Inside Google DeepMind's Quest to Solve Science with AI
When Demis Hassabis accepted the Nobel Prize in Chemistry in December 2024, it marked something unprecedented: an AI researcher winning science's highest honor for a tool that predicts protein structures. AlphaFold had solved a problem that biologists had struggled with for fifty years. But for Hassabis and Google DeepMind, it was just the beginning.
DeepMind's journey started with games. AlphaGo defeated the world champion at Go in 2016: a feat many experts said was a decade away. AlphaZero followed, mastering chess, Go, and shogi from scratch in hours. These weren't just stunts. They were proof that AI could discover strategies no human had conceived.
The pivot to science was Hassabis's master plan all along. Games were training grounds. The real target was always nature itself.
AlphaFold: The Breakthrough That Shook Biology
Proteins are the machinery of life. Their function depends on their three-dimensional shape. But predicting shape from the amino acid sequence alone: the "protein folding problem": had defied the best efforts of structural biologists for decades.
AlphaFold 2, released in 2020, predicted protein structures with accuracy rivaling experimental methods. AlphaFold 3 extended this to protein complexes, DNA, RNA, and drug-like molecules. The database now contains predicted structures for over 200 million proteins: essentially every protein known to science.
Please enable JavaScript to read the full article.