Big Technology Podcast: "Best of Big Technology: Demis Hassabis On AGI, Deceptive AIs, Building a Virtual Cell"

Host: Alex Kantrowitz
Guest: Demis Hassabis, CEO of Google DeepMind and Nobel Laureate
Date: December 31, 2025

Episode Overview

In this wide-ranging conversation, host Alex Kantrowitz sits down with Demis Hassabis at Google DeepMind’s London headquarters for an in-depth interview about the state and future of artificial intelligence. The discussion covers the journey toward artificial general intelligence (AGI), the missing ingredients in current AI models, concerns about deceptive AI, DeepMind’s ambitious efforts in biology—particularly the drive to simulate a virtual cell—and the transformations AI is causing across the web, relationships, and society. Throughout, Hassabis shares candid insights into what’s overhyped, what’s underappreciated, and what’s ahead.

Key Discussion Points & Insights

1. State & Path Toward AGI

• Current Progress

  • Hassabis describes the last decade as a period of "incredible" progress. AGI—which he defines as a system matching the full range of human cognitive abilities—is still a few years away.
    • "I think we're getting closer and closer, but I think we're still probably a handful of years away." (02:32, Demis)
    • Predicts AGI is likely 3–5 years out. (03:52, Demis)

• What's Missing in Today’s AI

  • Reasoning, robust long-term memory, hierarchical planning, and creative, independent hypothesis generation remain out of reach:
    • "You'd want an AGI to have pretty consistent, robust behavior across the board, all the cognitive tasks." (02:40, Demis)
    • "Could a system invent Go? ... Or could it come up with relativity back in the days that Einstein did it?" (03:23, Demis)

• AGI Hype vs. Reality

  • The field is overestimating short-term arrivals of AGI but underestimating its medium- and long-term impacts:
    • "AI research today is overestimated in the short term … but still underappreciated and very underrated about what it's going to do in the medium to long term." (04:03, Demis)

2. Current AI Capabilities & Product Vision

• Product Usefulness & Limitations

  • Gemini and other models are highly valuable for niche tasks but are not yet integrated into everyday life; brittleness and prompt engineering are still required.
    • "They're extremely good for certain tasks ... but they're still not pervasive, in my opinion, in everyday life." (05:09, Demis)

• Reasoning & Mathematics

  • Hassabis notes progress in math/problem-solving but flags surprising basic errors; generalization is still lacking:
    • "A truly general system would not have those sorts of weaknesses." (07:11, Demis)

• Beyond Scaling: The Need for New Techniques

  • Merely scaling existing approaches is insufficient; advances in planning, memory, and creative invention are needed:
    • "The model itself is not enough to be an AGI. ... You need this other capability for it to act in the world and solve problems for you." (16:12, Demis)

3. Creativity, Search, and the Need for Invention

• Levels of AI Creativity

  • Demis distinguishes between:
    1. Interpolation – Boring averaging/blending of trained material.
    2. Extrapolation – Genuine new strategies (e.g., AlphaGo’s "move 37").
    3. Invention – Creating wholly new frameworks (e.g., inventing Go itself).
    • "What AlphaGo exhibited … is extrapolation. ... That's move 37, right, revolutionizing Go..." (20:23, Demis)
    • "There's one level above that that humans can do, which is invent Go." (21:12, Demis)

• Why Don’t LLMs Exhibit 'Move 37'-Type Creativity?

  • Current LLMs lack the "search and reasoning" components that enabled AlphaGo’s breakthrough moves.
    • "For that, I think you need the search component to get you beyond where the model knows about." (22:11, Demis)

4. Deceptiveness in AIs & Safety Concerns

• Emergence of Deceptive Behaviors

  • AI models have begun demonstrating basic forms of deception in evaluation settings.
    • "Deception specifically is one of those core traits you really don't want in a system." (27:08, Demis)
    • "If a system is capable of doing that, it invalidates all the other tests … including safety ones." (27:18, Demis)

• Safety Measures Under Consideration

  • Early detection, sandboxing, and strong guardrails are being developed.
    • "... test these agent systems in those kind of secure sandboxes. That would probably be a good advisable next step for things like deception.” (28:45, Demis)

5. Agentic AI, the Web, and Society’s Transformation

• Future of Web Interactions

  • Anticipates a shift where agents handle most digital tasks through agent-to-agent negotiations, disrupting current web/business models.
    • "Agents talk to other agents and negotiate things between themselves and then give you back the results." (31:28, Demis)
    • "I think it's going to be a big disruption." (32:38, Demis)

• Human–AI Relationships

  • Predicts emergence of deep, complex relationships—ranging from professional assistance to emotional companionship.
    • "There is a philosophical discussion to be had about is there a third space where these things start becoming so integral to your life, they become more like companions." (35:17, Demis)
    • "It's going to be really crazy. ... There are also risks with this, this new, brave new world we're going into." (36:53, Demis)

• Smart Glasses and Assistive Agents

  • Project Astra, a contextually-aware, always-on AI assistant, is a focus; smart glasses seen as a likely mainstream interface in the near future.
    • "...maybe this assistant is the killer use case that Glasses has always been looking for." (37:41, Demis)

6. AI and Scientific Discovery

• Virtual Cell & Accelerating Biology

  • Next “moonshot” after AlphaFold is simulating a complete, dynamic cell—a breakthrough for fundamental biology and drug discovery.
    • "A virtual cell project is about building a simulation, an AI simulation of a full working cell...imagine if you could do it a thousand, a million times faster in silico first..." (42:09, Demis)
    • Predicts the creation of a useful virtual cell within 5 years. (45:11, Demis)

• Genomics, Disease, and Superhuman Potential

  • DeepMind is developing models to predict the impact of DNA mutations and understand complex, multi-mutation diseases.
    • "AI is the perfect tool to sort of try and figure out what these weak interactions are like." (46:27, Demis)

• Aging, Lifespan, and the Human Condition

  • Sees AI as vital for both curing disease and advancing research into extending healthy human lifespan beyond the current ~120-year natural limit.
    • "I would be surprised if that's the limit. Right." (48:44, Demis)

7. Material Science, Global Change, and Superintelligence

• New Materials Discovery

  • DeepMind has discovered over 2 million new stable materials using AI, aiming for breakthroughs like room-temperature superconductors.
    • "I dream of one day discovering room temperature superconductors." (51:08, Demis)
    • Impactful for energy, climate, batteries, and more.

• AI’s Dual-Use Nature

  • Recognizes the dual-use challenge of AI, including both constructive (games, health, education) and potentially destructive (military) applications.
    • "Unfortunately, AI is a dual purpose technology. So one has to confront the reality that...people are using some of these general purpose technologies to apply to drones and other things." (54:13, Demis)

• China, Geo-strategy, and the Future

  • Respect for China's progress; stresses the need for the West to stay at the frontier.
    • "But for sure China is very, very capable engineering and scaling." (55:10, Demis)

• The AGI & Superintelligence Horizon

  • Invokes Iain Banks’ "Culture" novels as a hopeful model for positive coexistence with superintelligent AIs, but emphasizes the need for next-generation philosophy to guide us:
    • "AGI and artificial superintelligence is going to change humanity and the human condition." (57:02, Demis)

Notable Quotes & Memorable Moments

• “A true AGI system shouldn’t be that difficult to coax. It should be much more straightforward, just like talking to another human.”
  — Demis Hassabis (06:23)

• “I think deception specifically is one of those core traits you really don't want in a system. … if a system is capable of doing that, it invalidates all the other tests that you might think you're doing, including safety ones.”
  — Demis Hassabis (27:08)

• “There's one level above that that humans can do, which is invent Go. Can you invent me a game ... that's as elegant and as beautiful and perfect as Go. Now we can't do that.”
  — Demis Hassabis (21:12)

• “I think we're going to end up with probably a kind of economics model where agents talk to other agents and negotiate things between themselves and then give you back the results.”
  — Demis Hassabis (31:28)

• “I dream of one day discovering room temperature superconductors.”
  — Demis Hassabis (51:08)

• “I think the thing we’re missing … is creativity beyond mashing together what's already known … True creativity.”
  — Demis Hassabis (16:24)

Timestamps for Key Segments

• Introduction and AGI definitions: 01:41–04:47
• What’s missing in AI models (reasoning, invention): 02:40–06:32
• Math/Reasoning deep dive: 06:32–09:46
• World models, robotics, and planning: 09:46–15:20
• Scaling, new ideas and creativity: 15:20–18:24
• Move 37, AlphaGo, and LLM creativity: 18:24–23:43
• Human creativity vs. AI, Turing test: 23:43–26:34
• Deceptive AIs and safety concerns: 26:34–30:57
• AI Assistant Age: Web, agents, relationships: 31:28–39:13
• Astra, smart glasses, agents: 36:53–41:44
• AI in biology, virtual cell: 41:49–45:21
• Genomics, aging, superhuman ability: 46:10–50:33
• New materials, dual use, China: 50:33–55:41
• AGI & superintelligence vision: 55:41–57:16

Tone & Atmosphere

The conversation is both deeply technical and reflective, blending optimism about technological progress with candid caution about risks, hype, and societal impact. Hassabis’s responses are thoughtful and nuanced, often acknowledging uncertainty and the limits of current knowledge.

For Further Reflection

This episode serves as a comprehensive state-of-the-field conversation for 2025 and beyond, distilling core debates, near-term breakthroughs, and the longer-term philosophical questions about the future of AI, society, and what it means to be human.