Demis Hassabis on the Frontier of Intelligence: What AGI Could Unlock and What Could Go Wrong

In a wide-ranging SXSW London conversation with Bloomberg’s Francine Lacqua, Demis Hassabis, co-founder and CEO of Google DeepMind, offered a candid, often philosophical view of where AI stands today and what lies just ahead. From protein folding to personal agents, Hassabis argued that AI isn’t just another platform shift. It’s a force of nature that could usher in a new scientific era or, if mishandled, one of unchecked disruption.

AGI in 5 to 10 Years? It’s No Longer Unthinkable

While the exact definition of artificial general intelligence (AGI) varies, Hassabis believes we are just 5–10 years away from systems that rival or surpass human general reasoning. He described this threshold as more than just technical. “It’ll be like another industrial revolution,” he said. “Maybe bigger.”

But current models still fall short. They’re creative but inconsistent, capable of generating new strategies (like AlphaGo did) but not yet able to invent entirely new frameworks. “We’re missing coherence. Consistency. The kind of abstract thinking humans do automatically.”

The Endgame: Radical Abundance, If We Get It Right

Hassabis’ long-standing mission has been to “solve intelligence, then use it to solve everything else.” That vision, once derided as “career suicide” in academic circles, now underpins many of DeepMind’s highest-profile breakthroughs.

His long-term hope is that AGI will unlock “root node problems” in science breakthroughs that act as multipliers across disciplines. For example:

• Protein structure prediction: AlphaFold has already become a standard research tool, used by over 2.5M researchers.

• Drug discovery: Isomorphic Labs is working to compress years of R&D into weeks.

• Climate: AI is helping model weather systems, optimize energy use in data centers, and even accelerate materials science (e.g., battery tech, superconductors, fusion containment).

“Imagine curing all diseases,” Hassabis said. “Imagine free, clean energy. That’s the upside we’re playing for.”

From Sci-Fi to Infrastructure: AI as Electricity

While AGI still lies ahead, AI’s current impact is already broad and deep. Hassabis compared it to electricity or the internet general-purpose, ubiquitous, and foundational. “It’s an application layer across everything,” he said.

His personal obsession? A next-gen AI assistant that handles scheduling, triages email, protects your attention, and genuinely understands context. “I’d pay thousands to never look at my inbox again.”

The Risks: Hype Now, Underestimation Later

Despite today’s buzz, Hassabis warned that AI’s short-term power is often overstated, while its long-term impact remains profoundly misunderstood. “It’s overhyped in the short term. But still underestimated in the long term.”

He remains wary of the “move fast and break things” mindset. While that ethos worked for web apps, it’s dangerous when building systems that could impact global infrastructure, healthcare, or democracy.

To mitigate risks, Hassabis advocates for:

• Scientific grounding: Understanding how models learn—not just what they produce.

• Adaptive regulation: Rules that evolve as the technology does.

• International cooperation: Cross-border collaboration on safety and standards.

But he acknowledged that geopolitics might complicate these ideals, especially as the U.S.–China AI race accelerates.

On Creativity, Values, and the Human Edge

Hassabis also addressed the elephant in the auditorium: will AI ever be truly creative or emotionally resonant?

Tools like DeepMind’s Veo (video generation) are impressive, but still lack soul. “They can’t yet tell moving stories,” he said. “There’s still a missing human component.”

That gap, he believes, is both a limitation and an opportunity. The best future lies in tools that amplify, not replace, human expression. “We want to work with creatives, not around them.”

Key Takeaways

1. AGI is closer than we think, likely within a decade, but major capabilities like creativity and consistency are still missing.

2. AI’s biggest potential lies in solving foundational problems in science, medicine, and energy not just automation.

3. DeepMind’s AlphaFold and Isomorphic Labs show what’s possible when AI and science merge at “digital speed.”

4. AI must be deployed carefully: scientific foresight, thoughtful regulation, and international collaboration are critical.

5. In creativity and values, human contribution still matters deeply especially for storytelling, emotion, and purpose.

6. AI today is overhyped in the short term but still vastly underestimated in the long term.

Independently Created. Not affiliated with SXSW.

Unofficially SXSW is a ClickZ Media publication in the Events division