Hassabis Plans Beyond Chatbots: London's Bid to Lead the AGI Era

Demis Hassabis just told everyone to zoom out. While Silicon Valley obsesses over chatbot features and model sizes, the DeepMind CEO sees these as mere stepping stones to artificial general intelligence. In a sweeping interview with Fortune timed to his new biography "The Infinity Machine," Hassabis laid out why London—not Palo Alto—might host humanity's next great leap.

The numbers he throws around are staggering. Hassabis predicts AI's impact will be "10 times bigger than the Industrial Revolution—and maybe 10 times faster," according to The Guardian. This isn't another tech executive breathlessly hyping their product. It's a chess grandmaster (literally) mapping moves five turns ahead while competitors fight over the current piece.

What's refreshing is Hassabis doesn't pretend to have it all figured out. He admits DeepMind's current systems remain "narrow" despite their impressive capabilities. The path from today's chatbots to tomorrow's AGI requires breakthroughs in reasoning, planning, and what researchers call "world models"—systems that understand not just patterns but underlying reality.

Here's where it gets interesting. Most AI founders would've relocated to Silicon Valley years ago. Hassabis did the opposite, doubling down on London even as competitors like OpenAI centralized in the Bay Area. His reasoning cuts against every startup playbook you've read.

London offers something Silicon Valley can't: distance from the hype cycle. When every coffee shop conversation revolves around the latest model release, it's hard to think in decades instead of quarters. The British capital also provides access to world-class universities (UCL, Imperial, Cambridge) without the salary inflation that's made Bay Area hiring a blood sport.

But there's a cultural component too. Hassabis argues European sensibilities around AI safety and ethics create better long-term outcomes. While American companies race to ship first and regulate later, DeepMind can build with guardrails baked in. It's the difference between building a race car and building transportation infrastructure.

Hassabis outlined a concrete four-phase approach he's pitching internally at Google. First, nail the fundamentals—scaling laws, efficiency improvements, and safety frameworks that actually work. Second, expand beyond text to multimodal understanding that can process video, audio, and physical interactions simultaneously.

Third comes what he calls the "reasoning breakthrough"—systems that can plan multi-step solutions like humans do. Current models excel at pattern matching but stumble on tasks requiring sustained logical thought. The final phase involves connecting these capabilities to real-world applications, starting with drug discovery through DeepMind's spinoff Isomorphic Labs.

Each phase builds on the previous, creating what Hassabis describes as a "renaissance" in AI capabilities. The timeline? Deliberately vague, but he suggests we're entering phase two now with Gemini's multimodal improvements.

While competitors chase consumer applications, Hassabis made a strategic bet that sounds almost quaint: curing diseases. Through Isomorphic Labs, DeepMind applies its AI to protein folding and drug discovery—problems with clear success metrics and enormous societal impact.

This isn't corporate charity. Drug discovery offers the perfect test bed for AGI capabilities. Success requires reasoning across multiple domains—chemistry, biology, physics—while handling incomplete information and massive search spaces. If an AI system can design novel treatments for diseases like Alzheimer's, it's demonstrated capabilities far beyond chatbot conversation.

The approach has already yielded results. DeepMind's AlphaFold predicted structures for nearly every protein known to science, creating what researchers call "the Google Maps of biology." Hassabis views this as proof that AGI won't emerge from bigger chatbots, but from systems that can reason about the physical world.

The competition isn't taking this lying down. Multiple sources confirm OpenAI declared "code red" after DeepMind's recent advances, accelerating their own AGI research timeline. The irony isn't lost on industry watchers—OpenAI, founded to prevent AGI concentration, now races to beat Google at its own game.

Hassabis welcomes the pressure. He argues competition drives innovation, but worries about safety shortcuts when companies feel existential threat. His solution? Open research and safety standards that benefit everyone, even competitors. It's classic Hassabis: playing infinite games while others fight finite battles.

The London advantage shows here too. British regulators have signaled willingness to create AGI-specific oversight frameworks, potentially giving DeepMind clearer rules while Silicon Valley companies navigate regulatory uncertainty. Sometimes being first means waiting for the right framework.

For developers watching from the sidelines, Hassabis drops some practical wisdom. Stop optimizing for benchmark scores and start building applications that matter. The next wave won't be about who has the biggest model, but who connects AI capabilities to real human problems most effectively.

His advice echoes what we're seeing in the field: multimodal AI applications are becoming table stakes, while AI agent frameworks that can reason across domains represent the real frontier. The winners won't be prompt engineers, but systems designers who understand both AI capabilities and human needs.

Most importantly, Hassabis suggests building for the transition. Today's applications should be designed with upgrade paths for more capable systems. If he's right about the timeline, we're not decades away from AGI—we're in the decade that creates it.

Here's what might be Hassabis's most contrarian take: Silicon Valley is optimizing for the wrong culture. The move-fast-and-break-things ethos that served social media companies won't work for AGI. When your product might literally reshape human civilization, breaking things isn't acceptable.

This cultural difference shows up everywhere. European researchers tend to publish more safety-focused papers, while American labs push capability boundaries faster. Hassabis argues both approaches are necessary, but need better integration. His London base provides neutral ground where these perspectives can collide productively.

The bet is paying off. DeepMind now attracts researchers who want to work on AGI without the pressure cooker environment of Silicon Valley startups. Sometimes the best way to move fast is to stop constantly looking over your shoulder at competitors.

Hassabis won't give timelines, but the tea leaves suggest major announcements within 18 months. DeepMind's Gemini updates have accelerated, safety research is publishing faster, and Isomorphic Labs is moving toward clinical trials. The pieces are aligning for something significant.

The smart money watches London, not Palo Alto. While American companies compete on model size and chatbot features, Hassabis is building the infrastructure for humanity's next chapter. When AGI arrives, it might speak with a British accent—and that's exactly how he planned it.