Demis Hassabis: We're Three Quarters of the Way to AGI

Summary

Demis Hassabis, co-founder and CEO of Google DeepMind, speaks with Sequoia partner Konstantine Buhler about his path to AI, DeepMind's founding, and his belief that AGI is achievable by 2030. He explains how AI for science, AlphaFold, and Isomorphic Labs could compress drug discovery timelines, and how learned simulators like WeatherNext and a virtual cell could unlock new sciences. He also discusses information as a fundamental substance, classical Turing machines versus quantum systems, and the philosophical questions around AI consciousness and agency.

• Demis Hassabis describes his early chess, gaming, and neuroscience background as preparation for founding DeepMind.
• He says DeepMind's 20-year AGI mission is on track and reaffirms 2030 as his AGI timeline.
• AlphaFold is presented as an AI biology breakthrough; Isomorphic Labs aims to automate drug design.
• He argues drug discovery could shrink from 10 years to months, weeks, or days through in silico exploration.
• AI learned simulators, including WeatherNext and a virtual cell, may create new scientific fields.
• He suggests information may be more fundamental than matter or energy.
• Classical Turing machines can approximate some quantum systems, as shown by AlphaFold.
• He advocates building an intelligent tool before addressing agency and consciousness.