Descriptions:
Demis Hassabis, CEO of Google DeepMind, sits down with Harry Stebbings on 20VC for a wide-ranging conversation on AGI timelines, compute bottlenecks, scaling laws, and the future of scientific discovery. Hassabis defines AGI as a system exhibiting the full range of human cognitive capabilities—the only known proof that general intelligence is possible—and estimates a strong probability of achieving it within five years, consistent with extrapolations he and co-founder Shane Legg made when founding DeepMind in 2010.
On scaling laws, Hassabis pushes back against plateau narratives, arguing that while exponential gains have moderated, compute expansion still yields substantial returns. He also frames compute as an essential experimental workbench for testing new algorithmic ideas at scale, predicting that labs with strong algorithmic invention capacity will pull ahead as the current generation of techniques matures. The conversation covers Isomorphic Labs—spun out from DeepMind following AlphaFold—with Hassabis projecting a complete AI-driven drug design engine within 5–10 years and outlining a two-stage path to accelerating clinical trials through patient stratification and metabolic simulation.
The interview closes on AI safety, where Hassabis echoes Stephen Hawking’s warning that humanity may not get a second chance to get it right, citing dual-use risks as one of his two core concerns alongside systemic misuse by bad actors. The episode is essential viewing for anyone tracking how DeepMind’s leadership thinks about the near-term trajectory of frontier AI.
📺 Source: 20VC with Harry Stebbings · Published April 07, 2026
🏷️ Format: Interview
