The Quest to ‘Solve All Diseases’ with AI: Isomorphic Labs’ Max Jaderberg

Isomorphic Labs’ Max Jaderberg argues drug design needs six AlphaFold-scale breakthroughs and generative agents to navigate 10^60 possible molecules.

• Drug-like molecular space is ~10^60; even screening 1 billion molecules (10^9) leaves 10^51 unexplored — generative agents, not just predictive models, are required.
• Isomorphic needs roughly six AlphaFold-equivalent breakthroughs to build a complete drug design engine; AlphaFold is just one piece.
• Isomorphic’s internal generative models are already producing molecule designs that human chemists doubt, but lab tests confirm the model is right and the human is wrong.
• AlphaFold 3’s diffusion-based architecture predicts atomic coordinates for proteins, small molecules, DNA, and RNA together — enabling structure-based design that previously required six-month lab crystallization.
• Isomorphic focuses on oncology and immunology; Novartis brought targets the field has worked on for 10+ years; the partnership expanded after year one.
• The GPT-3 moment in AI biology will look more like AlphaGo’s Move 37 — outputs that are correct but beyond human interpretation, not just human-quality.
• 60–80% of Isomorphic’s ML team has no prior chemistry or biology background; deliberate naivety plus curiosity is treated as an asset for first-principles breakthroughs.
• Demis Hassabis predicted dozens of Nature papers and a Nobel Prize during a toast for DeepMind’s first Nature paper (Atari DQN) — roughly 10 years before it happened.

2025-04-29 · Watch on YouTube