The future of intelligence | Demis Hassabis (Co-founder and CEO of DeepMind)

Quick Overview

Demis Hassabis believes achieving AGI requires a dual focus, allocating 50% effort to scaling current models and 50% to fundamental innovation, while also emphasizing that accurate world models are crucial for robotics and universal assistance, as current language models lack the necessary understanding of spatial dynamics and physical context.

Key Points: Hassabis divides DeepMind's effort into 50% scaling and 50% innovation, stating both are necessary to reach AGI. The progress of models like Gemini 3 and advancements in world models, especially over the summer, are sources of excitement. AlphaFold served as the major proof point for solving 'root node' problems, and DeepMind is now exploring similar problems in material science, targeting room temperature superconductors and better batteries. DeepMind partners with Commonwealth Fusion to accelerate fusion energy development, aiming to help contain plasma and assist with material design. Current AI exhibits 'jagged intelligences,' performing at PhD level in some areas (like math Olympiads) but making trivial mistakes in others, highlighting a missing consistency that AGI must resolve. World models, like those powering Genie, are essential because language models struggle to capture spatial dynamics and physical context that must be learned from online, sensory experience. Hassabis suggests society needs to prepare for post-AGI economic restructuring, possibly involving new models like direct democracy or Universal Basic Income, as the transformation might occur ten times faster than the Industrial Revolution.

Context: This transcript captures an interview between Professor Hannah Fry and Demis Hassabis, Co-founder and CEO of Google DeepMind, discussing the extraordinary progress in AI over the past year, including the shift towards agentic AI and the release of models like Gemini 3. The discussion moves beyond immediate product launches to explore the scientific and technological questions defining the next phase of AI development, focusing heavily on the path to Artificial General Intelligence (AGI), the role of world models, and the societal implications of such powerful technology.