Nov 2020, model, DeepMind

AlphaFold 2 at CASP14

DeepMind's AlphaFold 2 predicted protein structures at close to experimental accuracy in the CASP14 blind assessment, the result Demis Hassabis cites as the first proof that AI can advance science.

What it was

CASP (Critical Assessment of protein Structure Prediction) is a blind test, run every two years since 1994, in which groups predict the 3D shapes of proteins whose structures have been determined experimentally but not yet released. At CASP14, whose results were announced on November 30, 2020, DeepMind's AlphaFold 2 reached a median score of 92.4 GDT across all targets, an average error of about 1.6 angstroms, roughly the width of an atom. John Jumper led the work, with Demis Hassabis as DeepMind's chief executive.

CASP co-founder John Moult said in DeepMind's announcement that the field had been "stuck on this one problem" of how proteins fold "for nearly 50 years." The method was published in Nature on July 15, 2021, with the code released alongside it.

Read the original

What it changed

On July 22, 2021, DeepMind and the European Bioinformatics Institute (EMBL-EBI) launched the AlphaFold Protein Structure Database with about 350,000 predicted structures, including the entire human proteome, free to use. The same day Nature published the human-proteome predictions. By the time of the 2024 Nobel announcement, AlphaFold 2 had been used to predict the structures of "virtually all the 200 million proteins that researchers have identified." Hassabis said at the 2021 launch, "We believe this represents the most significant contribution AI has made to advancing scientific knowledge to date."

The Royal Swedish Academy of Sciences awarded half of the 2024 Nobel Prize in Chemistry to Hassabis and Jumper for protein structure prediction, and the other half to David Baker for computational protein design. Its announcement said AlphaFold 2 had been used by more than two million people from 190 countries.

The arguments it moved

Who should have access to powerful models?

AlphaFold 2 was released openly, and its successor was not, which put DeepMind on both sides of the open-science argument. When AlphaFold 2 was published, "the full underlying code was made accessible to all researchers," Nature's editors wrote on May 22, 2024. AlphaFold 3, published in Nature that month, came with pseudocode and a web server with daily limits and restrictions on drug-development use, and DeepMind had partnered with its sister company Isomorphic Labs. Nature's editorial acknowledged "criticism, of both the AlphaFold team at Google DeepMind in London and Nature." DeepMind published AlphaFold 3's inference code on November 11, 2024, with model weights available on request for non-commercial use.

Compare every leader's position on this

Who benefits from AI's economic gains?

Hassabis's case that AI's gains can be broadly shared leans on AlphaFold. Its database was free and, by the Nobel committee's count, used in 190 countries. He told The Guardian in August 2025 that if AI is stewarded safely "we should be in a world of what I sometimes call radical abundance." The commercial side runs through Isomorphic Labs, the Alphabet-owned drug-discovery company Hassabis also leads; Nature's May 2024 editorial noted that AlphaFold 3 came with restrictions on use "in drug development."

Compare every leader's position on this

Positions it bears on

Demis Hassabis, AI for science

Sees AI as the most powerful tool yet for advancing knowledge, with AlphaFold as the first proof point and root-node problems as the targets.

AlphaFold 2 is the first proof point in his case that AI is the "ultimate tool to advance the frontier of knowledge."

Critics and counterpoints
Demis Hassabis, Radical abundance

Argues that a safely managed transition to AGI could end scarcity, cure disease and make the economy no longer zero-sum, while conceding that distribution is a political problem.

The August 2025 Guardian interview in which he described radical abundance presented AlphaFold as the proof of the benefits he promises.

Critics and counterpoints

Sources

Last verified 2026-09-21.