pressures that pushed DeepMind toward Google, as well as Google’s pivot away from initial promises to keep the technology isolated from weapons and surveillance. Hassabis spent a decade building AI as a scientific instrument, only to see it conscripted into a commercial gold rush as Google raced to catch up with OpenAI and Anthropic through the Bard and Gemini model families.

Regulatory Failures and Global Realities

Hassabis attempted to install safeguards—such as an ethics board that never met and an independent spin-out charter that Google ultimately smothered—to protect the world from unregulated AI, according to Mallaby’s late chapters. Mallaby initially concluded that AI safety rests on trust among builders, but geopolitical realities quickly outpaced that doctrine. Within months of the book’s publication, the US Department of War blacklisted Google rival Anthropic for refusing surveillance work, while the Department of Commerce restricted Anthropic’s frontier model Fable through export controls. Meanwhile, Chinese labs operating under US sanctions on Huawei chips have scored competitively against proprietary models using open models.

Frequently Asked Questions

Who wrote The Infinity Machine?

The book was written by Sebastian Mallaby and published on August 2, 2026, documenting the career of Demis Hassabis.

What is Demis Hassabis’s biggest achievement?

According to biographical records and his 2024 Nobel Prize in Chemistry, Hassabis’s singular achievement is AlphaFold, which predicted the 3D structures of nearly all known proteins.

How did DeepMind become part of Google?

DeepMind joined Google because scaling complex AI architectures like AlphaFold required massive capital and infrastructure that a standalone lab could not sustain independently.