The same machinery that can fake a voice or forge a photograph also cracked a problem biology had carried for fifty years. In the autumn of 2024, the Nobel committee certified both halves of the technology Geoffrey Hinton helped invent: the counterfeit, and the cure.
The breakthrough, scored
CASP accuracy (GDT, 0–100). A score near 90 is roughly as good as a physical experiment. For decades, the field crawled forward a point or two per cycle. Then it ended.
"Decades" bar is an illustrative midpoint of the 40s–50s range the chapter gives qualitatively; 2018 and 2020 values are from the source dossier.
The gift, at scale
DeepMind published the method and gave the structures away — turning years of lab work into a free, instant lookup.
~200M
protein structures in the open AlphaFold database within two years of launch — very nearly every catalogued protein on Earth.
Used by >1,000,000 researchers across 190+ countries — free of charge.
OCT 8, 2024 · STOCKHOLM
Nobel Prize in Physics
John Hopfield & Geoffrey Hinton
For foundational discoveries that enabled machine learning with neural networks — the Hopfield network and the Boltzmann machine. Hinton used the moment to repeat his warnings.
OCT 9, 2024 · STOCKHOLM
Nobel Prize in Chemistry
Demis Hassabis & John Jumper · with David Baker
For AlphaFold's protein-structure prediction (Hassabis, Jumper) and for computational protein design (Baker). Hassabis argued the opposite case: built carefully, AI would be the most beneficial technology in history.
SOURCE · NEURON MAKERS, CH. 32 — drawn from research dossier 10 (CASP/GDT results, AlphaFold Nature paper July 2021, AlphaFold Protein Structure Database with EMBL-EBI, AlphaFold3 May 2024, Nobel announcements Oct 8–9 2024). ~200M and >1M-researcher figures as reported.