The deep mind of Demis Hassabis

Written after hundreds of hours of conversations with the DeepMind founder and his associates over three years, this could well be a definitive account of AI

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

By Sebastian Mallaby 

Published by 

Penguin

464 pages ₹1,299

 Since the release of the first iteration of ChatGPT in late 2022, artificial intelligence (AI) has dominated the conversation. While AI has been responsible for many startling developments, it has also been controversial with debates raging about the dangers of the technology.

 

One AI-driven achievement that is unequivocally beneficial is the work on protein-folding. An AI algorithm, Alphafold, developed a compendium of 260 million protein structures that is publicly available. This has helped accelerate drug-development and the understanding of diseases. The man responsible, Demis Hassabis — and John Jumper, who led his team at DeepMind — received a well-deserved 2024 Nobel Prize for chemistry. 

 

 

The author of this book is a veteran journalist who has covered the relevant beats for decades for The Washington Post and The Economist. He had deep access (if readers will forgive the pun) spending hundreds of hours chatting with Mr Hassabis and his associates over a three-year period. So the book could be described as definitive, if there can be such a thing in AI where things evolve with such rapidity.

 

Sebastian Mallaby covers Mr Hassabis’ life, the creation of DeepMind, and his struggles to keep the company’s (financially unrewarding) research focus, while convincing investors to keep funding it. The book also covers the race between different schools of AI as developed by different companies and researchers. It attempts to explain much of the technology in non-technical language. It maps corporate battles between the world’s mega-billionaires as they jumped on to the AI bandwagon and tried to steal marches on each other.

 

The corporate shenanigans are well documented, along with vignettes that throw light on the tech bro culture and somewhat childish tantrums that are part and parcel of Silicon Valley. The author struggles, however, when it comes to explaining the technology, sometimes moving into near-meaningless verbiage. He also struggles to stay neutral when charting the course of ethical deployment of AI.

 

Mr Mallaby  is also sympathetic to Mustafa Suleyman, a former associate and co-founder of DeepMind and once a close friend of Mr Hassabis, who was a pioneer in the battle to keep AI ethical. Mr Suleyman is also a formidable thinker, whose heart is probably in the right place. But he caused big fissures in DeepMind and may have permanently damaged the AI ethics movement and AI’s applications to health care, with his abrasive style of functioning. 

 

So who is Demis Hassabis? He’s a Londoner through and through who’s refused to migrate to Silicon Valley. He’s the son of immigrants with a Greek-Cypriot father, who was a street musician, and a Singaporean Chinese mother who worked as a nurse. At age five, it became obvious that he was extraordinarily gifted. One of his professors remarked that while most Nobel winners need to be talented and lucky (in picking research focus), it was obvious Mr Hassabis would make it to the top in whatever field he chose.

 

By 11, he was the second-best chess-player of his generation (Judit Polgar was no 1). By 15, he was a multiple world gaming champion and designing some of the most popular games in history. He refused a seven figure salary from Bullfrog (a famous design studio) to study computer science at Cambridge. At 20, he emerged with a double first. By then, he was obsessed with “solving intelligence”. That led to a switch to neuroscience, which he considered crucial for the next stage in developing computer intelligence. His PhD thesis in neuroscience found a critical link between memory and imagination and has been cited thousands of times. After that, he did post-doctoral research at Harvard and the Massachusetts Institute of Technology.

 

When he founded DeepMind, many observers were puzzled by the company and its unusual focus. DeepMind remained London-based. It was a for-profit. Mr Hassabis convinced a few London angels to come up with seed funding, and then Peter Thiel and Elon Musk gave him money. Finally, Google bought the company. Amazingly, Mr Hassabis was allowed to continue running it, and to continue being headquartered in London and focused on whatever he considered interesting in his quest to solve intelligence.

 

In 2016, Mr Hassabis and his team cracked the most mathematically intractable game — Go — creating an algorithm AlphaGo, which beat a reigning world champion (and drove him into retirement within two years). A refinement created the Alphazero algorithm, which taught itself with almost no human assistance. That revolutionised the computer approach to chess, Go, and shogi. The Alpha algorithm was how DeepMind initially attacked protein folding which is orders of magnitude more difficult.

 

This account could not have been written without the level of access

 

Mr Mallaby developed. The book is carefully researched and almost obsessively footnoted, with independent verification and sources for almost every factual statement and every quote.

 

But access of this order can lead to a version of Stockholm Syndrome where the writer develops too much empathy for a subject. Mr Mallaby openly admires Mr Hassabis and that means he is almost uncritical in his assessment of an admittedly extraordinary individual.

---

 The reviewer is a New Delhi-based independent journalist