The reader's edition
The Infinity Machine: Demis Hassabis, DeepMind, and the Quest for Superintelligence: the summary
Sebastian Mallaby is a journalist and a senior fellow at the Council on Foreign Relations, and his earlier books include The Power Law, about venture capital. The Infinity Machine is a biography of Demis Hassabis, the chess prodigy and game designer who co-founded DeepMind in London in 2010 and now leads Google DeepMind. Mallaby follows the lab's two-step aim: build general intelligence, then use it to speed up science. The story runs from Hassabis's childhood in North London through the sale to Google, AlphaGo's 2016 match against Lee Sedol, the AlphaFold protein work that earned him a share of the 2024 Nobel Prize in Chemistry, and the race that followed ChatGPT. It draws on long interviews with Hassabis and with colleagues, rivals and critics, and weighs what he has built against the risks of building it. Read a chapter at a time, with a printable action book at the end.
- Sebastian Mallaby
- 2026
- 4 chapters
- 5 min read
- 9 takeaways
This is HourLife's reading of The Infinity Machine: Demis Hassabis, DeepMind, and the Quest for Superintelligence, in our own words. It is not a substitute for the book.
Chapter 01
Before you start
The book in a minute
Tap a paragraph to highlight it. Turn pages with the arrows, the arrow keys, or a swipe.
Sebastian Mallaby is a journalist and a senior fellow at the Council on Foreign Relations, and his earlier books include The Power Law, about venture capital. The Infinity Machine is a biography of Demis Hassabis, the chess prodigy and game designer who co-founded DeepMind in London in 2010 and now leads Google DeepMind. Mallaby follows the lab's two-step aim: build general intelligence, then use it to speed up science. The story runs from Hassabis's childhood in North London through the sale to Google, AlphaGo's 2016 match against Lee Sedol, the AlphaFold protein work that earned him a share of the 2024 Nobel Prize in Chemistry, and the race that followed ChatGPT. It draws on long interviews with Hassabis and with colleagues, rivals and critics, and weighs what he has built against the risks of building it.
Below are the 9 ideas HourLife readers vote up most, each in our words with a note on what it asks of you, and then the steps that put them to work.
Chapter 02
What readers keep
The takeaways
-
1
HourLife takeaway
For Hassabis, games were never a detour from science. They were the training ground.
Chess as a boy, designing video games as a teenager, then Atari, Go and chess again for DeepMind's programs. Mallaby shows one thread running through all of it: a game gives you clear rules and a clear score, which is exactly what a learning system needs to practise on.
-
2
HourLife takeaway
A two-step mission kept the lab pointed at the far goal: solve intelligence, then use it on everything else.
DeepMind's founding aim was unusually long-range for a start-up. Mallaby uses it to explain choices that looked odd at the time, such as spending years on games with no product in sight. The far goal set the order of the near steps.
-
3
HourLife takeaway
Studying the brain was Hassabis's way of studying how to build a mind.
Between his games company and DeepMind, Hassabis did a PhD in cognitive neuroscience at University College London, working on memory and imagination. The book treats this as groundwork, not a career change: he wanted to know how the one working example of general intelligence does it.
-
4
HourLife takeaway
Sometimes the move that looks like a mistake is the one that changes the game.
In the second game of the 2016 match in Seoul, AlphaGo played a move that strong players first read as an error. It turned out to be part of a winning plan. Mallaby uses the match to show the moment many people saw that a machine could find ideas humans had missed.
-
5
HourLife takeaway
Pick problems where you can keep score, and progress becomes something you can see.
Protein folding had a huge space of possible shapes, a clear measure of how close a prediction is, and a public blind test, the CASP competition, that showed who was ahead. That made it a natural next target after Go, and AlphaFold's results there led to the 2024 Nobel Prize in Chemistry.
-
6
HourLife takeaway
Joining a giant buys you computing power, and it costs you some say over your own future.
Google bought DeepMind in 2014, and the money and computers that came with it paid for the work that followed. Mallaby traces how the lab tried to keep its independence and its safeguards inside a large company, and how that became harder as the stakes rose.
-
7
HourLife takeaway
A rival's launch can rewrite a patient plan almost overnight.
When ChatGPT came out in late 2022, a lab known for careful research found itself behind in the product race. The book follows the scramble that followed, including Google merging its two AI groups into Google DeepMind under Hassabis in 2023 and the push to ship Gemini.
-
8
HourLife takeaway
Caution is easy to promise before the race starts, and hard to keep once it has.
Mallaby describes people around DeepMind disagreeing about how fast to go, and how competitive pressure moved safety down the list across the industry. He presents the debate from several sides rather than settling it, which leaves the reader with the question of who should decide.
-
9
HourLife takeaway
The drive that builds something extraordinary usually comes with a price.
Mallaby writes about Hassabis with admiration and with a critic's eye. The perfectionism and intensity that took him from chess tournaments to a Nobel Prize also shaped how he ran a lab and treated rivals. The portrait is of a person, not a legend.
Chapter 03
Put it to work
Your action book
Do the setup steps once, then keep the habits going. The setup changes your week; the habits keep it changed after the first good one.
The download puts this whole reading and a printable action book in one PDF: the setup checklist, then a page of real calendar days for each habit, starting on the day you choose.
Set up once
Run the three tests on one problem
Pick one stubborn problem at work. Ask whether it has a huge number of possible answers, a clear score for a good one, and enough data or a way to test answers cheaply. Use the checker on this page and note which test fails.
Write your own two-step mission
On one card, write the far goal you care about, then the one capability you would need to build first to reach it. Keep the card where you plan your week.
Turn a skill into a game
Choose a skill you want to grow and give it rules and a score. For example: ten minutes of drafting, scored by how many sentences survive the next day's edit.
Decide the score before you start
Before your next project begins, write down the one number that would tell you it is working, and when you will check it. Do this before anyone has a stake in the answer.
Look up a protein in the AlphaFold database
Search the free AlphaFold Protein Structure Database for a protein you have heard of, such as insulin or haemoglobin. Notice how each part of the prediction is shaded by how confident the model is.
Keep it up
Keep a surprise log
When a tool, a colleague or your own work does something you did not expect, write it down. A week later, decide whether it was a mistake or a better idea you had missed.
Read AI news with the race in mind
Each time you read about a new AI release, ask two questions: who is this company racing, and what might have been rushed or skipped to ship it first?
Check your scoreboard every Friday
Spend five minutes looking at the number you chose for your main project. Write one line on what moved it and one on what you will try next week.
Chapter 04
The last word
Before you go
Pick one step from the action book and do it today. An idea from a book starts to count once it changes what one ordinary day looks like.
Learn the game first. Then aim the player at something bigger than the board.
That is the end of our reading. Where to go from here, or go back to the full The Infinity Machine: Demis Hassabis, DeepMind, and the Quest for Superintelligence page.
1 / 4 · 5 min left
