Vol 12 ★ SPECIAL EDITION: The Man Building God's Search Engine

Here's the thing that makes Hassabis different from every other AI leader on the planet. He wasn't reacting to a trend or riding a wave. He was building the wave... from 1993. 1993. AI couldn't recognise a picture of a cat until 2012. This guy was 19 years ahead of the cat.

He went to Cambridge. Built a video game company in London called Elixir as a way to smuggle AI into a consumer product. Then did a PhD in neuroscience cos he figured if you want to build artificial intelligence, maybe start by understanding the real thing. And only THEN, in 2010, did he found DeepMind.

OpenAI didn't show up until 2015. By which point AI was already impressive and founding a lab was, shall we say, a slightly easier bet. Mallaby didn't mince it; everyone else, relatively speaking, is a bandwagon jumper.

The book opens with a quote about the atomic bomb. Which is a choice. But it's deliberate. Back in the 1940s, a group of the world's most brilliant scientists built a weapon so powerful it could wipe out civilisation... and they did it anyway, because the science was too exciting to resist. Oppenheimer called it "technically sweet." As in, I know this might end us all but I literally cannot stop myself. AI, Mallaby argues, has the same energy. Different century… same compulsion.

Mallaby didn't have to bring up the comparison with any of them, they all did it themselves. Sam Altman volunteers that he shares a birthday with Oppenheimer. Hassabis, standing outside his first DeepMind office in Russell Square, pointed out it was three doors down from where Turing gave his lectures on AI. And around the corner… The exact crossing where a physicist first had the idea that led to the bomb.

Hassabis told Mallaby: "And now we're doing the new version. Isn't that amazing?" To which Mallaby thought... OK I guess so, but doesn't it worry you?

Mallaby told a story that made all out ears prickle They were sitting in a café in a park. It was a normal day; people at the next table chatting about a friend who went to hospital. And Hassabis starts riffing about how at 2 o'clock in the morning, he's at his desk and... well, let me just give you the quote.

"Reality is staring me in the face. It's calling out to me, saying... discover me, understand me. And if I can understand nature better, I'm understanding the creation of some intelligent being. It might be God."

Demis Hassabis, as recounted by Mallaby

I mean…this man isn't building a chatbot. He's trying to decode the universe. He told Mallaby that Einstein, Newton... none of them fully solved the mysteries of reality. And he wants to do better. He's building AI to finish the job they started.

But here's what makes it complicated, because it seems he's also the most competitive person in the room. When ChatGPT launched and stole the spotlight, Mallaby went to see him a month later. "How does it feel?" he asked. Hassabis looked at him and said: "This is war. They've parked their tanks on my lawn."

So we've got a man who speaks about the divine at 2am and declares war by lunchtime. Got it.

The emotional weirdness of building world-changing technology doesn't stop with Hassabis. Mallaby collected some stories that sound like they belong in a film noir, not a tech co. Here’s a couple that I have to share.

Ilya Sutskever, then chief scientist at OpenAI, held a ceremony with colleagues. They wore monk-like cloaks. Gathered around a fire pit, he produced an effigy and told them: "Imagine this is the evil AI. It must be destroyed." And they all watched it burn. More pagan than Christian, but definitely out there. Mallaby checked with people who were present. Yup, it happened.

Shane Legg, Hassabis's co-founder at DeepMind, gave a lecture called "The Halloween Scenario" predicting that AI would start to become genuinely dangerous around... well, around now. Thirty years from when he gave the talk. And as he described the potential annihilation of humanity, he laughed. Because apparently when humans contemplate their own extinction, it feels so absurd that the brain reaches for humour.

Mallaby put it perfectly: "It's partly a book about machine intelligence. It's also a book about human intelligence and how humans deal with it."

Mallaby was honest about his own journey. Going in, he didn't buy the doomsday narrative. His logic was simple. Machines aren't evolved. They don't want to survive. They don't have a motive to fight us. Why would they bother?

Then he went to Toronto and sat in Geoffrey Hinton's kitchen. Hinton is basically the academic godfather of deep learning. And Hinton said this:

Imagine you build a very powerful AI and you're worried an enemy AI might attack it. You're too slow and not clever enough to defend against it yourself. So you give your AI instructions to defend itself. To survive. You've just given it a survival instinct. And once it has that... combined with superintelligence... combined with the capacity to deceive, which labs are already documenting in experiments... how comfortable do you feel now?

Mallaby admitted it made sense. He doesn't wake up every morning in a cold sweat. But he no longer dismisses the risk either.

Mallaby gave the best plain-English explanation of AI I've heard in a while, so I'm stealing it for y'all.

Deep learning is like going to a library. You read everything, absorb the patterns, and now you can answer questions based on what you've absorbed. It's indirect. Crystallised knowledge from other sources. This is mostly what ChatGPT and Claude Chat are doing.

Reinforcement learning is like actually living in the world. You try something, you get feedback, you learn by doing. Drop a glass on concrete... it breaks. Touch a hot pan... ouch. You learn through experience, not books. This is DeepMind's way.

The key to really powerful AI, Mallaby says, is combining both. Library brain plus street smarts. And that's exactly what Hassabis has been building toward for over a decade.

For someone building what might be the most consequential technology in human history, Hassabis sounds very normal. Married his Cambridge girlfriend. Two kids. Same house for 15 years. When a publisher friend offered him unsolicited advice about AI, Hassabis politely told him it was "very useful input." Reader, it was not useful input.

He refused years of pressure from Google and Peter Thiel to move DeepMind to California. He kept the lab in London. Mallaby frames the Google acquisition not as a sellout but as a brilliant hustle... nearly a billion dollars a year in R&D money flowing into Britain, no product requirement - the research stays in London. And the entire British AI ecosystem that has grown around DeepMind, from Cambridge to UCL to Imperial to a constellation of spinout companies, is something Mallaby thinks we don't celebrate nearly enough.

An audience member asked about copyright and the scraping of creative work. Mallaby didn't flinch. He confirmed that the early AI labs straight up pirated hundreds of thousands of books to train their models. One researcher told him they just downloaded everything without asking a boss or a lawyer. At the time it felt like a harmless experiment. Now those models are everywhere, ingesting writers' work and spitting it back to millions of users. That, he said, is an outrage.

He believes, what survives is the deeply human stuff. The in person interviews. Being in the room. The research stays human. Actual writing... that's the part most under threat.

His literary agent is going to court over her authors' scraped work. Someone approached her saying they're writing ten books a day with AI. She told them she couldn't help. She needs human-written material.

Mallaby doesn't hedge. He thinks AI is either the most important moment in world history since the dawn of human thought 70,000 years ago... or at minimum, comparable to the Industrial Revolution. Either way, the current level of excitement about AI, which already feels high, is actually lower than it should be.

The path to AGI? Hassabis is betting on two horses. One is just scaling up what already works... more chips, more data, bigger models. The other is waiting for a new algorithmic breakthrough, some flash of genius that unlocks the next level. He doesn't know which one gets there first. (So he's funding both).

At the end of the talk, we are told that the book ends with Hassabis offers an answer to where he believed all of this is heading. But for that... we will all have to read The Infinity Machine.

✦ ✦ ✦

Pastries consumed. Brain expanded. Slight existential dread acquired.…and pls dont forget to subscribe 🙂

Sxoxo