Google DeepMind | History, Innovations, & Controversies - Encyclopedia Britannica

Please select which sections you would like to print:

verifiedCite

While every effort has been made to follow citation style rules, there may be some discrepancies.

Please refer to the appropriate style manual or other sources if you have any questions.

Google DeepMind is an artificial intelligence (AI) research laboratory created through the merger of Google AI’s Google Brain division and the tech start-up DeepMind. As a subsidiary of Alphabet, it focuses on advancing artificial general intelligence (AGI) and developing AI systems for use across Google’s products. The organization is headquartered in London and operates internationally.

Who founded DeepMind?

DeepMind was founded in 2010 by Demis Hassabis, Mustafa Suleyman, and Shane Legg. The founders brought together expertise in neuroscience, computer science, and machine learning with the goal of advancing artificial intelligence. Their shared ambition was to create systems capable of general learning and reasoning, rather than narrowly specialized software designed for specific tasks.

What were DeepMind’s early breakthroughs in AI?

DeepMind’s early breakthroughs in artificial intelligence (AI) included its use of deep reinforcement learning, a method that combines neural networks with trial-and-error learning. Using this approach, the company developed a system capable of learning to play about 50 different Atari 2600 games directly from raw screen input, without prior knowledge of the rules. In 2016 DeepMind introduced another major advance with WaveNet, a neural network capable of generating realistic human speech.

What are some applications of Google DeepMind’s AI systems?

Google DeepMind’s artificial intelligence (AI) systems have been applied across a wide range of fields, including genetics, computer programming, and weather prediction. Notable examples include AlphaFold, which predicts protein structures and has aided biological research, and Aeneas, which assists historians in analyzing ancient inscriptions.

What ethical concerns are associated with AI research at Google DeepMind?

Ethical concerns surrounding Google DeepMind’s artificial intelligence (AI) research include the broader societal impact of advanced AI, particularly the development of highly autonomous systems and how they should be governed. Additional issues have involved the use of personal data in partnerships such as those with health care providers. In response, the organization has supported initiatives focused on AI safety, ethics, and responsible development.

DeepMind was founded in 2010 by Demis Hassabis, Mustafa Suleyman, and Shane Legg to “solve intelligence” by creating artificial general intelligence (AGI). Its approach centered on using machine learning to train neural networks to learn and make decisions in ways that resemble human cognition, rather than programming actions and reactions individually or specializing software for specific tasks (such as playing chess or modeling climate systems).

In its early days the company employed about 100 leading scientists in machine learning and computational neuroscience from dozens of countries to work on these problems. It was acquired by Google in 2014 for a reported total between $400 million and $650 million.

Google Brain was launched in 2011 at X, the Moonshot Factory, Google’s division for experimental technologies, including early work on self-driving cars. It was intended to use AI to enhance Google’s existing products and search capabilities and was merged with DeepMind in 2023 to form Google DeepMind.

Innovations

In an early breakthrough DeepMind used deep reinforcement learning, a combination of reinforcement learning (trial and error with rewards and punishments) and deep learning (using layered neural networks), to create a program that was able to learn how to play some 50 different games on the Atari 2600 console without any prior knowledge of the game rules. The company continued applying its methods to games with AlphaGo, released in 2015, which became the first computer program to beat a human world champion in the game go, and AlphaStar, first unveiled in 2019, which became the first computer program to defeat a top-ranked player in StarCraft II.

DeepMind’s research has been central to advances in generative AI and natural language processing within Google’s AI ecosystem. In 2016 DeepMind introduced WaveNet, one of the first AI models to generate natural-sounding speech; an updated version was deployed as the voice of Google Assistant in 2017. Google Brain developed LaMDA (Language Model for Dialogue Applications) in 2021, the conversational AI that powered the early version of Google Bard, which was later succeeded by the Gemini family of models. Innovations at DeepMind have led to improvements in Google products, including search, Google Photos, and Google Maps.

DeepMind has also applied its AI systems to fields including genetics, computer coding, and weather prediction. Its AlphaFold system, which predicts the structure of proteins from their amino acid sequences, helped scientists better understand the virus that causes COVID-19 and contributed to advances in protein-structure prediction that were recognized with the 2024 Nobel Prize in Chemistry. In 2025 DeepMind released Aeneas, a program that can help historians analyze ancient engravings and determine when and where they were inscribed. That year a version of Gemini 2.5 became the first AI model to win gold at a computer programming competition for humans, solving a complex problem that none of the human teams were able to solve.

Explore Britannica Premium!

The trusted destination for professionals, college students, and lifelong learners.

AI research has frequently raised ethical and safety concerns because of its potential impact on society and human life. Some critics have also expressed concern about the possibility of creating highly autonomous—or sentient—systems and related ethical and safety questions, including how they can be stopped from overpowering humans. In 2016 DeepMind developed what it calls the “big red button” to stop AIs from learning to resist humans. The so-called “button” involves building “interruption policies” into a machine that will safely pause or shut down its workings.

Beginning in 2015, DeepMind partnered with the U.K.’s National Health Service in optimizing health care systems and developing methods of identifying patients who would benefit from assessments for illness, but in 2017 the U.K. Information Commissioner’s Office ruled that DeepMind’s use of personal data had breached the Data Protection Act. Later that year DeepMind formed a research group to address ethics and society and consider the impacts of artificial intelligence.