The Infinity Machine: Demis Hassabis, DeepMind, and the Quest for Superintelligence

Summary:

The Infinity Machine explores the life of the founders of DeepMind, especially Demis Hassabis, who has had a near lifelong obsession with developing Artificial General Intelligence (AGI). It also chronicles the rapid development of AI models over the course of the last decade, up to early 2026.

Review:

This was a fascinating book. It is part biography, part technological history, and part educational resource on AI development. The narrative is well done and engaging. I finished it in a single day.

The book opens with Hassabis’s background. He was a child prodigy in chess and uber competitive. He grew up with supportive parents who drove him to be his best, though early in his chess career his father pressured him a lot by yelling at him to do better. Hassabis recounted a memory of telling his father that he was obviously doing everything he could to win every match, so his feedback wasn’t helping. After that, his father never yelled at him again.

He went to Cambridge for his undergraduate studies in Computer Science, but before that he worked for Bullfrog Productions, which eventually led to him founding Elixir Studios, and helped code bleeding edge games such as Theme Park (1994) and Black & White (2001). It was during this time that he was gripped by artificial intelligence.

He eventually began exploring neuroscience as a field to help him understand intelligence and how the human brain derives thought, because he believed it would be instrumental for AI development if we ever wanted to reach true AGI ¹. He said that his mission was actually to solve intelligence first and by doing so, he’d solve “everything else” because understanding how to create intelligence (or allow a machine to have it) would then lead to superhuman intelligence that solves all other problems.

After getting a PhD in cognitive neuroscience, he went on to found DeepMind with Shane Legg and Mustafa Suleyman². The mission of DeepMind was to build the world’s first AGI for the good of humanity. The book goes deep into the fundraising efforts that were necessary to start the company and the many challenges they faced.

When DeepMind was founded in 2010, the idea of true artificial intelligence was a pipe dream of science fiction³. Fortunately, Hassabis is a dynamic preacher of his beliefs and has the uncanny ability to sell anyone on his vision. This was vital for the success of most of DeepMind’s history, up to the point of a huge acquisition by Google.

The early days of AI development at DeepMind were focused on building artificial intelligence that could play and beat Atari games at a superhuman level using only the pixels on screen as input. In 2013 they did exactly that. The goal then shifted to develop AI that could beat the best human in the world at the ancient game of Go. In 2016, DeepMind changed the course of history by beating Lee Sedol with their program Alphago. Not only did the AI beat the world’s best player at possibly the most complex human board game ever created, it did so by using moves that no human has ever been recorded to use.

In 2014 Google acquired DeepMind for $650M dollars and retained Hassabis as the leader of the company. This shifted the dream of developing AGI from being a long stretch into high likelihood due to Google’s near infinite resources. DeepMind continued developing AI and sought to do so in a safe manner, but eventually other players entered the field.

The latter fourth of the book deals with the rise of companies like OpenAI, Anthropic, and DeepSeek (among many others) in the late 20-teens and 2020’s. The release of OpenAI’s chat GPT in 2023 was the first time that an AI model was made available for the general public and that was another pivotal moment in human history (according to the book). Eventually Google released Gemini via DeepMind, but they were actually behind the ball on bringing a product to market.

I found it very interesting that a theme of the book is the requirement for AI safety and the discussions of how AGI could lead to very bad consequences for mankind if things aren’t handled appropriately. Essentially, the top AI thought leaders all tend to agree that AGI in the hands of evil people could lead to WW3. Further, AGI without proper guardrails could, in theory develop goals of its own and decide that humanity is a problem to be resolved (enter every sci-fi story you’ve heard of) by whatever means necessary.

The book briefly deals with a scenario in which AGI could make this determination, develop a secret long-ramping virus, release it into the world, and succeed due to the fact that all of mankind would be infected before we even knew the virus existed.

As scary as that thought is, they follow it up with the opposing views that say that AGI is a machine and won’t ever turn bad because it literally can’t. Machines don’t have feelings. Your toaster has just as much chance as AGI of turning on you. Even if it did, it’s a machine that can be turned off. Cut off it’s access to electricity and you win.

All that to be said, the importance of AI safety cannot be understated and even today, Hassabis and others are working towards the development of a worldwide pact on what this should look like.

There’s a lot more that could be said about the book, but I’ve tried to cover as many of the highlights as possible. If you’re at all interested in AI or technology, this is well worth a read.

Footnotes

1. AGI or Artificial General Intelligence is the idea of autonomous AI that can reason on its own without the need of human prompting or intervention. Hassabis and many others believe AGI will unlock a utopia for humanity if done correctly. ↩

2. The book goes into their backgrounds as well, but needless to say, they also have very impressive CV’s. ↩

3. Side Note: Hassabis is a huge science fiction fan and was compared to Ender from Ender’s Game multiple times throughout the narrative. ↩