Promoting achievement in science
By proving that it can navigate the giant search space of the Go board, AlphaGo has demonstrated the power of AI to help us better understand the complexities of the world. We started by trying to solve the protein folding problem, a 50-year-old major challenge of predicting the 3D structure of proteins – information that is important for understanding disease and developing new drugs.
In 2020, we finally solved this long-standing scientific problem with our AlphaFold 2 program. From there, we folded the structures of all 200 million proteins known to science and made them freely available to scientists in an open database. Today, more than 3 million researchers worldwide use the AlphaFold database to accelerate their critical work on everything from malaria vaccines to plastic-eating enzymes. And in 2024, it was the honor of a lifetime for me and John Jumper to be awarded the Nobel Prize in Chemistry for leading this project, on behalf of the entire AlphaFold team.
Since AlphaGo won, we have applied its basic method to many other areas of science and mathematics, including:
Mathematical thinking: The most direct descendant of AlphaGo's architecture, AlphaProof learned to prove formal mathematical statements using a combination of language models and the reinforcement learning and search of AlphaZero. Alongside AlphaGeometry 2, it became the first program to receive a standard-medal (silver) at the International Mathematical Olympiad (IMO), proving that AlphaGo's methods can open up advanced mathematical reasoning and lay the foundation for our efficient general models.
Gemini, our largest and most capable model, has recently moved forward. An improved version of its Deep Think mode achieved gold medal-level performance at the 2025 IMO using a method inspired by AlphaGo. Since then, Deep Think has been applied to increasingly complex, open-ended challenges throughout science and engineering.
Discovery algorithm: As AlphaGo sought out the best moves in the game, our coding agent AlphaEvolve explores the computer code space to find the most efficient algorithms. It had its Move 37 moment when it discovered a new way to multiply matrices, the basic mathematical operation that powers almost all modern neural networks. AlphaEvolve is now being tested on problems ranging from data center development to quantum computing.
Scientific collaboration: We are combining the search and reasoning principles we pioneered with AlphaGo into a collaborative AI scientist. By having ambassadors 'debate' scientific ideas and hypotheses, the program serves as a participant who is able to do the critical thinking required to identify patterns in data and solve complex problems. In a validation study at Imperial College London, it analyzed decades of literature and independently arrived at a common hypothesis about antimicrobial resistance that researchers had spent years developing and experimentally validating.
We have also used AI to better understand the genome, advanced fusion energy research, improved weather forecasting and more.
Although our scientific models are amazing, they are very special. In order to achieve important achievements like creating unlimited clean energy or solving diseases that we don't understand today, we need general AI systems that can find the basic structure and connections between different subjects, and help us come up with new ideas like leading scientists.
The future of intelligence
For AI to be truly general, it needs to understand the virtual world. We built Gemini to be multimodal from the ground up to understand not only language, but also audio, video, images and code to build a model of the world.
To think and think in all these ways, the latest models of Gemini use some of the techniques we have opened with AlphaGo and AlphaZero.
The next generation of AI systems will also need to be able to summon specialized tools. For example, if a model needs to know the structure of a protein it can use AlphaFold for that.
We think that the combination of Gemini's world models, AlphaGo's search and programming, and the special use of AI tools will prove to be very important for AGI.
True intelligence is a key capability that such an AGI system would need to demonstrate. Move 37 was a vision of AI's ability to think outside the box, but a true innovation will require something more. It won't necessarily come up with a novel strategy for Go, as AlphaGo did so admirably, but it will actually create a game as deep and beautiful, and as worthy of study as Go.
Ten years after AlphaGo's victory, our ultimate goal is at hand. The spark of creativity first seen in the advanced Move 37 has now coalesced to pave the way towards AGI – and usher in a new era of discovery science.
