The Infinite Machine — A "Chicken Game" Over AI Control and a Survival Guide for the Rest of Us

I just finished the biography of Demis Hassabis, co-founder of DeepMind.

The book opens with a chess competition from Hassabis's childhood — and it's immediately clear he was simply gifted at that kind of strategic contest. He later moved into other board games and even entered the games industry to develop simulation games. Those competitive and game-design instincts, it turns out, laid the direct groundwork for what would become deep learning. People say genius always comes with a touch of madness, and the early Hassabis genuinely carried that young person's burning energy — passionate, rebellious against the system. To pursue his dream of AGI, he went out of his way to earn a PhD in neuroscience, determined to understand the low-level logic of the brain. The moment he had that top-tier academic credential in hand, he stepped out and founded DeepMind.

By mid-book, the story enters the battle between deep learning and classical machine learning. The AlphaGo chapter — AI entering competitive board games — reads like the early phase of a scouting exercise. They challenged the world's top players, iteratively refined the model architecture, and in match after match, the human champions retreated. It's a quietly moving section. When "being the strongest" becomes the goal in itself, and humans — working within every biological limit they have — get effortlessly surpassed by a machine, you can feel the quiet devastation in the retirement of those grandmasters.

I imagine Hassabis, watching it all unfold in person, felt a deep satisfaction. And that satisfaction is part of why he later hesitated when Facebook and Google came with massive acquisition offers — which made early investor Elon Musk anxious enough to team up with Sam Altman and found OpenAI as a counterweight. Because Hassabis had seen AI's destructive potential in a specific domain with his own eyes, and because he genuinely believed in the possibility of AGI, his attitude toward the technology's development became more reverential and cautious — not less.

The interesting twist comes when he turns to medicine and science, tackling protein folding. He encountered the arrogance and inefficiency baked into academic establishments and traditional gatekeepers. The leap was enormous, and it exposed him to fierce skepticism from the traditional academic world. But that friction ended up smoothing out his edges — Hassabis's style became notably more measured afterward.

As machine learning, deep learning, and AGI ambitions kept expanding, the oversight committees that everyone had nominally agreed to put in place gradually turned into a full-blown chicken game.

Markets are not structured competitions. Nobody politely follows the rules. Rather than trusting that your opponent would develop AI responsibly, it made more strategic sense to keep AI development in your own hands. And so, one by one, the names we now know — Google, Musk, Sam Altman — entered the arena.

When everyone expects someone else might launch something new at any moment, you can't afford to be too late. You have to strike a balance between first-mover advantage and stability. That tension is what gave us ChatGPT arriving in the public eye — and the word "emergence" suddenly exploding everywhere.

For Hassabis, though, these AIs are absolutely not just predicting the next character. They operate at a deeper level: understanding the semantics behind language, and predicting the next meaning from that. Maybe at its core it's still statistics — but it means machines can now statistically model something that was previously invisible beneath human language.

It reminds me of what we often call Meta-learning: teaching machines how to learn to learn. And honestly, how similar is that to how humans develop? A baby makes a random sound; the adult responds with expression and tone; the baby learns whether that sound was correct. That's a method humans refined through experience, one tuned to our DNA. Machine learning today is trying to find an equivalently effective learning method — and once found, we may have actually touched the underlying logic of how humans learn.

Back to the first-mover vs. stability tension. Google, as an incumbent giant, along with Hassabis (at least as the book portrays him) genuinely wanting responsible AI, lost the pole position in this race — and was relentlessly written off by commentators at the time.

Looking back from 2026, even when Google releases new models, they're inevitably compared to OpenAI's products. And anyone who's watched Sam Altman present knows his ability to move a crowd is extraordinary. The same tools, launched by him, always land with more pull on the public.

But through all of this, what we're witnessing is a sequence of paradigm shifts.

First DeepMind demonstrated AI's precursor through disruptive innovation; Google used its standard capital playbook to acquire those brilliant minds; other funds saw Google move and piled in; then China used a new technique — knowledge distillation — to claim another seat at the table.

Where we stand today, AI still looks like the Warring States period: powers rising and falling, and we are genuinely still in the middle of the development arc.

For ordinary people like us, there's no need to play the role of those chess grandmasters — rigidly defending the limits of human cognition until AI relentlessly understands, and surpasses, each one. What we should be doing now is learning to use AI, and finding — together with AI — a new equilibrium in our lives and work.