The Modern Oppenheimer, Demis Hassabis, and the Quest for Superintelligence

An episode in two halves. First Rebecca Patterson turns interviewer and questions Sebastian Mallaby about his new book on Demis Hassabis, the un-famous scientist who has been chasing superintelligence since the 1990s. Then a listener mailbag ranges across the dollar’s reserve-currency dominance, the economics of frontier AI, and how drones are rewriting warfare.

Key ideas

1. Hassabis is the OG of modern AI, and the un-famous counterpoint to Sam Altman. Mallaby’s book paints him as a real scientist — a Nobel laureate for AlphaFold, a chess prodigy and coding phenomenon who founded DeepMind in London in 2010, years before OpenAI existed. Where Altman wants power and Zuckerberg wants product, Hassabis is driven by science: at 17 he decided ‘Isaac Newton was a failure’ for never grasping the full ‘fabric of reality’, and set out to build a tool — superintelligence — that would carry him past Newton. He hopes for a second Nobel.
2. The Oppenheimer parallel is one the AI founders reach for themselves. Mallaby did not have to impose it: Altman volunteers that he shares Oppenheimer’s birthday, and Hassabis tells the story of Szilárd conceiving the nuclear chain reaction at a Russell Square crossing near his first office. Mallaby’s own point is harder — a lab cannot self-impose safety in a race dynamic, because a race is a race to the bottom: Google DeepMind can make its own system safer, but that does nothing if rival labs do not follow. The only lever left is to build clearly beneficial applications, above all in medicine, so society will accept the technology.
3. AlphaFold and Isomorphic Labs turn protein biology into a drug-discovery platform. Mallaby explains the breakthrough as ‘amino-acid origami’ — predicting from a DNA sequence how a protein self-folds into its intricate shape, a computational-biology challenge unsolved for decades until DeepMind cracked it in 2020. The 200-million-plus structures are open-sourced free to any scientist. The work has now moved to the proprietary sister company Isomorphic Labs (AlphaFold 4), modelling not just protein shapes but their dynamic interactions with other molecules — the foundation for accelerated drug discovery.
4. The dollar’s reserve status rests on four pillars China cannot match. Patterson lays them out: economic dominance that pulls in foreign investment; deep, liquid Treasury markets backed by an independent Fed; an innovative, profitable corporate sector; and the network effects that make dollar trades cheapest even between third countries. China’s closed capital account precludes it from supplanting the dollar — money cannot move freely in and out. The likelier counterweight is the euro, which could challenge ‘tomorrow’ if Europe harmonised its national bond markets into a single ~$30tn pool at Treasury scale — but that political agreement is near-zero-probability for now. Stablecoins, backed by short-term T-bills, are a US tool to juice Treasury demand and extend dollar dominance; the ECB, seeing a threat to its monetary policy, is pushing back with its own digital currency.
5. Frontier AI has a ‘first-mover disadvantage’, like pharma without the patents. US labs and consumers pay to be first — training runs cost hundreds of billions — while others free-ride by distillation, reverse-engineering a finished model to build a cheap replica. Huawei, which Mallaby visited in China, has explicitly chosen to skip the frontier and win on verticals and scale (AI to repair high-speed trains, to run phone health scans, for logistics). Monetisation therefore shifts to productised AI embedded in things people already use — Gemini across Google Drive and Gmail, Anthropic’s enterprise revenue — and the open question is whether that revenue can sustain frontier-model investment.
6. Drones have rewritten the arithmetic of war. Three-quarters of Russian casualties in Ukraine come from drones. They also overturn the Strait of Hormuz calculus: Iran can now selectively threaten hostile ships while keeping the waterway open for its own, without the post-war hangover of self-laid mines. Drones first look like an underdog’s weapon, but AI-controlled swarms — too numerous for interceptors — may swing the edge back to the manufacturing superpowers, the US and China, with Turkey an underrated third. The choke point is rare earths: China supplies 90% and could, today, stop a US drone programme.
7. Buyer’s remorse, choke points, and the productivity thread that ties it together. Polls show non-MAGA 2024 swing voters souring on the Gulf war as it hits their wallets, while diehard MAGA support holds firm. Mallaby flags colleague Eddie Fishman’s Choke Points for naming the weaponisable node in a supply chain — a phrase now everywhere. And the whole grab-bag loops back to the book: with US job-growth break-even near zero, the only escape from stagnant GDP is more productivity per worker — ‘so God bless AI.‘

Context

This is a grab-bag mailbag episode, but its centrepiece is Mallaby’s new book, The Infinity Machine. The AI-safety thread — a well-meaning founder trapped in a race to the bottom — and the China-race economics both connect to the wiki’s other Mallaby and Spillover coverage; the free-rider distillation argument in particular recurs in Sebastian Mallaby on OpenAI's Cash Crunch, the AI Bubble Debate, and the China AI Race.

Related

• Sebastian Mallaby — co-host; the book is his
• Rebecca Patterson — co-host; the reserve-currency answer is hers
• Demis Hassabis — subject of the book
• The Infinity Machine — the book
• Sebastian Mallaby on OpenAI's Cash Crunch, the AI Bubble Debate, and the China AI Race — the AI-race economics recur there