DeepMind London serves as the headquarters of Google DeepMind, a British-American artificial intelligence research laboratory and subsidiary of Alphabet Inc. Founded in the United Kingdom in 2010, the company was acquired by Google in 2014 and merged with Google Brain in April 2023 to form Google DeepMind. The London office remains the primary hub for its research activities, with additional centres in the United States, Canada, France, Germany, and Switzerland.
The laboratory is renowned for pioneering work in machine learning, particularly deep learning and reinforcement learning. Its early algorithms, such as the neural networks used for playing Atari video games, demonstrated a general-purpose approach that could learn from raw pixels. This philosophy culminated in high-profile achievements like AlphaGo, which defeated a world champion in the board game Go in 2016, and AlphaFold, which made significant advances in protein folding prediction in 2020. In the 2020s, Google DeepMind shifted focus toward generative AI, developing the Gemini and Gemma large language model families.
History
DeepMind was founded in November 2010 by Demis Hassabis, Shane Legg, and Mustafa Suleyman. Hassabis and Legg first met at the Gatsby Computational Neuroscience Unit at University College London. The founders aimed to create a general-purpose AI that could be useful for almost anything. Initial work involved teaching AI to play Atari games like Breakout, Pong, and Space Invaders, without any programmed knowledge of how to play.
Early investors included venture capital firms Horizons Ventures and Founders Fund, as well as entrepreneurs Scott Banister, Peter Thiel, and Elon Musk. Jaan Tallinn was an early investor and adviser. In January 2014, Google confirmed its acquisition of DeepMind for a reported $400 million to $650 million. The company was renamed Google DeepMind and later became a subsidiary of Alphabet Inc.
DeepMind also engaged in societal projects, such as a pilot program with the UK's National Health Service (NHS) to modernize information systems. In September 2015, it signed an information sharing agreement with the Royal Free NHS Trust to co-develop a clinical task management app called Streams. The company established an AI ethics board and a unit called DeepMind Ethics and Society, with philosopher Nick Bostrom as an adviser.
In December 2019, co-founder Mustafa Suleyman left DeepMind to join Google in a policy role. In April 2023, DeepMind merged with Google AI's Google Brain division to form Google DeepMind, a move seen as a response to the public release of ChatGPT by OpenAI. In the summer of 2026, DeepMind moved into Google's Platform 37 building in King's Cross Central, London.
Research and Achievements
DeepMind has published over a thousand papers, including thirteen in Nature or Science as of 2020. Its research spans multiple domains, from game-playing to scientific discovery. In 2014, it introduced neural Turing machines, which are neural networks that can access external memory like a conventional Turing machine. This concept influenced later work on memory-augmented networks.
The company's game-playing AI includes AlphaGo, which beat European champion Fan Hui in 2015 and world champion Lee Sedol in 2016. A more general program, AlphaZero, mastered Go, chess, and shogi through self-play. Subsequent models like MuZero and AlphaStar extended this to other games, and in 2020, Agent57 surpassed human-level performance on all 57 Atari 2600 games. In July 2022, DeepMind announced DeepNash, a system that plays the board game Stratego at expert level.
In mathematics and algorithm discovery, DeepMind developed AlphaGeometry, FunSearch, AlphaEvolve, AlphaDev, and AlphaTensor. These systems have contributed to solving mathematical problems and discovering more efficient algorithms, such as for matrix multiplication.
AlphaFold and Scientific Impact
In 2020, DeepMind achieved a major breakthrough in protein folding with AlphaFold, which set state-of-the-art records on benchmark tests for protein structure prediction. By July 2022, over 200 million predicted protein structures, representing virtually all known proteins, were released on the AlphaFold database. This has had significant implications for biology and medicine, aiding researchers in understanding diseases and developing treatments.
Large Language Models and Generative AI
Following the merger with Google Brain, Google DeepMind became responsible for developing two large language model families: the proprietary Gemini and the open-weight Gemma. These models are used across Google's products and services. The company also developed other generative AI models, including Imagen for text-to-image, Veo for text-to-video, and Lyria for text-to-music. These efforts position Google DeepMind as a key player in the competitive landscape of AI research, alongside organizations like Anthropic and OpenAI.
Ethics and Governance
DeepMind has been attentive to the ethical implications of AI. It established an ethics board after Google's acquisition, though its membership was kept confidential. The DeepMind Ethics and Society unit was created to study ethical and societal questions, with prominent figures like Nick Bostrom as advisers. In October 2017, a dedicated research team was launched to investigate AI ethics. These initiatives reflect the company's commitment to responsible AI development, a topic of growing importance in the field.