Jeff Hirsh
3
Read: August 2026
Inspiration: heard much about DeepMind and wanted to understand its history and place in the AI race
Summary
Written with the help of ChatGPT, below is a brief summary to understand what is covered in the book.
| “The Infinity Machine”, published in 2026 by journalist and author Sebastian Mallaby, chronicles the life of Demis Hassabis and the rise of DeepMind as one of the world’s most influential artificial intelligence laboratories. Mallaby follows Hassabis from his childhood as a chess prodigy and video-game programmer through his studies in computer science and neuroscience and the founding of DeepMind in 2010. The book traces breakthroughs including DeepMind’s reinforcement-learning systems, AlphaGo, AlphaZero, and AlphaFold, while also exploring the company’s acquisition by Google and the tensions between scientific ambition and commercial priorities. It places Hassabis and DeepMind within the broader race to develop artificial general intelligence and examines the scientific, economic, and ethical implications of increasingly capable AI systems. Ultimately, the book portrays Hassabis’s pursuit of artificial intelligence as an attempt to build a general-purpose tool capable of accelerating discoveries across science and solving some of humanity’s hardest problems. |
Unedited Notes
Direct from my original book log, below are my unedited notes (abbreviations and misspellings included) to show how I take notes as I read.
Demis Hassabis obsessed from childhood with intelligence itself, chess prodigy by ~5 and master-level player young, saw games as laboratory for thinking/problem solving, intensely competitive but bigger question became whether intelligence could be understood + recreated in machine, grew up North London to Greek Cypriot father/Chinese Singaporean mother, unusual childhood with parents not from academic/tech elite, coding ability early and joined Peter Molyneux/Bullfrog as teenager working on games, Cambridge computer science then founded Elixir Studios, eventually stepped away and did neuroscience PhD because thought understanding human brain/memory could inform artificial intelligence, DeepMind founded 2010 with Mustafa Suleyman + Shane Legg, mission intentionally enormous—“solve intelligence” then use intelligence to solve everything else, early pitch was AGI at time when term sounded sci-fi, needed unusual investors willing to fund expensive research with little near-term product, Peter Thiel/Founders Fund among key early believers, early DeepMind work centered on reinforcement learning where agent learns by trial/error + reward rather than being explicitly programmed, breakthrough Atari work showed same system could learn multiple games from raw pixels with minimal game-specific instruction—important bc suggested general learning algorithm vs handcrafted AI, Google/Facebook become interested as AI talent race heats up, Google acquire DeepMind in 2014 for hundreds of millions, unusual structure because team cared about independence/safety and wanted protection against pure commercialization, tension thereafter between being scientific mission-driven lab and part of giant corporation that actually needs products, AlphaGo became public proof point—Go viewed much harder than chess due massive number of possible positions, DeepMind combine neural networks + reinforcement learning/tree search, beat Lee Sedol in Seoul in 2016 watched globally, famous Move 37 looked bizarre to humans but brilliant, machine not merely copying human play but finding strategies people had not conceived, next AlphaZero learn chess/Go/shogi largely from self-play and outperform specialized predecessors, Demis always saw games as stepping stone not endpoint, AlphaFold attacked protein folding where amino acid sequence known but 3D structure incredibly hard to determine, AlphaFold2 breakthrough around 2020 dramatically increased accuracy and eventually predicted structures for enormous number of known proteins, Demis + John Jumper awarded 2024 Nobel Prize in Chemistry for work, parallel story is DeepMind increasingly challenged by OpenAI/transformer/LLM revolution—ChatGPT in late 2022 shifts public perception and Google appears behind despite years as AI leader, DeepMind and Google Brain combined 2023 into Google DeepMind with Demis leading, Gemini becomes effort to compete at frontier across language/multimodal/reasoning, book also tracks personalities around modern AI—Hinton, Sutskever, Suleyman, Legg, Altman, Musk etc and different views on AGI/safety/commercialization, Demis’ vision somewhat distinct from pure chatbot race—AGI as scientific instrument that can discover new knowledge in biology/materials/physics and ultimately help solve fundamental mysteries, tension throughout is same thing that makes technology valuable also makes it potentially uncontrollable/power-concentrating, DeepMind began as idealistic independent research lab but to pursue frontier models requires huge compute/capital so ultimately needs Google-scale infrastructure, recurring question whether AGI can remain science project once embedded in one of world’s biggest commercial companies
