View source for Google DeepMind

{{Use British English|date=September 2016}}
{{Infobox company
| name = DeepMind Technologies Limited
| logo = [[File:DeepMind new logo.svg|frameless|upright=1.15|class=skin-invert]]
| image = 
| image_size = 
| image_caption = 
| trading_name = {{Ubl
| Google DeepMind
}}
| former_name = 
| founded = {{Start date and age|df=y|2010|09|23}} (incorporation)<ref name="CompaniesHouse">[https://find-and-update.company-information.service.gov.uk/company/07386350 "DeepMind Technologies Limited overview - Find and update company information - Gov.uk"]. ''[[Companies House]]''. 2010-09-23.</ref><br>
{{Start date and age|df=y|2010|11|15}} (official launch)<ref name="economist">[https://www.economist.com/1843/2019/03/01/deepmind-and-google-the-battle-to-control-artificial-intelligence "DeepMind and Google: the battle to control artificial intelligence"]. ''The Economist''. 1 March 2019.</ref>
| location = [[London]], England<ref>[https://www.ses-ltd.co.uk/case-study/kings-cross-s2-building/ "King's Cross – S2 Building – SES Engineering Services"]. ''ses-ltd.co.uk''.</ref>
| founders = {{plain list|
*[[Demis Hassabis]]
*[[Shane Legg]]
*[[Mustafa Suleyman]]}}
| key_people = {{Unbulleted list
  |[[Demis Hassabis]] ([[chief executive officer|CEO]])
  |[[Lila Ibrahim]] ([[chief operating officer|COO]])}}
| industry = [[Artificial intelligence]]
| parent = DeepMind Holdings Limited<ref>[https://find-and-update.company-information.service.gov.uk/company/07386350/persons-with-significant-control "Deepmind Technologies Limited persons with significant control – Find and update company information – Gov.uk"]. ''[[Companies House]]''. 2019-11-04.</ref>
| subsid = [[Google AI]]
| type = [[Subsidiary]]
| owner = [[Alphabet Inc.]]<ref>[https://find-and-update.company-information.service.gov.uk/company/12181850/persons-with-significant-control "Deepmind Holdings Limited persons with significant control – Find and update company information – GOV.UK"]. ''[[Companies House]]''. 2019-08-30.</ref>
| num_employees = c. 6,000 (2025)<ref>Herrera and Blunt, Sebastian and Katherine. [https://www.wsj.com/tech/ai/microsoft-google-deepmind-ai-recruitment-fcc60b67 "Microsoft Raids Google's DeepMind AI Unit With Promise of Less Bureaucracy"]. ''The Wall Street Journal''. 7 August 2025.</ref>
| products = {{UBL
| [[Google Gemini|Gemini]]
| [[Veo (text-to-video model)|Veo (video)]]
| [[Nano Banana]]
| [[Imagen (text-to-image model)|Imagen (image)]]
| [[Lyria (text-to-music model)|Lyria (music)]]
| [[AlphaFold]]
| [[AlphaGo]]
}}
| revenue = {{decrease}} £1.33&nbsp;billion (2024)<ref name=AR24>[https://find-and-update.company-information.service.gov.uk/company/07386350/filing-history/MzQ4Mjk4NjY3OWFkaXF6a2N4/document?format=pdf&download=0 "Full accounts made up to 31 December 2024"]. Companies House. 2 October 2025.</ref>
| operating_income = {{increase}} £217&nbsp;million (2024)<ref name=AR24 />
| net_income = {{increase}} £174&nbsp;million (2024)<ref name=AR24 />
| website = {{URL|https://deepmind.google/}}
}}
{{Artificial intelligence}}

'''DeepMind Technologies Limited''',<ref name="CompaniesHouse" /><!--full legal name per [[MOS:FIRSTCORP]]--> [[trading as]] '''Google DeepMind''' or simply '''DeepMind''', is a British-American [[artificial intelligence]] research laboratory which serves as a [[subsidiary]] of [[Alphabet Inc.]] Founded in the UK in 2010, it was [[List of mergers and acquisitions by Alphabet|acquired]] by Google in 2014<ref name=":4">Bray, Chad. [https://dealbook.nytimes.com/2014/01/27/google-acquires-british-artificial-intelligence-developer/ "Google Acquires British Artificial Intelligence Developer"]. ''DealBook''. 27 January 2014.</ref> and merged with [[Google AI]]'s [[Google Brain]] division to become Google DeepMind in April 2023. The company is headquartered in [[London]], with research centres in the United States, Canada,<ref name=":5">[https://deepmind.com/about/ "About Us"]. ''DeepMind''. 14 May 2024.</ref> France,<ref>[https://deepmind.com/blog/a-return-to-paris/ "A return to Paris"]. ''DeepMind''. 14 May 2024.</ref> Germany, and Switzerland.

In 2014, DeepMind introduced [[neural Turing machine]]s ([[Neural network (machine learning)|neural networks]] that can access external memory like a conventional [[Turing machine]]).<ref name="arxiv">Graves, Alex. "Neural Turing Machines".</ref><ref>[http://www.technologyreview.com/view/533741/best-of-2014-googles-secretive-deepmind-startup-unveils-a-neural-turing-machine/ "Best of 2014: Google's Secretive DeepMind Startup Unveils a "Neural Turing Machine" {{!"]. ''MIT Technology Review''. 2014-12-29.</ref> The company has created many neural network models trained with [[reinforcement learning]] to play [[video games]] and [[board games]]. It made headlines in 2016 after its [[AlphaGo]] program beat [[Lee Sedol]], a [[Go (game)|Go]] world champion, in [[AlphaGo versus Lee Sedol|a five-game match]], which was later featured in the documentary ''[[AlphaGo (film)|AlphaGo]]''.<ref>{{Citation|last=Kohs|first=Greg|title=AlphaGo|date=29 September 2017|url=https://www.imdb.com/title/tt6700846/|others=Ioannis Antonoglou, Lucas Baker, Nick Bostrom|access-date=9 January 2018|archive-date=6 April 2017|archive-url=https://web.archive.org/web/20170406000346/https://www.imdb.com/title/tt6700846/|url-status=live}}</ref> A more general program, [[AlphaZero]], beat the most powerful programs playing go, [[chess]] and [[shogi]] (Japanese chess) after a few days of play against itself using reinforcement learning.<ref>Silver, David. "Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm". 5 December 2017.</ref> DeepMind has since trained models for game-playing ([[MuZero]], [[AlphaStar (software)|AlphaStar]]), for geometry ([[AlphaGeometry]]), and for algorithm discovery ([[AlphaEvolve]], [[AlphaDev]], AlphaTensor).

In 2020, DeepMind made significant advances in the problem of [[protein structure prediction|protein folding]] with [[AlphaFold]], which achieved [[state of the art]] records on [[Benchmark (computing)|benchmark tests]] for protein folding prediction.<ref>Callaway, Ewen. [https://www.nature.com/articles/d41586-020-03348-4 "'It will change everything': DeepMind's AI makes gigantic leap in solving protein structuress"]. ''Nature''. 30 November 2020.</ref> In July 2022, it was announced that over 200 million predicted protein structures, representing virtually all known proteins, would be released on the AlphaFold database.<ref name="geddes">Geddes, Linda. [https://www.theguardian.com/technology/2022/jul/28/deepmind-uncovers-structure-of-200m-proteins-in-scientific-leap-forward "DeepMind uncovers structure of 200m proteins in scientific leap forward"]. ''The Guardian''. 28 July 2022.</ref><ref name="alphafold DB">[https://www.deepmind.com/blog/alphafold-reveals-the-structure-of-the-protein-universe "AlphaFold reveals the structure of the protein universe"]. ''DeepMind''. 28 July 2022.</ref>

Google DeepMind has become responsible for the development of [[Gemini (language model)|Gemini]] (Google's family of [[Large language model|large language models]]) and other [[generative AI]] tools, such as the [[Text-to-image model|text-to-image]] model [[Imagen (text-to-image model)|Imagen]], the [[Text-to-video model|text-to-video]] model [[Veo (text-to-video model)|Veo]], and the [[Text-to-music model|text-to-music]] model Lyria.

{{toclimit|3}}

== History ==

The [[Startup company|start-up]] was founded by [[Demis Hassabis]], [[Shane Legg]] and [[Mustafa Suleyman]] in November 2010.<ref name="economist"></ref> Hassabis and Legg first met at the Gatsby Computational Neuroscience Unit at [[University College London]] (UCL).<ref>Gibbs, Samuel. [https://www.theguardian.com/technology/shortcuts/2014/jan/28/demis-hassabis-15-facts-deepmind-technologies-founder-google "Demis Hassabis: 15 facts about the DeepMind Technologies founder"]. ''The Guardian''. 28 January 2014.</ref>

Demis Hassabis has said that the start-up began working on artificial intelligence technology by teaching it how to play old video games from the seventies and eighties, which are relatively primitive compared to the ones that are available today. Some of those games included ''[[Breakout (video game)|Breakout]]'', ''[[Pong]]'', and ''[[Space Invaders]]''. AI was introduced to one game at a time, without any prior knowledge of its rules. After spending some time on learning the game, AI would eventually become an expert in it. "The cognitive processes which the AI goes through are said to be very like those of a human who had never seen the game would use to understand and attempt to master it."<ref>Marr, Bernard. [https://www.forbes.com/sites/bernardmarr/2017/02/02/how-googles-amazing-ai-start-up-deepmind-is-making-our-world-a-smarter-place/#3f5f079ddfff "How Google's Amazing AI Start-Up 'DeepMind' Is Making Our World A Smarter Place"]. ''Forbes''.</ref> The goal of the founders is to create a general-purpose AI that can be useful and effective for almost anything.

Major venture capital firms [[Horizons Ventures]] and [[Founders Fund]] invested in the company,<ref>Cookson, Robert. [https://www.ft.com/content/b09dbd40-876a-11e3-9c5c-00144feab7de "DeepMind buy heralds rise of the machines"]. ''Financial Times''. 27 January 2014.</ref> as well as entrepreneurs [[Scott Banister]],<ref>[https://angel.co/deepmind-technologies-limited "DeepMind Technologies Investors"].</ref> [[Peter Thiel]],<ref>Shead, Sam. [https://www.businessinsider.com/how-deepmind-convinced-peter-thiel-to-invest-outside-silicon-valley-2017-7 "How DeepMind convinced billionaire Peter Thiel to invest without moving the company to Silicon Valley"]. Business Insider.</ref> and [[Elon Musk]].<ref>Rowan, David. [https://www.wired.co.uk/article/deepmind "DeepMind: inside Google's super-brain"]. ''Wired UK''. 22 June 2015.</ref> [[Jaan Tallinn]] was an early investor and an adviser to the company.<ref>[http://recode.net/2014/01/26/exclusive-google-to-buy-artificial-intelligence-startup-deepmind-for-400m/ "Recode.net – DeepMind Technologies Acquisition"]. 26 January 2014.</ref> On 26 January 2014, Google confirmed its acquisition of DeepMind for a price reportedly ranging between $400 million and $650 million.<ref>[https://www.reuters.com/article/google-deepmind-idUSL2N0L102A20140127 "Google to buy artificial intelligence company DeepMind"]. ''Reuters''. 26 January 2014.</ref><ref>Gibbs, Samuel. [https://www.theguardian.com/technology/2014/jan/27/google-acquires-uk-artificial-intelligence-startup-deepmind "Google Acquires UK AI startup Deepmind"]. ''The Guardian''. 27 January 2014.</ref><ref>Shu, Catherine. [https://techcrunch.com/2014/01/26/google-deepmind/ "Report of Acquisition, TechCrunch"]. ''TechCrunch''. 26 January 2014.</ref> and that it had agreed to take over DeepMind Technologies. The sale to Google took place after [[Facebook]] reportedly ended negotiations with DeepMind Technologies in 2013.<ref>Efrati, Amir. [https://www.theinformation.com/Google-beat-Facebook-For-DeepMind-Creates-Ethics-Board "Google beats Facebook for Acquisition of DeepMind Technologies"]. January 26, 2014.</ref> The company was afterwards renamed Google DeepMind and kept that name for about two years.<ref name="nature2015" />

In 2014, DeepMind received the "Company of the Year" award from [[Cambridge Computer Laboratory]].<ref>[https://www.cl.cam.ac.uk/ring/awards.html "Hall of Fame Awards: To celebrate the success of companies founded by Computer Laboratory graduates."]. University of Cambridge.</ref>
{{Multiple image|align=right|direction=vertical|width=260px|image1=Google DeepMind logo.svg|caption1=Logo from 2015–2016|image2=DeepMind logo.png|caption2=Logo from 2016–2019}}
In September 2015, DeepMind and the [[Royal Free London NHS Foundation Trust|Royal Free NHS Trust]] signed their initial information sharing agreement to co-develop a clinical task management app, Streams.<ref>Lomas, Natasha. [https://techcrunch.com/2017/08/31/documents-detail-deepminds-plan-to-apply-ai-to-nhs-data-in-2015/ "Documents detail DeepMind's plan to apply AI to NHS data in 2015"]. ''TechCrunch''.</ref>

After Google's acquisition the company established an [[Ethics of artificial intelligence|artificial intelligence ethics]] board.<ref>Selinger, Evan. [https://www.forbes.com/sites/privacynotice/2014/02/03/inside-googles-mysterious-ethics-board/ "Inside Google's Mysterious Ethics Board"]. ''Forbes''. 3 February 2014.</ref> The ethics board for AI research remains a mystery, with both Google and DeepMind declining to reveal who sits on the board.<ref name="theguardian.com 2016-05-04">Ramesh, Randeep. [https://www.theguardian.com/commentisfree/2016/may/04/googles-deepmind-shouldnt-be-sucking-up-our-nhs-records-in-secret "Google's DeepMind shouldn't suck up our NHS records in secret"]. ''The Guardian''. 4 May 2016.</ref> DeepMind has opened a new unit called DeepMind Ethics and Society and focused on the ethical and societal questions raised by artificial intelligence featuring prominent philosopher [[Nick Bostrom]] as advisor.<ref name=":9" /> In October 2017, DeepMind launched a new research team to investigate AI ethics.<ref>Shead, Sam. [http://www.businessinsider.com/deepmind-has-launched-a-new-ethics-and-society-research-team-2017-10 "DeepMind has launched a new 'ethics and society' research team"]. ''Business Insider''. October 4, 2017.</ref><ref>Vincent, James. [https://www.theverge.com/2017/10/4/16417978/deepmind-ai-ethics-society-research-group "DeepMind launches new research team to investigate AI ethics"]. ''The Verge''. October 4, 2017.</ref>

In December 2019, co-founder Suleyman announced he would be leaving DeepMind to join Google, working in a policy role.<ref name="SuleymanDeparture">Murgia, Madhumita. [https://www.ft.com/content/02757f12-1780-11ea-9ee4-11f260415385 "Client Challenge"]. ''www.ft.com''. 2019-12-05.</ref> In March 2024, [[Microsoft]] appointed him as the EVP and CEO of its newly created consumer AI unit, Microsoft AI.<ref>Blogs, Microsoft Corporate. [https://blogs.microsoft.com/blog/2024/03/19/mustafa-suleyman-deepmind-and-inflection-co-founder-joins-microsoft-to-lead-copilot/ "Mustafa Suleyman, DeepMind and Inflection Co-founder, joins Microsoft to lead Copilot"]. ''The Official Microsoft Blog''. 2024-03-19.</ref>

In April 2023, DeepMind merged with [[Google AI]]'s [[Google Brain]] division to form Google DeepMind, as part of the company's continued efforts to accelerate work on AI in response to [[OpenAI]]'s [[ChatGPT]].<ref>Roth, Emma. [https://www.theverge.com/2023/4/20/23691468/google-ai-deepmind-brain-merger "Google's big AI push will combine Brain and DeepMind into one team"]. ''[[The Verge]]''. 20 April 2023.</ref> This marked the end of a years-long struggle from DeepMind executives to secure greater autonomy from Google.<ref>Olson, Parmy. [https://www.wsj.com/articles/google-unit-deepmind-triedand-failedto-win-ai-autonomy-from-parent-11621592951 "Google Unit DeepMind Tried—and Failed—to Win AI Autonomy From Parent"]. ''[[The Wall Street Journal]]''. 21 May 2023.</ref>

== Products and technologies ==
As of 2020, DeepMind has published over a thousand papers, including thirteen papers that were accepted by ''[[Nature (journal)|Nature]]'' or ''[[Science (journal)|Science]]''. DeepMind received media attention during the AlphaGo period; according to a [[LexisNexis]] search, 1842 published news stories mentioned DeepMind in 2016, declining to 1363 in 2019.<ref>Shead, Sam. [https://www.cnbc.com/2020/06/05/google-deepmind-alphago-buzz-dissipates.html "Why the buzz around DeepMind is dissipating as it transitions from games to science"]. ''CNBC''. 5 June 2020.</ref>

===Games===
Unlike earlier AIs, such as [[IBM]]'s [[Deep Blue (chess computer)|Deep Blue]] or [[Watson (computer)|Watson]], which were developed for a pre-defined purpose and only function within that scope, DeepMind's initial algorithms were intended to be general. They used [[reinforcement learning]], an algorithm that learns from experience using only raw pixels as data input. Their initial approach used [[Q-learning#Deep Q-learning|deep Q-learning]] with a [[convolutional neural network]].<ref name="nature2015" /><ref name="Atari Paper">Mnih, Volodymyr. "Playing Atari with Deep Reinforcement Learning". 12 December 2013.</ref> They tested the system on video games, notably early [[arcade games]], such as ''[[Space Invaders]]'' or ''[[Breakout (video game)|Breakout]]''.<ref name="Atari Paper" /><ref name="hassabis talk" /> Without altering the code, the same AI was able to play certain games more efficiently than any human ever could.<ref name="hassabis talk">[https://www.youtube.com/watch?v=EfGD2qveGdQ "Deepmind artificial intelligence @ FDOT14"]. 19 April 2014.</ref>

In July 2018, researchers from DeepMind trained one of its systems to play the computer game ''[[Quake III Arena]]''.<ref>[https://www.engadget.com/2018/07/03/deepmind-ai-quake-iii-arena-human/ "DeepMind AI's new trick is playing 'Quake III Arena' like a human"] . ''Engadget''. 3 July 2018.</ref>

In 2013, DeepMind published research on an AI system that surpassed human abilities in games such as [[Pong]], [[Breakout (video game)|Breakout]] and [[Enduro (video game)|Enduro]], while surpassing state of the art performance on [[Seaquest (video game)|Seaquest]], [[Beamrider]], and [[Q*bert]].<ref>Wiggers, Kyle. [https://venturebeat.com/2018/12/29/a-look-back-at-some-of-ais-biggest-video-game-wins-in-2018/ "A look back at some of AI's biggest video game wins in 2018"]. ''VentureBeat''. 29 December 2018.</ref><ref>Mnih, Volodymyr. "Playing Atari with Deep Reinforcement Learning". 19 December 2013.</ref> This work reportedly led to the company's acquisition by Google.<ref name="arxiv medium">[https://medium.com/the-physics-arxiv-blog/the-last-ai-breakthrough-deepmind-made-before-google-bought-it-for-400m-7952031ee5e1 "The Last AI Breakthrough DeepMind Made Before Google Bought It For $400m"]. The Physics [[arXiv]] Blog. 29 January 2014.</ref> DeepMind's AI had been applied to video games made in the 1970s and [[History of video games#1980s|1980s]]; work was ongoing for more complex 3D games such as ''[[Quake (video game)|Quake]]'', which first appeared in the 1990s.<ref name="hassabis talk" />

In 2020, DeepMind published Agent57,<ref>Piot, Bilal. "Agent57: Outperforming the Atari Human Benchmark". 30 March 2020.</ref><ref>[https://deepmind.com/blog/article/Agent57-Outperforming-the-human-Atari-benchmark "Agent57: Outperforming the Atari Human Benchmark"]. ''DeepMind''. 31 March 2020.</ref> an AI Agent which surpasses human level performance on all 57 games of the Atari 2600 suite.<ref>Linder, Courtney. [https://www.popularmechanics.com/culture/gaming/a32006038/deepmind-ai-atari-agent57/ "This AI Can Beat Humans At All 57 Atari Games"]. ''Popular Mechanics''. 2 April 2020.</ref> In July 2022, DeepMind announced the development of DeepNash, a model-free [[multi-agent reinforcement learning]] system capable of playing the board game [[Stratego]] at the level of a human expert.<ref>Israni, Priyanka. [https://www.marktechpost.com/2022/07/09/deepmind-ai-researchers-introduce-deepnash-an-autonomous-agent-trained-with-model-free-multiagent-reinforcement-learning-that-learns-to-play-the-game-of-stratego-at-expert-level/ "Deepmind AI Researchers Introduce 'DeepNash', An Autonomous Agent Trained With Model-Free Multiagent Reinforcement Learning That Learns To Play The Game Of Stratego At Expert Level"]. ''MarkTechPost''. 9 July 2022.</ref>

==== AlphaGo and successors ====
''Main article: [[AlphaGo|AlphaGo Zero|AlphaZero|MuZero]]''
In October 2015, a [[computer Go]] program called AlphaGo, developed by DeepMind, beat the European Go champion [[Fan Hui]], a [[Go ranks and ratings|2 dan]] (out of 9 dan possible) professional, five to zero.<ref name="bbcgo">[https://www.bbc.com/news/technology-35420579 "Google achieves AI 'breakthrough' by beating Go champion"]. ''BBC News''. 27 January 2016.</ref> This was the first time an artificial intelligence (AI) defeated a professional Go player.<ref name="lemondego">Larousserie, David. [http://www.lemonde.fr/pixels/article/2016/01/27/premiere-defaite-d-un-professionnel-du-go-contre-une-intelligence-artificielle_4854886_4408996.html "Première défaite d'un professionnel du go contre une intelligence artificielle"]. ''Le Monde''. 27 January 2016.</ref> Previously, computers were only known to have played Go at "amateur" level.<ref name="bbcgo" /><ref name="googlego">[http://googleresearch.blogspot.com/2016/01/alphago-mastering-ancient-game-of-go.html "Research Blog: AlphaGo: Mastering the ancient game of Go with Machine Learning"]. ''Google Research Blog''. 27 January 2016.</ref> Go is considered much more difficult for computers to win compared to other games like [[chess]], due to the much larger number of possibilities, making it prohibitively difficult for traditional AI methods such as [[Brute-force search|brute-force]].<ref name="bbcgo" /><ref name="googlego" />

In March 2016 it beat [[Lee Sedol]], a [[Go ranks and ratings|9-dan]] professional player, with a score of 4 to 1 in a [[AlphaGo versus Lee Sedol|five-game match]]. In the 2017 [[Future of Go Summit]], AlphaGo won a [[AlphaGo versus Ke Jie|three-game match with Ke Jie]], who had been the world's highest-ranked player for two years.<ref>[http://www.goratings.org/ "World's Go Player Ratings"]. May 2017.</ref><ref>[http://sports.sina.com.cn/go/2016-08-02/doc-ifxunyya3020238.shtml "柯洁迎19岁生日 雄踞人类世界排名第一已两年"]. May 2017.</ref> In 2017, an improved version, [[AlphaGo Zero]], defeated AlphaGo in a hundred out of a hundred games. Later that year, [[AlphaZero]], a modified version of AlphaGo Zero, gained superhuman abilities at chess and shogi. In 2019, DeepMind released a new model named [[MuZero]] that mastered the domains of [[Go (game)|Go]], [[chess]], [[shogi]], and [[Atari 2600|Atari 2600 games]] without human data, domain knowledge, or known rules.<ref>[https://www.deepmind.com/blog/muzero-mastering-go-chess-shogi-and-atari-without-rules "MuZero: Mastering Go, chess, shogi and Atari without rules"]. ''www.deepmind.com''. 23 December 2020.</ref><ref>Schrittwieser, Julian. [https://www.nature.com/articles/s41586-020-03051-4.epdf?sharing_token=kTk-xTZpQOF8Ym8nTQK6EdRgN0jAjWel9jnR3ZoTv0PMSWGj38iNIyNOw_ooNp2BvzZ4nIcedo7GEXD7UmLqb0M_V_fop31mMY9VBBLNmGbm0K9jETKkZnJ9SgJ8Rwhp3ySvLuTcUr888puIYbngQ0fiMf45ZGDAQ7fUI66-u7Y= "Mastering Atari, Go, chess and shogi by planning with a learned model"]. ''Nature''. 23 December 2020.</ref>

AlphaGo technology was developed based on [[deep reinforcement learning]], making it different from the AI technologies then on the market. The data fed into the AlphaGo algorithm consisted of various moves based on historical tournament data. The number of moves was increased gradually until over 30 million of them were processed. The aim was to have the system mimic the human player, as represented by the input data, and eventually become better. It played against itself and learned from the outcomes; thus, it learned to improve itself over the time and increased its winning rate as a result.<ref>[https://www.economist.com/news/science-and-technology/21730391-learning-play-go-only-start-latest-ai-can-work-things-out-without "The latest AI can work things out without being taught"]. ''The Economist''.</ref>

AlphaGo used two deep neural networks: a policy network to evaluate move probabilities and a value network to assess positions. The policy network trained via supervised learning, and was subsequently refined by policy-gradient [[reinforcement learning]]. The value network learned to predict winners of games played by the policy network against itself. After training, these networks employed a lookahead [[Monte Carlo tree search]], using the policy network to identify candidate high-probability moves, while the value network (in conjunction with Monte Carlo rollouts using a fast rollout policy) evaluated tree positions.<ref name=":0">Silver, David. [http://discovery.ucl.ac.uk/10045895/1/agz_unformatted_nature.pdf "Mastering the game of Go without human knowledge"]. ''[[Nature (journal)''. 19 October 2017.{{closed access}}</ref>

In contrast, AlphaGo Zero was trained without being fed data of human-played games. Instead it generated its own data, playing millions of games against itself. It used a single neural network, rather than separate policy and value networks. Its simplified tree search relied upon this neural network to evaluate positions and sample moves. A new reinforcement learning algorithm incorporated lookahead search inside the training loop.<ref name=":0" /> AlphaGo Zero employed around 15 people and millions in computing resources.<ref>Knight, Will. [https://www.technologyreview.com/s/609141/alphago-zero-shows-machines-can-become-superhuman-without-any-help/ "The world's smartest game-playing AI—DeepMind's AlphaGo—just got way smarter"]. ''MIT Technology Review''.</ref> Ultimately, it needed much less computing power than AlphaGo, running on four specialized AI processors (Google [[Tensor processing unit|TPUs]]), instead of AlphaGo's 48.<ref>Vincent, James. [https://www.theverge.com/2017/10/18/16495548/deepmind-ai-go-alphago-zero-self-taught "DeepMind's Go-playing AI doesn't need human help to beat us anymore"]. ''The Verge''. 18 October 2017.</ref> It also required less training time, being able to beat its predecessor after just three days, compared with months required for the original AlphaGo.<ref>Cellan-Jones, Rory. [https://www.bbc.com/news/technology-41668701 "Google DeepMind: AI becomes more alien"]. ''BBC News''. 18 October 2017.</ref> Similarly, AlphaZero also learned via [[Self-play (reinforcement learning technique)|self-play]].

Researchers applied MuZero to solve the real world challenge of video compression with a set number of bits with respect to Internet traffic on sites such as [[YouTube]], [[Twitch (service)|Twitch]], and [[Google Meet]]. The goal of MuZero is to optimally compress the video so the quality of the video is maintained with a reduction in data. The final result using MuZero was a 6.28% average reduction in bitrate.<ref>[https://www.deepmind.com/blog/muzeros-first-step-from-research-into-the-real-world "MuZero's first step from research into the real world"]. ''www.deepmind.com''. 11 February 2022.</ref><ref>Mandhane, Amol. "MuZero with Self-competition for Rate Control in VP9 Video Compression". 14 February 2022.</ref>

==== AlphaStar ====
''Main article: [[AlphaStar (software)]]''

In 2016, Hassabis discussed the game ''[[StarCraft]]'' as a future challenge, since it requires strategic thinking and handling imperfect information.<ref>Byford, Sam. [https://www.theverge.com/2016/3/10/11192774/demis-hassabis-interview-alphago-google-deepmind-ai "DeepMind founder Demis Hassabis on how AI will shape the future"]. ''The Verge''. 10 March 2016.</ref>

In January 2019, DeepMind introduced AlphaStar, a program playing the real-time strategy game ''[[StarCraft II]]''. AlphaStar used reinforcement learning based on replays from human players, and then played against itself to enhance its skills. At the time of the presentation, AlphaStar had knowledge equivalent to 200 years of playing time. It won 10 consecutive matches against two professional players, although it had the unfair advantage of being able to see the entire field, unlike a human player who has to move the camera manually. A preliminary version in which that advantage was fixed lost a subsequent match.<ref>Whitwam, Ryan. [http://www.extremetech.com/gaming/284441-deepmind-ai-challenges-pro-starcraft-ii-players-wins-almost-every-match "DeepMind AI Challenges Pro StarCraft II Players, Wins Almost Every Match"]. ''Extreme Tech''. 24 January 2019.</ref>

In July 2019, AlphaStar began playing against random humans on the public 1v1 European multiplayer ladder. Unlike the first iteration of AlphaStar, which played only [[Protoss]] v. Protoss, this one played as all of the game's races, and had earlier unfair advantages fixed.<ref>Amadeo, Ron. [https://arstechnica.com/gadgets/2019/07/deepmind-ai-takes-on-the-public-in-starcraft-ii-multiplayer/ "DeepMind AI is secretly lurking on the public StarCraft II 1v1 ladder"]. ''Ars Technica''. 11 July 2019.</ref> By October 2019, AlphaStar had reached Grandmaster level on the ''StarCraft II'' ladder on all three ''StarCraft'' races, becoming the first AI to reach the top league of a widely popular [[Esports|esport]] without any game restrictions.<ref>[https://deepmind.com/blog/article/AlphaStar-Grandmaster-level-in-StarCraft-II-using-multi-agent-reinforcement-learning "AlphaStar: Grandmaster level in StarCraft II using multi-agent reinforcement learning"]. ''DeepMind Blog''. 31 October 2019.</ref>

=== Datacenter operation ===
In 2014, a datacenter engineer at Google began using supervised machine learning to predict [[power usage effectiveness]] (PUE) of datacenters at Google. The system was deployed in production to allow operators to simulate control strategies and pick the one that saves the most energy.<ref>Gao, Jim. [https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/42542.pdf "Machine Learning Applications for Data Center Optimization"]. ''Google White Paper''. 2014.</ref><ref>Kava, Joe. [https://blog.google/inside-google/infrastructure/better-data-centers-through-machine/ "Better data centers through machine learning"]. ''The Keyword''. 2014-05-28.</ref> In 2016, inspired by AlphaGo, he contacted DeepMind to apply [[reinforcement learning]] (RL) to train a system that could also recommend actions. It was tested on a live datacenter. The system read from sensor readings and recommended actions to take, and human engineers would implement the actions. Though the human engineers found its recommendations unintuitive, they satisfied all safety constraints, and led to a 15% saving in PUE.<ref>Evans, Rich. [https://blog.google/outreach-initiatives/environment/deepmind-ai-reduces-energy-used-for/ "DeepMind AI reduces energy used for cooling Google data centers by 40%"]. ''The Keyword''. 2016-07-20.</ref> The system was deployed more widely across Google, with datacenter controllers receiving email recommendations from the system every 15 minutes.<ref name=":8">[https://www.sequoiacap.com/podcast/training-data-jim-gao/ "Phaidra's Jim Gao on Building for the Fourth Industrial Revolution"]. ''Sequoia Capital''. 2024-08-20.</ref>

Eventually a more mature and more autonomous system was deployed, where the AI's actions are checked against safety constraints and implemented autonomously if verified safe, and human operators would supervise the AI and may override. The system led to a 30% saving in PUE. The system produced cooling strategies that surprised long-time operators, such as exploiting winter conditions to produce colder than normal water.<ref name=":8" /><ref>Gamble, Chris. [https://www.deepmind.com/blog/safety-first-ai-for-autonomous-data-centre-cooling-and-industrial-control "Safety-first AI for autonomous data centre cooling and industrial control"]. ''DeepMind''. 2018-08-17.</ref> Google subsequently collaborated with [[Trane Technologies]] to deploy similar RL-based systems on [[Heating, ventilation, and air conditioning|HVAC]] of facilities outside of Google.<ref>Luo, Jerry. "Controlling Commercial Cooling Systems Using Reinforcement Learning". 2022-12-14.</ref>

=== Protein folding ===
''Main article: [[AlphaFold]]''

In 2016, DeepMind turned its artificial intelligence to [[protein structure prediction|protein folding]], a long-standing problem in [[molecular biology]]. In December 2018, DeepMind's AlphaFold won the 13th [[Critical Assessment of Techniques for Protein Structure Prediction]] (CASP) by successfully predicting the most accurate structure for 25 out of 43 proteins. "This is a lighthouse project, our first major investment in terms of people and resources into a fundamental, very important, real-world scientific problem," Hassabis said to ''The Guardian''.<ref>Sample, Ian. [https://www.theguardian.com/science/2018/dec/02/google-deepminds-ai-program-alphafold-predicts-3d-shapes-of-proteins "Google's DeepMind predicts 3D shapes of proteins"]. ''[[The Guardian]]''. 2 December 2018.</ref> In 2020, in the 14th CASP, AlphaFold's predictions achieved an accuracy score regarded as comparable with lab techniques. Andriy Kryshtafovych, one of the panel of scientific adjudicators, described the achievement as "truly remarkable", and said the problem of predicting how proteins fold had been "largely solved".<ref>Briggs, Helen. [https://www.bbc.co.uk/news/science-environment-55133972 "One of biology's biggest mysteries 'largely solved' by AI"]. ''[[BBC News]]''. 30 November 2020.</ref><ref>[https://deepmind.com/blog/article/alphafold-a-solution-to-a-50-year-old-grand-challenge-in-biology "AlphaFold: a solution to a 50-year-old grand challenge in biology"]. ''DeepMind''. 30 November 2020.</ref><ref>Shead, Sam. [https://www.cnbc.com/2020/11/30/deepmind-solves-protein-folding-grand-challenge-with-alphafold-ai.html "DeepMind solves 50-year-old 'grand challenge' with protein folding A.I."]. cnbc.com. 30 November 2020.</ref>

In July 2021, the open-source RoseTTAFold and AlphaFold2 were released to allow scientists to run their own versions of the tools. A week later DeepMind announced that AlphaFold had completed its prediction of nearly all human proteins as well as the entire [[proteome]]s of 20 other widely studied organisms.<ref>Callaway, Ewen. "What's next for AlphaFold and the AI protein-folding revolution". ''Nature''. 2022.</ref> The structures were released on the AlphaFold Protein Structure Database. In July 2022, it was announced that the predictions of over 200 million proteins, representing virtually all known proteins, would be released on the AlphaFold database.<ref name=geddes /><ref name="alphafold DB"/>

The most recent update, AlphaFold3, was released in May 2024, predicting the interactions of proteins with DNA, RNA, and various other molecules. In a particular [[Benchmark (computing)|benchmark test]] on the problem of DNA interactions, AlphaFold3's attained an accuracy of 65%, significantly improving the previous state of the art of 28%.<ref>Sullivan, Mark. [https://www.fastcompany.com/91120456/deepmind-alphafold-3-dna-rna-modeling "DeepMind's new AlphaFold 3 expands to DNA, RNA modeling"]. ''[[Fast Company]]''. May 8, 2024.</ref>

In October 2024, Hassabis and [[John M. Jumper|John Jumper]] received half of the 2024 [[Nobel Prize in Chemistry]] jointly for protein structure prediction, citing AlphaFold2 achievement.<ref>[https://www.nobelprize.org/prizes/chemistry/2024/press-release/ "The Nobel Prize in Chemistry 2024"]. ''NobelPrize.org''.</ref>

===Language models===
In 2016, DeepMind introduced [[WaveNet]], a [[text-to-speech]] system. It was originally too computationally intensive for use in consumer products, but in late 2017 it became ready for use in consumer applications such as [[Google Assistant]].<ref>[http://fortune.com/2017/10/05/google-assistant-deepmind-wavenet-speech-ai/ "Here's Why Google's Assistant Sounds More Realistic Than Ever Before"]. ''Fortune''. 5 October 2017.</ref><ref>Gershgorn, Dave. [https://qz.com/1165775/googles-voice-generating-ai-is-now-indistinguishable-from-humans/ "Google's voice-generating AI is now indistinguishable from humans"]. ''Quartz''.</ref> In 2018 Google launched a commercial text-to-speech product, Cloud Text-to-Speech, based on WaveNet.<ref name="cnbc money">Novet, Jordan. [https://www.cnbc.com/2018/03/31/how-google-makes-money-from-alphabets-deepmind-ai-research-group.html "Google is finding ways to make money from Alphabet's DeepMind A.I. technology"]. ''CNBC''. 31 March 2018.</ref><ref>[https://cloudplatform.googleblog.com/2018/03/introducing-Cloud-Text-to-Speech-powered-by-Deepmind-WaveNet-technology.html "Introducing Cloud Text-to-Speech powered by DeepMind WaveNet technology"]. ''Google Cloud Platform Blog''.</ref> In 2018, DeepMind introduced a more efficient model called WaveRNN co-developed with [[Google AI]].<ref>[https://deepmind.com/research/publications/efficient-neural-audio-synthesis "Efficient Neural Audio Synthesis"]. ''Deepmind''.</ref><ref>[https://deepmind.com/blog/article/Using-WaveNet-technology-to-reunite-speech-impaired-users-with-their-original-voices "Using WaveNet technology to reunite speech-impaired users with their original voices"]. ''Deepmind''. 18 December 2019.</ref> In 2020 WaveNetEQ, a packet loss concealment method based on a WaveRNN architecture, was presented.<ref>Citation needed.</ref> In 2019, Google started to roll WaveRNN with WavenetEQ out to [[Google Duo]] users.<ref>[http://ai.googleblog.com/2020/04/improving-audio-quality-in-duo-with.html "Improving Audio Quality in Duo with WaveNetEQ"]. ''Google AI Blog''. April 2020.</ref>

Released in May 2022, [[Gato (DeepMind)|Gato]] is a polyvalent [[multimodal learning|multimodal]] model. It was trained on 604 tasks, such as image captioning, dialogue, or stacking blocks. On 450 of these tasks, Gato outperformed human experts at least half of the time, according to DeepMind.<ref>Wiggers, Kyle. [https://techcrunch.com/2022/05/13/deepminds-new-ai-can-perform-over-600-tasks-from-playing-games-to-controlling-robots/ "DeepMind's new AI system can perform over 600 tasks"]. ''TechCrunch''. 13 May 2022.</ref> Unlike models like MuZero, Gato does not need to be retrained to switch from one task to the other.

[[Sparrow (chatbot)|Sparrow]] is an artificial intelligence-powered [[chatbot]] developed by DeepMind to build safer machine learning systems by using a mix of human feedback and Google search suggestions.<ref>Gupta, Khushboo. [https://www.marktechpost.com/2022/09/28/deepmind-introduces-sparrow-an-artificial-intelligence-powered-chatbot-developed-to-build-safer-machine-learning-systems/ "Deepmind Introduces 'Sparrow,' An Artificial Intelligence-Powered Chatbot Developed To Build Safer Machine Learning Systems"]. 28 September 2022.</ref>

[[Chinchilla (language model)|Chinchilla]] is a language model developed by DeepMind.<ref>[https://dataconomy.com/2023/01/12/what-is-chinchilla-ai-chatbot-deepmind/ "What Is Chinchilla AI: Chatbot Language Model Rival By Deepmind To GPT-3"]. ''Dataconomy''. 12 January 2023.</ref>

DeepMind posted a blog post on 28 April 2022 on a single visual language model (VLM) named Flamingo that can accurately describe a picture of something with just a few training images.<ref>Alayrac, Jean-Baptiste. [https://www.deepmind.com/blog/tackling-multiple-tasks-with-a-single-visual-language-model "Tackling multiple tasks with a single visual language model"]. ''www.deepmind.com''. 28 April 2022.</ref><ref>Alayrac, Jean-Baptiste. [https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/tackling-multiple-tasks-with-a-single-visual-language-model/flamingo.pdf "Flamingo: a Visual Language Model for Few-Shot Learning"]. ''''.</ref>

==== AlphaCode ====
In 2022, DeepMind unveiled AlphaCode, an [[AI-assisted software development|AI-powered coding engine]] that creates [[computer programs]] at a rate comparable to that of an average programmer, with the company testing the system against coding challenges created by [[Codeforces]] utilized in human [[competitive programming]] competitions.<ref>Vincent, James. [https://www.theverge.com/2022/2/2/22914085/alphacode-ai-coding-program-automatic-deepmind-codeforce "DeepMind says its new AI coding engine is as good as an average human programmer"]. ''[[The Verge]]''. 2 February 2022.</ref> AlphaCode earned a rank equivalent to 54% of the median score on Codeforces after being trained on [[GitHub]] data and Codeforce problems and solutions. The program was required to come up with a unique solution and stopped from duplicating answers.

====Gemini====
''Main article: [[Gemini (language model)]]''

Gemini is a [[Multimodal learning|multimodal]] [[large language model]] which was released on 6 December 2023.<ref>Kruppa, Miles. [https://www.wsj.com/tech/ai/google-announces-ai-system-gemini-after-turmoil-at-rival-openai-10835335 "Google Announces AI System Gemini After Turmoil at Rival OpenAI"]. ''[[The Wall Street Journal]]''. 6 December 2023.</ref> It is the successor of Google's [[LaMDA]] and [[PaLM|PaLM 2]] language models and sought to challenge OpenAI's [[GPT-4]].<ref>Knight, Will. [https://www.wired.com/story/google-deepmind-demis-hassabis-chatgpt/ "Google DeepMind's CEO Says Its Next Algorithm Will Eclipse ChatGPT"]. ''[[Wired (magazine)''. 26 June 2023.</ref> Gemini comes in 3 sizes: Nano, Pro, and Ultra.<ref>Pierce, David. [https://www.theverge.com/2023/12/6/23990466/google-gemini-llm-ai-model "Google launches Gemini, the AI model it hopes will take down GPT-4"]. ''The Verge''. 6 December 2023.</ref> Gemini is also the name of the chatbot that integrates Gemini (and which was previously called [[Google Bard|Bard]]).<ref>[https://www.cbsnews.com/news/google-gemini-ai-bard/ "Google is rebranding its Bard AI service as Gemini. Here's what it means."]. ''CBS News''. 8 February 2024.</ref>

On 12 December 2024, Google released Gemini 2.0 Flash, the first model in the Gemini 2.0 series. It notably features expanded multimodality, with the ability to also generate images and audio,<ref>Haddad, C. J.. [https://www.cnbc.com/2024/12/11/google-releases-the-first-of-its-gemini-2point0-ai-models.html "Google releases the first of its Gemini 2.0 AI models"]. ''CNBC''. 2024-12-11.</ref> and is part of Google's broader plans to integrate advanced AI into [[Autonomous agent|autonomous agents]].<ref>[https://blog.google/technology/google-deepmind/google-gemini-ai-update-december-2024/?utm_content=#agents-for-developers "Introducing Gemini 2.0: our new AI model for the agentic era"]. ''Google''. 2024-12-11.</ref>

On 25 March 2025, Google released Gemini 2.5, a reasoning model that stops to "think" before giving a response. Google announced that all future models will also have reasoning ability.<ref>Zeff, Maxwell. [https://techcrunch.com/2025/03/25/google-unveils-a-next-gen-ai-reasoning-model/ "Google unveils a next-gen family of AI reasoning models"]. ''TechCrunch''. 2025-03-25.</ref><ref>[https://blog.google/technology/google-deepmind/gemini-model-thinking-updates-march-2025/#enhanced-reasoning "Gemini 2.5: Our most intelligent AI model"]. ''Google''. 2025-03-25.</ref> On 30 March 2025, Google released Gemini 2.5 to all free users.<ref>Kumari, Sweta. [https://www.business-standard.com/technology/tech-news/google-rolls-out-custom-chatbots-gems-for-free-tier-gemini-users-details-125032600913_1.html "Google rolls-out custom chatbots 'Gems' for free-tier Gemini users: Details"]. ''Business Standard''. 2025-03-26.</ref>

On 18 November 2025, Google released Gemini 3 Pro, a reasoning model which is fully multimodal.<ref>[https://blog.google/products/gemini/gemini-3/ "A new era of intelligence with Gemini 3"]. ''Google''. 2025-11-18.</ref> It was fully integrated with Google Search and AI Mode the same day.<ref>Knight, Will. [https://www.wired.com/story/google-launches-gemini-3-ai-bubble-search/ "Gemini 3 Is Here—and Google Says It Will Make Search Smarter"]. ''Wired''.</ref>

====Gemma====
''Main article: [[Gemma (language model)]]''

Gemma is a collection of open-weight large language models. The first ones were released on 21 February 2024 and are available in two distinct sizes: a 7 billion parameter model optimized for GPU and TPU usage, and a 2 billion parameter model designed for CPU and on-device applications. Gemma models were trained on up to 6 trillion tokens of text, employing similar architectures, datasets, and training methodologies as the Gemini model set.<ref>Quach, Katyanna. [https://www.theregister.com/2024/02/22/google_gemma_llms/ "Google Gemma LLMs small enough to run on your computer"]. ''The Register''. 2024-02-22.</ref>

In June 2024, Google started releasing Gemma 2 models.<ref>Yeung, Ken. [https://venturebeat.com/ai/googles-gemma-2-series-launches-with-not-one-but-two-lightweight-model-options-a-9b-and-27b/ "Google's Gemma 2 series launches with not one, but two lightweight model options—a 9B and 27B"]. ''VentureBeat''. 2024-06-27.</ref> In December 2024, Google introduced ''PaliGemma 2'', an upgraded vision-language model.<ref>Wiggers, Kyle. [https://techcrunch.com/2024/12/05/google-says-its-new-open-models-can-identify-emotions-and-that-has-experts-worried/ "Google says its new AI models can identify emotions — and that has experts worried"]. ''TechCrunch''. 5 December 2024.</ref> In February 2025, they launched ''PaliGemma 2 Mix'', a version fine-tuned for multiple tasks. It is available in 3B, 10B, and 28B parameters with 224px and 448px resolutions.<ref>Barron, Jenna. [https://sdtimes.com/ai/feb-21-2025-development-tools-that-have-recently-added-new-ai-capabilities/ "Feb 21, 2025: Development tools that have recently added new AI capabilities"]. ''SD Times''. 2025-02-21.</ref>

In March 2025, Google released Gemma 3, calling it the most capable model that can be run on a single GPU.<ref>Lawler, Richard. [https://www.theverge.com/ai-artificial-intelligence/627968/google-gemma-3-open-ai-model "Google calls Gemma 3 the most powerful AI model you can run on one GPU"]. ''The Verge''. 2025-03-12.</ref> It has four available sizes: 1B, 4B, 12B, and 27B.<ref>David, Emilia. [https://venturebeat.com/ai/google-unveils-open-source-gemma-3-model-with-128k-context-window/ "Google unveils open source Gemma 3 model with 128k context window"]. ''VentureBeat''. 2025-03-12.</ref> In March 2025, Google introduced TxGemma, an open-source model designed to improve the efficiency of therapeutics development.<ref>Azizi, Shekoofeh. [https://developers.googleblog.com/en/introducing-txgemma-open-models-improving-therapeutics-development/#:~:text=TxGemma%20models,%20fine-tuned%20from,:%202B,%209B%20and%2027B. "Introducing TxGemma: Open models to improve therapeutics development"]. ''Google Developers Blog''. 25 March 2025.</ref>

In April 2025, Google introduced DolphinGemma, a research artificial intelligence model designed to hopefully decode dolphin communication. They want to train a foundation model that can learn the structure of dolphin vocalizations and generate novel dolphin-like sound sequences.<ref>Starner, Thad. [https://blog.google/technology/ai/dolphingemma/ "DolphinGemma: How Google AI is helping decode dolphin communication"]. ''Google''. 2025-04-14.</ref><ref>Aadeetya, S. [https://www.news18.com/tech/dolphingemma-google-using-ai-to-understand-what-dolphins-are-saying-9299420.html "DolphinGemma: Google Using AI And Pixel 9 Phone To Understand What Dolphins Are Saying"]. ''News18''.</ref>

====SIMA====
In March 2024, DeepMind introduced Scalable Instructable Multiword Agent, or SIMA, an AI agent capable of understanding and following natural language instructions to complete tasks across various 3D virtual environments. Trained on nine video games from eight studios and four research environments, SIMA demonstrated adaptability to new tasks and settings without requiring access to game source code or APIs. The agent comprises pre-trained computer vision and language models fine-tuned on gaming data, with language being crucial for understanding and completing given tasks as instructed. DeepMind's research aimed to develop more helpful AI agents by translating advanced AI capabilities into real-world actions through a language interface.<ref>[https://deepmind.google/discover/blog/sima-generalist-ai-agent-for-3d-virtual-environments/ "A generalist AI agent for 3D virtual environments"]. ''Google DeepMind''. 13 March 2024.</ref><ref>David, Emilia. [https://www.theverge.com/2024/3/13/24099024/google-deepmind-ai-agent-sima-video-games "Google's new AI will play video games with you — but not to win"]. ''The Verge''. 13 March 2024.</ref>

==== Habermas machine ====
''See also: [[Pol.is|Deliberative opinion poll]]''
In 2024, Google Deepmind published the results of an experiment where they trained two [[Large language model|large language models]] to help identify and present areas of overlap among a few thousand group members they had recruited online using techniques like [[sortition]] to get a representative sample of participants. The project is named in honor of [[Jürgen Habermas]].<ref>Williams, Rhiannon. [https://www.technologyreview.com/2024/10/17/1105810/ai-could-help-people-find-common-ground-during-deliberations/ "AI could help people find common ground during deliberations"]. ''MIT Technology Review''. October 17, 2024.</ref><ref name=":6">Davis, Nicola. [https://www.theguardian.com/technology/2024/oct/17/ai-mediation-tool-may-help-reduce-culture-war-rifts-say-researchers "AI mediation tool may help reduce culture war rifts, say researchers"]. ''The Guardian''. 2024-10-17.</ref> In one experiment, the participants rated the summaries by the AI higher than the human moderator 56% of the time.<ref name=":6" />

=== Generative AI ===

==== Video generation ====
''Main article: [[Veo (text-to-video model)]]''

In May 2024, a [[Multimodality|multimodal]] [[Text-to-video model|video generation model]] called Veo was announced at [[Google I/O]] 2024. Google claimed that it could generate [[1080p]] videos beyond a minute long.<ref name="Wiggers 14 May 2024">Wiggers, Kyle. [https://techcrunch.com/2024/05/14/google-veo-a-serious-swing-at-ai-generated-video-debuts-at-google-io-2024/ "Google Veo, a serious swing at AI-generated video, debuts at Google I/O 2024"]. ''TechCrunch''. 14 May 2024.</ref> In December 2024, [[Google]] released Veo 2, available via VideoFX. It supports [[4K resolution]] video generation, and has an improved understanding of physics.<ref>, . [https://www.thehindu.com/sci-tech/technology/google-unveils-improved-ai-video-generator-veo-2-to-rival-openais-sora/article68994621.ece "Google unveils improved AI video generator Veo 2 to rival OpenAI's Sora"]. ''The Hindu''. 2024-12-17.</ref> In April 2025, Google announced that Veo 2 became available for advanced users on Gemini App.<ref>Wiggers, Kyle. [https://techcrunch.com/2025/04/15/googles-veo-2-video-generator-comes-to-gemini/ "Google's Veo 2 video generating model comes to Gemini"]. ''TechCrunch''. 2025-04-15.</ref>

In May 2025, Google released Veo 3, which not only generates videos but also creates synchronized audio — including dialogue, sound effects, and ambient noise — to match the visuals.<ref>Elias, Jennifer. [https://www.cnbc.com/2025/05/20/google-ai-video-generator-audio-veo-3.html "Google launches Veo 3, an AI video generator that incorporates audio"]. ''CNBC''. 2025-05-20.</ref><ref>Wiggers, Kyle. [https://techcrunch.com/2025/05/20/googles-veo-3-can-generate-videos-and-soundtracks-to-go-along-with-them/ "Veo 3 can generate videos — and soundtracks to go along with them"]. ''TechCrunch''. 2025-05-20.</ref>

Google also announced Flow, a video-creation tool powered by Veo and [[Imagen (text-to-image model)|Imagen]].<ref>Peters, Jay. [https://www.theverge.com/news/670181/google-deepmind-ai-videos-app-flow-veo-3-2-imagen-4-io-2025 "Google has a new tool just for making AI videos"]. ''[[The Verge]]''. May 20, 2025.</ref>

==== Music generation ====
[[File:Digital Sonata of the Speed God (Lyria 3).opus|thumb|A music generated with Lyria 3]]
Google DeepMind developed Lyria, a text-to-music model. As of August 2025, it is available on Vertex AI and the Gemini API.<ref>[https://cloud.google.com/vertex-ai/generative-ai/docs/music/generate-music "Vertex AI {{!"]. ''Google Cloud''.</ref><ref>[https://ai.google.dev/gemini-api/docs/music-generation "Music generation using Lyria RealTime"].</ref><ref>Wiggers, Kyle. [https://techcrunch.com/2025/04/09/google-brings-a-music-generating-ai-model-to-its-enterprise-cloud/ "Google's enterprise cloud gets a music-generating AI model"]. ''TechCrunch''. 2025-04-09.</ref> On February 18, 2026, DeepMind released Lyria 3.<ref>Whitwam, Ryan. [https://arstechnica.com/google/2026/02/gemini-can-now-generate-ai-music-for-you-no-lyrics-required/ "Record scratch—Google's Lyria 3 AI music model is coming to Gemini today"]. ''Ars Technica''. 2026-02-18.</ref> On March 25, 2026, DeepMind released Lyria 3 Pro which allows users to create longer tracks with more structural awareness.<ref>Mehta, Ivan. [https://techcrunch.com/2026/03/25/google-launches-lyria-3-pro-music-generation-model/ "Google launches Lyria 3 Pro music generation model"]. ''TechCrunch''. 2026-03-25.</ref> 

==== Environment generation ====
''Main article: [[Genie (AI model)]]''
In March 2024, DeepMind introduced "[[Genie (text-to-video model)|Genie]]" (Generative Interactive Environments), an AI model that can generate game-like, action-controllable virtual worlds based on textual descriptions, images, or sketches. Built as an autoregressive [[latent diffusion model]], Genie enables frame-by-frame interactivity without requiring labeled action data for training. Its successor, Genie 2, released in December 2024, expanded these capabilities to generate diverse and interactive 3D environments.<ref>Orland, Kyle. [https://arstechnica.com/ai/2024/12/googles-genie-2-world-model-reveal-leaves-more-questions-than-answers/ "Google's Genie 2 "world model" reveal leaves more questions than answers"]. ''Ars Technica''. 2024-12-06.</ref> Genie 3 was released in August 2025, with higher-resolution world generations and multiple minutes of visual consistency.<ref>Whitwam, Ryan. [https://arstechnica.com/ai/2025/08/deepmind-reveals-genie-3-world-model-that-creates-real-time-interactive-simulations/ "DeepMind reveals Genie 3 "world model" that creates real-time interactive simulations"]. ''Ars Technica''. 2025-08-05.</ref> On January 29, 2026, DeepMind released Project Genie to AI Ultra subscribers.<ref>Whitwam, Ryan. [https://arstechnica.com/google/2026/01/google-project-genie-lets-you-create-interactive-worlds-from-a-photo-or-prompt/ "Google Project Genie lets you create interactive worlds from a photo or prompt"]. ''Ars Technica''. 2026-01-29.</ref>

=== Robotics ===
Released in June 2023, RoboCat is an AI model that can control robotic arms. The model can adapt to new models of robotic arms, and to new types of tasks.<ref>Wiggers, Kyle. [https://techcrunch.com/2023/06/21/deepminds-robocat-learns-to-perform-a-range-of-robotics-tasks/ "DeepMind's RoboCat learns to perform a range of robotics tasks"]. ''TechCrunch''. 21 June 2023.</ref><ref>Cuthbertson, Anthony. [https://www.independent.co.uk/tech/google-deepmind-ai-robot-robocat-b2362892.html "Google's DeepMind unveils AI robot that can teach itself unsupervised"]. ''The Independent''. 23 June 2023.</ref> In March 2025, DeepMind launched two AI models, Gemini Robotics and Gemini Robotics-ER, aimed at improving how robots interact with the physical world<ref>Wiggers, Kyle. [https://techcrunch.com/2025/03/12/google-deepmind-unveils-new-ai-models-for-controlling-robots/ "Google DeepMind unveils new AI models for controlling robots"]. ''TechCrunch''. 2025-03-12.</ref> and released Gemini Robotics 1.5 in September 2025.<ref>Schreiner, Maximilian. [https://the-decoder.com/google-deepmind-taps-boston-dynamics-former-cto-to-build-the-android-of-robots/ "Google Deepmind taps Boston Dynamics' former CTO to build the 'Android' of robots"]. ''the decoder''. 2025-11-20.</ref>

=== Others ===

==== Football ====
DeepMind researchers have applied machine learning models to the sport of [[Association football|football]], often referred to as soccer in North America, modelling the behaviour of football players, including the goalkeeper, defenders, and strikers during different scenarios such as penalty kicks. The researchers used heat maps and cluster analysis to organize players based on their tendency to behave a certain way during the game when confronted with a decision on how to score or prevent the other team from scoring. 

The researchers mention that machine learning models could be used to democratize the football industry by automatically selecting interesting video clips of the game that serve as highlights. This can be done by searching videos for certain events, which is possible because video analysis is an established field of machine learning. This is also possible because of extensive sports analytics based on data including annotated passes or shots, sensors that capture data about the players movements many times over the course of a game, and game theory models.<ref>Tuyls, Karl. [https://www.deepmind.com/blog/advancing-sports-analytics-through-ai-research "Advancing sports analytics through AI research"]. ''DeepMind''. 7 May 2021.</ref><ref>Tuyls, Karl. [https://www.jair.org/index.php/jair/article/view/12505 "Game Plan: What AI can do for Football, and What Football can do for AI"]. ''Journal of Artificial Intelligence Research''. 6 May 2021.</ref>

==== Archaeology ====
Google has unveiled a new archaeology document program, named Ithaca after [[Homer's Ithaca|the Greek island]] in Homer's [[Odyssey]].<ref name=":1">Assael, Yannis. [https://deepmind.google/discover/blog/predicting-the-past-with-ithaca/ "Predicting the past with Ithaca"]. ''Google DeepMind''. 9 March 2022.</ref> This deep neural network helps researchers restore the empty text of damaged Greek documents, and to identify their date and geographical origin.<ref name=":2">Vincent, James. [https://www.theverge.com/2022/3/9/22968773/ai-machine-learning-ancient-inscriptions-texts-deepmind-ithaca-model "DeepMind's new AI model helps decipher, date, and locate ancient inscriptions"]. ''The Verge''. 9 March 2022.</ref> The work builds on another text analysis network that DeepMind released in 2019, named Pythia.<ref name=":2" /> Ithaca achieves 62% accuracy in restoring damaged texts and 71% location accuracy, and has a dating precision of 30 years.<ref name=":2" /> The authors claimed that the use of Ithaca by "expert historians" raised the accuracy of their work from 25 to 72 percent.<ref name=":1" /> However, [[Eleanor Dickey]] noted that this test was actually only made of students, saying that it wasn't clear how helpful Ithaca would be to "genuinely qualified editors".<ref name=":2" /> 

The team is working on extending the model to other ancient languages, including [[Demotic Egyptian language|Demotic]], [[Akkadian language|Akkadian]], [[Hebrew language|Hebrew]], and [[Mayan languages|Mayan]].<ref name=":1" />

==== Materials science ====
In November 2023, Google DeepMind announced an Open Source Graph Network for Materials Exploration (GNoME). The tool proposes millions of materials previously unknown to chemistry, including several hundred thousand stable crystalline structures, of which 736 had been experimentally produced by the Massachusetts Institute of Technology, at the time of the release.<ref>Merchant, Amil. "Scaling deep learning for materials discovery". ''Nature''. December 2023.</ref><ref>Kim, June. [https://www.technologyreview.com/2023/11/29/1084061/deepmind-ai-tool-for-new-materials-discovery/ "Google DeepMind's new AI tool helped create more than 700 new materials"]. ''MIT Technology Review''. 29 November 2023.</ref> However, according to [[Anthony Cheetham]], GNoME did not make "a useful, practical contribution to the experimental materials scientists."<ref name="404media">Koebler, Jason. [https://www.404media.co/google-says-it-discovered-millions-of-new-materials-with-ai-human-researchers/ "Is Google's AI Actually Discovering 'Millions of New Materials?'"]. ''[[404 Media]]''. April 11, 2024.</ref> A review article by Cheetham and Ram Seshadri were unable to identify any "strikingly novel" materials found by GNoME, with most being minor variants of already-known materials.<ref name="404media" /><ref>Cheetham, Anthony K.. "Artificial intelligence driving materials discovery? Perspective on the article: Scaling Deep Learning for Materials Discovery". ''[[Chemistry of Materials]]''.</ref>

=== Mathematics ===
====AlphaTensor====
In October 2022, DeepMind released [[AlphaTensor]], which used reinforcement learning techniques similar to those in AlphaGo, to find novel [[Matrix multiplication algorithm|algorithms for matrix multiplication]].<ref name=AlphaTensor1>Hutson, Matthew. [https://www.nature.com/articles/d41586-022-03166-w "DeepMind AI invents faster algorithms to solve tough maths puzzles"]. ''[[Nature (journal)''. 5 October 2022.</ref><ref name=AlphaTensor2>Heaven, Will Douglas. [https://www.technologyreview.com/2022/10/05/1060717/deepmind-uses-its-game-playing-ai-to-best-a-50-year-old-record-in-computer-science/ "DeepMind's game-playing AI has beaten a 50-year-old record in computer science"]. ''[[MIT Technology Review]]''. 5 October 2022.</ref> In the special case of multiplying two 4×4 matrices with [[integer]] entries, where only the evenness or oddness of the entries is recorded, AlphaTensor found an algorithm requiring only 47 distinct multiplications; the previous optimum, known since 1969, was the more general [[Strassen algorithm]], using 49 multiplications.<ref name="quantamag">Brubaker, Ben. [https://www.quantamagazine.org/ai-reveals-new-possibilities-in-matrix-multiplication-20221123/ "AI Reveals New Possibilities in Matrix Multiplication"]. ''Quanta Magazine''. November 2022.</ref> Computer scientist Josh Alman described AlphaTensor as "a proof of concept for something that could become a breakthrough", while [[Virginia Vassilevska Williams|Vassilevska Williams]] called it "a little overhyped"<ref name="quantamag" /> despite also acknowledging its basis in reinforcement learning as "something completely different" from previous approaches.<ref name=AlphaTensor2 />

====AlphaGeometry====
''Main article: [[AlphaGeometry]]''

AlphaGeometry is a [[neuro-symbolic AI]] that was able to solve 25 out of 30 geometry problems of the [[International Mathematical Olympiad]], a performance comparable to that of a gold medalist.<ref name=":3">Zia, Tehseen. [https://www.unite.ai/alphageometry-how-deepminds-ai-masters-geometry-problems-at-olympian-levels/ "AlphaGeometry: DeepMind's AI Masters Geometry Problems at Olympiad Levels"]. ''Unite.ai''. January 24, 2024.</ref>

Traditional geometry programs are [[Symbolic artificial intelligence|symbolic engines]] that rely exclusively on human-coded [[Rule-based system|rules]] to generate rigorous proofs, which makes them lack flexibility in unusual situations. AlphaGeometry combines such a symbolic engine with a specialized [[large language model]] trained on [[synthetic data]] of geometrical proofs. When the symbolic engine doesn't manage to find a formal and rigorous proof on its own, it solicits the large language model, which suggests a geometrical construct to move forward. However, it is unclear how applicable this method is to other domains of mathematics or reasoning, because symbolic engines rely on domain-specific rules and because of the need for synthetic data.<ref name=":3" />

====AlphaProof====
AlphaProof is an AI model, which couples a pre-trained language model with the AlphaZero reinforcement learning algorithm. AlphaZero has previously taught itself how to master games. The pre-trained language model used in this combination is the [[Fine-tuning (deep learning)|fine-tuning]] of a [[Gemini (language model)|Gemini]] model to automatically translate natural language problem statements into formal statements, creating a large library of formal problems of varying difficulty. For this purpose, mathematical statements are defined in the formal language [[Lean (proof assistant)|Lean]]. At the 2024 International Mathematical Olympiad, AlphaProof together with an adapted version of AlphaGeometry have reached the same level of solving problems in the combined categories as a silver medalist in that competition for the first time.<ref name="NYT">Roberts, Siobhan. [https://www.nytimes.com/2024/07/25/science/ai-math-alphaproof-deepmind.html/ "AI achieves silver-medal standard solving International Mathematical Olympiad problems"]. ''[[The New York Times]]''. July 25, 2024.</ref><ref name="IMO">AlphaProof and AlphaGeometry teams. [https://deepmind.google/discover/blog/ai-solves-imo-problems-at-silver-medal-level/ "AI achieves silver-medal standard solving International Mathematical Olympiad problems"]. ''deepmind.google''. July 25, 2024.</ref>

===AlphaDev===
{{Main article|AlphaDev}}

In June 2023, Deepmind announced that [[AlphaDev]], which searches for improved computer science algorithms using [[reinforcement learning]], discovered a more efficient way of coding a sorting algorithm and a hashing algorithm. The new sorting algorithm was 70% faster for shorter sequences and 1.7% faster for sequences exceeding 250,000 elements, and the new hashing algorithm was 30% faster in some cases. The sorting algorithm was accepted into the [[C++ Standard Library]] [[sorting algorithm]]s, and was the first change to those algorithms in more than a decade and the first update to involve an algorithm discovered using AI.<ref name="mit">Heaven, Will Douglas. [https://www.technologyreview.com/2023/06/07/1074184/google-deepmind-game-ai-alphadev-algorithm-code-faster/ "Google DeepMind's game-playing AI just found another way to make code faster"]. [[MIT Technology Review]]. June 7, 2023.</ref> The hashing algorithm was released to an opensource library.<ref>Mankowitz, Daniel J.. [https://deepmind.google/discover/blog/alphadev-discovers-faster-sorting-algorithms "AlphaDev discovers faster sorting algorithms"]. ''DeepMind Blog''. 7 June 2023.</ref> Google estimates that these two algorithms are used trillions of times every day.<ref>Sparkes, Matthew. [https://www.newscientist.com/article/2376512-deepmind-ais-new-way-to-sort-objects-could-speed-up-global-computing/ "DeepMind AI's new way to sort objects could speed up global computing"]. ''New Scientist''. 7 June 2023.</ref>

=== AlphaEvolve ===
{{Main article|AlphaEvolve}}

In May 2025, Google DeepMind unveiled [[AlphaEvolve]], an [[Evolutionary computation|evolutionary]] coding agent using LLMs like Gemini to design optimized algorithms. AlphaEvolve begins each optimization process with an initial algorithm and metrics to evaluate the quality of a solution. At each step, it uses the LLM to generate variations of the algorithms or combine them, and selects the best candidates for further iterations.<ref name=":7">Tardif, Antoine. [https://www.unite.ai/alphaevolve-google-deepminds-groundbreaking-step-toward-agi/ "AlphaEvolve: Google DeepMind's Groundbreaking Step Toward AGI"]. ''Unite.AI''. 2025-05-17.</ref>

AlphaEvolve has made several algorithmic discoveries, including in matrix multiplication. According to Google, when tested on 50 open [[mathematical problems]], AlphaEvolve was able to match the efficiency of state-of-the-art algorithms in 75% of cases, and discovered improved solutions 20% of the time, such as with the [[kissing number problem]] in 11 dimensions. It also developed a new heuristic for data centre scheduling, recovering on average 0.7% of Google's worldwide compute resources.<ref name=":7" />

=== Chip design ===
''Main article: [[AlphaChip (controversy)]]''

AlphaChip is a [[reinforcement learning]]-based neural architecture that guides the task of chip [[Placement (electronic design automation)|placement]]. DeepMind claimed that the technique reduced the time needed to create chip layouts from weeks to hours. According to the company, its chip designs were used in every [[Tensor Processing Unit]] (TPU) iteration since 2020.<ref>Ghoshal, Abhimanyu. [https://newatlas.com/ai-humanoids/3-mind-blowing-ways-ai-chip-design-singularity/ "Singularity alert: AIs are already designing their own chips"]. ''New Atlas''. 2024-11-30.</ref><ref>Shilov, Anton. [https://www.tomshardware.com/tech-industry/google-unveils-alphachip-ai-assisted-chip-design-technology-chip-layout-as-a-game-for-a-computer "Google unveils AlphaChip AI-assisted chip design technology — chip layout as a game for a computer"]. ''Tom's Hardware''. 2024-09-28.</ref> Multiple independent researchers remained unconvinced, citing a lack of direct public benchmarks and independent proof of its claimed superiority over existing commercial chip design tools.<ref name="CACM">Markov, Igor L.. "Reevaluating Google's Reinforcement Learning for IC Macro Placement". ''Communications of the ACM''. 2024-10-23.</ref> The TPU chips were co-designed with [[Broadcom]].<ref>Mann, Tobias. [https://www.theregister.com/2023/09/22/google_broadcom_tpus/ "For your info, Broadcom helped Google make those TPU chips"]. ''The Register''. 22 September 2023.</ref><ref>[https://www.reuters.com/technology/google-discussed-dropping-broadcom-ai-chips-supplier-the-information-2023-09-21/ "Google expects no change in its relationship with AI chip supplier Broadcom"]. ''Reuters''. 2023-09-21.</ref> [[Communications of the ACM]] noted that despite substantial publicity, DeepMind had not provided the comparative benchmarks long requested by experts, leaving some skepticism in the field.<ref name="cacm2024">Halper, Mark. [https://cacm.acm.org/news/updates-spark-uproar/ "Updates Spark Uproar"]. ''Communications of the ACM''. 2024-11-04.</ref> Similarly, [[New Scientist]] reported that while Google claims AlphaChip has produced “superhuman” chip layouts now used in production, external specialists called for transparent performance data to substantiate these assertions and enable fair comparisons with current state-of-the-art methods.<ref name="newscien2024">Hsu, Jeremy. [https://www.newscientist.com/article/2450402-google-says-its-ai-designs-chips-better-than-humans-experts-disagree/ "Google says its AI designs chips better than humans - Experts disagree"]. ''New Scientist''. 2024-10-14.</ref>

=== Safety ===
Google Research released a paper in 2016 regarding [[AI safety]] and avoiding undesirable behaviour during the AI learning process.<ref>Amodei, Dario. "Concrete Problems in AI Safety". 21 June 2016.</ref> In 2017 DeepMind released GridWorld, an open-source testbed for evaluating whether an algorithm learns to disable its [[kill switch]] or otherwise exhibits certain undesirable behaviours.<ref>Kahn, Jeremy. [https://www.bloomberg.com/news/articles/2017-12-11/deepmind-has-simple-tests-that-might-prevent-elon-musk-s-ai-apocalypse "DeepMind Has Simple Tests That Might Prevent Elon Musk's AI Apocalypse"]. ''Bloomberg.com''. 11 December 2017.</ref><ref>Meyer, David. [http://fortune.com/2017/12/12/alphabet-deepmind-ai-safety-musk-games/ "Alphabet's DeepMind Is Using Games to Discover If Artificial Intelligence Can Break Free and Kill Us All"]. ''Fortune''. 12 December 2017.</ref>

The Robot Constitution is a security ruleset part of AutoRT set by DeepMind in January 2024 for its AI products. The rules are inspired by [[Asimov]]'s [[Three Laws of Robotics]]. The rules are applied to the underlying [[Large language model|large language models]] of the helper robots.<ref>Khalid, Amrita. [https://www.theverge.com/2024/1/4/24025535/google-ai-robot-constitution-autort-deepmind-three-laws "Google wrote a "Robot Constitution" to make sure its new AI droids won't kill us"]. ''The Verge''. January 4, 2024.</ref><ref>Chowdhury, Hasan. [https://www.businessinsider.com/google-deepmind-rules-ai-robots-safer-in-your-home-2024-1 "Google DeepMind has new rules to make sure AI robots behave when tidying your home"]. ''Business Insider''.</ref><ref>Gowran, Leigh Mc. [https://www.siliconrepublic.com/machines/google-deepmind-robot-constitution-real-world-safety "DeepMind is training robots for real-world activities"]. ''Silicon Republic''. January 5, 2024.</ref><ref>[https://www.tomsguide.com/news/google-is-using-ai-to-teach-robots-household-chores-heres-the-result "Google's DeepMind is using AI to teach robots household chores — here's the result"]. ''Tom's Guide''. January 5, 2024.</ref><ref>[https://uk.pcmag.com/ai/150336/google-taps-asimovs-three-laws-of-robotics-for-real-robot-safety "Google Taps Asimov's Three Laws of Robotics for Real Robot Safety"]. ''PCMag UK''. January 4, 2024.</ref><ref>Citation needed.</ref>
Rule number 1 is a robot “may not injure a human being”.
=== Weather prediction ===
Google DeepMind developed an AI-based weather prediction system called Weather Lab, which significantly improved tropical cyclone forecasting. Launched in mid-2025, this model utilized stochastic neural networks trained on 45 years of global weather and cyclone data, enabling it to predict cyclone formation, track, intensity, and structure with multiple probabilistic forecasts up to 15 days in advance. During the 2025 Atlantic hurricane season, DeepMind's Weather Lab outperformed traditional physics-based models, including the [[US National Weather Service]]'s Global Forecast System, in both track and intensity predictions, earning notable recognition from meteorologists and aiding hurricane forecasting efforts by the [[US National Hurricane Center]]. This marked a substantial advancement in weather modeling, demonstrating the potential for AI to enhance the speed and accuracy of severe weather forecasts.<ref>Berger, Eric. [https://arstechnica.com/science/2025/11/googles-new-weather-model-impressed-during-its-first-hurricane-season/ "Google's new weather model impressed during its first hurricane season"]. ''Ars Technica''. 2025-11-04.</ref>

=== Miscellaneous contributions to Google ===
DeepMind (alongside other Alphabet AI researchers) assists [[Google Play]]'s personalized app recommendations.<ref name="cnbc money"/> DeepMind has also collaborated with the [[Android (operating system)|Android]] team at [[Google]] for the creation of two new features which were made available to people with devices running Android Pie, the ninth installment of Google's mobile operating system. These features, Adaptive Battery and Adaptive Brightness, use machine learning to conserve energy and make devices running the operating system easier to use. It is the first time DeepMind has used these techniques on such a small scale, with typical machine learning applications requiring orders of magnitude more computing power.<ref>[https://deepmind.com/blog/deepmind-meet-android/ "DeepMind, meet Android"]. ''DeepMind Blog''. 14 May 2024. 8 May 2018.</ref>

== DeepMind Health ==
In July 2016, a collaboration between DeepMind and [[Moorfields Eye Hospital]] was announced to develop [[Artificial intelligence in healthcare|AI applications for healthcare]].<ref>Baraniuk, Chris. [https://www.bbc.com/news/technology-36713308 "Google's DeepMind to peek at NHS eye scans for disease analysis"]. BBC. 6 July 2016.</ref> DeepMind would be applied to the analysis of [[Data anonymization|anonymised]] eye scans, searching for early signs of diseases leading to [[blindness]].

In August 2016, a research programme with [[University College Hospital|University College London Hospital]] was announced with the aim of developing an algorithm that can automatically differentiate between healthy and cancerous tissues in head and neck areas.<ref>Baraniuk, Chris. [https://www.bbc.co.uk/news/technology-37230806 "Google DeepMind targets NHS head and neck cancer treatment"]. BBC. 31 August 2016.</ref>

There are also projects with the [[Royal Free London NHS Foundation Trust]] and [[Imperial College Healthcare NHS Trust]] to develop new clinical mobile apps linked to [[electronic patient record]]s.<ref>[http://www.itpro.co.uk/public-sector/27833/deepmind-announces-second-nhs-partnership "DeepMind announces second NHS partnership"]. IT Pro. 23 December 2016.</ref> Staff at the [[Royal Free Hospital]] were reported as saying in December 2017 that access to patient data through the app had saved a 'huge amount of time' and made a 'phenomenal' difference to the management of patients with acute kidney injury. Test result data is sent to staff's mobile phones and alerts them to changes in the patient's condition. It also enables staff to see if someone else has responded, and to show patients their results in visual form.<ref>[https://www.digitalhealth.net/2017/12/google-deepmind-streams-royal-free/ "Google DeepMind's Streams technology branded 'phenomenal'"]. Digital Health. 4 December 2017.</ref><ref>[https://www.bmj.com/bmj/section-pdf/966505?path=/bmj/360/8141/This_Week.full.pdf "A dedicated WhatsApp for clinicians"]. ''the bmj''. 17 February 2018.</ref>

In November 2017, DeepMind announced a research partnership with the [[Cancer Research UK]] Centre at Imperial College London with the goal of improving breast cancer detection by applying machine learning to mammography.<ref>David, Eric. [https://siliconangle.com/blog/2017/11/24/google-deepmind-announces-new-research-partnership-fight-breast-cancer-ai/ "Google DeepMind announces new research partnership to fight breast cancer with AI"]. ''Silicon Angle''. 24 November 2017.</ref> Additionally, in February 2018, DeepMind announced it was working with the [[United States Department of Veterans Affairs|U.S. Department of Veterans Affairs]] in an attempt to use machine learning to predict the onset of acute kidney injury in patients, and also more broadly the general deterioration of patients during a hospital stay so that doctors and nurses can more quickly treat patients in need.<ref>Frank, ISG, Blair Hanley. [https://venturebeat.com/2018/02/22/googles-deepmind-wants-ai-to-spot-kidney-injuries/ "Google's DeepMind wants AI to spot kidney injuries"]. ''VentureBeat''. 22 February 2018.</ref>

DeepMind developed an app called Streams, which sends alerts to doctors about patients at risk of acute kidney injury.<ref>Evenstad, Lis. [https://www.computerweekly.com/news/252443164/DeepMind-Health-must-be-transparent-to-gain-public-trust-review-finds "DeepMind Health must be transparent to gain public trust, review finds"]. ''ComputerWeekly.com''. 15 June 2018.</ref> On 13 November 2018, DeepMind announced that its health division and the Streams app would be absorbed into [[Google Health]].<ref>Vincent, James. [https://www.theverge.com/2018/11/13/18091774/google-deepmind-health-absorbing-streams-team-ai-assistant-nurse-doctor "Google is absorbing DeepMind's health care unit to create an 'AI assistant for nurses and doctors'"]. ''The Verge''. 13 November 2018.</ref> Privacy advocates said the announcement betrayed patient trust and appeared to contradict previous statements by DeepMind that patient data would not be connected to Google accounts or services.<ref>Hern, Alex. [https://www.theguardian.com/technology/2018/nov/14/google-betrays-patient-trust-deepmind-healthcare-move "Google 'betrays patient trust' with DeepMind Health move"]. ''The Guardian''. 14 November 2018.</ref><ref>Stokel-Walker, Chris. [https://www.wired.co.uk/article/google-deepmind-nhs-health-data "Why Google consuming DeepMind Health is scaring privacy experts"]. ''Wired''. 14 November 2018.</ref> A spokesman for DeepMind said that patient data would still be kept separate from Google services or projects.<ref>Murphy, Margi. [https://www.telegraph.co.uk/technology/2018/11/14/deepmind-boss-defends-controversial-google-health-deal/ "DeepMind boss defends controversial Google health deal"]. ''The Telegraph''. 14 November 2018.{{cbignore}}</ref>

=== NHS data-sharing controversy ===
In April 2016, ''[[New Scientist]]'' obtained a copy of a [[data sharing]] agreement between DeepMind and the [[Royal Free London NHS Foundation Trust]]. The latter operates three London hospitals where an estimated 1.6 million patients are treated annually. The agreement shows DeepMind Health had access to admissions, discharge and transfer data, accident and emergency, pathology and radiology, and critical care at these hospitals. This included personal details such as whether patients had been diagnosed with [[HIV/AIDS|HIV]], suffered from [[major depressive disorder|depression]] or had ever undergone an [[abortion]] in order to conduct research to seek better outcomes in various health conditions.<ref>Hodson, Hal. [https://www.newscientist.com/article/2086454-revealed-google-ai-has-access-to-huge-haul-of-nhs-patient-data "Revealed: Google AI has access to huge haul of NHS patient data"]. ''[[New Scientist]]''. 29 April 2016.</ref><ref>[https://www.newscientist.com/article/mg23030722-900-big-data-if-theres-nothing-to-hide-why-be-secretive/ "Leader: If Google has nothing to hide about NHS data, why so secretive?"]. ''[[New Scientist]]''. 4 May 2016.</ref>

A complaint was filed to the [[Information Commissioner's Office]] (ICO), arguing that the data should be pseudonymised and encrypted.<ref>Donnelly, Caroline. [http://www.computerweekly.com/news/450296175/ICO-probes-Google-DeepMind-patient-data-sharing-deal-with-NHS-Hospital-Trust "ICO probes Google DeepMind patient data-sharing deal with NHS Hospital Trust"]. ''[[Computer Weekly]]''. 12 May 2016.</ref> In May 2016, ''New Scientist'' published a further article claiming that the project had failed to secure approval from the Confidentiality Advisory Group of the [[Medicines and Healthcare products Regulatory Agency]].<ref>Hodson, Hal. [https://www.newscientist.com/article/2088056-exclusive-googles-nhs-deal/ "Did Google's NHS patient data deal need ethical approval?"]. ''[[New Scientist]]''. 25 May 2016.</ref>

In 2017, the ICO concluded a year-long investigation that focused on how the Royal Free NHS Foundation Trust tested the app, Streams, in late 2015 and 2016.<ref>[https://ico.org.uk/about-the-ico/news-and-events/news-and-blogs/2017/07/royal-free-google-deepmind-trial-failed-to-comply-with-data-protection-law/ "Royal Free - Google DeepMind trial failed to comply with data protection law"]. ''ico.org.uk''. 17 August 2017.</ref> The ICO found that the Royal Free failed to comply with the Data Protection Act when it provided patient details to DeepMind, and found several shortcomings in how the data was handled, including that patients were not adequately informed that their data would be used as part of the test. DeepMind published its thoughts<ref>Suleyman, Mustafa. [https://deepmind.com/blog/ico-royal-free/ "The Information Commissioner, the Royal Free, and what we've learned"]. ''DeepMind''. 3 July 2017.</ref> on the investigation in July 2017, saying "we need to do better" and highlighting several activities and initiatives they had initiated for transparency, oversight and engagement. This included developing a patient and public involvement strategy<ref>[https://deepmind.com/applied/deepmind-health/patients/ "For Patients"]. ''DeepMind''.</ref> and being transparent in its partnerships.

In May 2017, ''Sky News'' published a leaked letter from the National Data Guardian, Dame [[Fiona Caldicott]], revealing that in her "considered opinion" the data-sharing agreement between DeepMind and the Royal Free took place on an "inappropriate legal basis".<ref>Martin, Alexander J. [http://news.sky.com/story/google-received-16-million-nhs-patients-data-on-an-inappropriate-legal-basis-10879142/ "Google received 1.6 million NHS patients' data on an 'inappropriate legal basis'"]. ''[[Sky News]]''. 15 May 2017.</ref> The Information Commissioner's Office ruled in July 2017 that the Royal Free hospital failed to comply with the Data Protection Act when it handed over personal data of 1.6 million patients to DeepMind.<ref>Hern, Alex. [https://www.theguardian.com/technology/2017/jul/03/google-deepmind-16m-patient-royal-free-deal-data-protection-act "Royal Free breached UK data law in 1.6m patient deal with Google's DeepMind"]. ''The Guardian''. 3 July 2017.</ref>

== DeepMind Ethics and Society ==
In October 2017, DeepMind announced a new research unit, DeepMind Ethics & Society.<ref>Legassick, Sean. [https://deepmind.com/blog/why-we-launched-deepmind-ethics-society/ "Why we launched DeepMind Ethics & Society"]. ''DeepMind Blog''. October 3, 2017.</ref> Their goal is to fund external research of the following themes: privacy, transparency, and fairness; economic impacts; governance and accountability; managing AI risk; AI morality and values; and how AI can address the world's challenges. As a result, the team hopes to further understand the ethical implications of AI and aid society to seeing AI can be beneficial.<ref>Temperton, James. [https://www.wired.co.uk/article/deepmind-ethics-and-society-artificial-intelligence "DeepMind's new AI ethics unit is the company's next big move"]. ''Wired (UK)''.</ref>

This new subdivision of DeepMind is a completely separate unit from the partnership of leading companies using AI, academia, civil society organizations and nonprofits of the name [[Partnership on AI|Partnership on Artificial Intelligence to Benefit People and Society]] of which DeepMind is also a part. The DeepMind Ethics and Society board is also distinct from the mooted AI Ethics Board that [[Google]] originally agreed to form when acquiring DeepMind.<ref name=":9">Hern, Alex. [https://www.theguardian.com/technology/2017/oct/04/google-deepmind-ai-artificial-intelligence-ethics-group-problems "DeepMind announces ethics group to focus on problems of AI"]. ''[[The Guardian]]''. 4 October 2017.</ref>
==DeepMind Professors of machine learning==
DeepMind sponsors three [[Academic ranks in the United Kingdom|chairs]] of machine learning:

# At the [[University of Cambridge]], held by [[Neil Lawrence]],<ref>[https://www.cam.ac.uk/research/news/cambridge-appoints-first-deepmind-professor-of-machine-learning "Cambridge appoints first DeepMind Professor of Machine Learning"]. ''University of Cambridge''. 18 September 2019.</ref> in the [[Department of Computer Science and Technology, University of Cambridge|Department of Computer Science and Technology]],
# At the [[University of Oxford]], held by [[Michael Bronstein]],<ref>[http://www.cs.ox.ac.uk/news/1862-full.html "DeepMind funds new post at Oxford University – the DeepMind Professorship of Artificial Intelligence"]. ''Department of Computer Science''.</ref> in the [[Department of Computer Science, University of Oxford|Department of Computer Science]], and
# At the [[University College London]], held by Marc Deisenroth,<ref>[https://www.ucl.ac.uk/news/2019/nov/deepmind-renews-its-commitment-ucl "DeepMind renews its commitment to UCL"]. ''University College London''. 29 March 2021.</ref> in the Department of Computer Science.

==See also==
* [[Anthropic]]
* [[Cohere]]
* [[Glossary of artificial intelligence]]
* [[Imagen (text-to-image model)|Imagen]]
* [[Model Context Protocol]]
* [[Robot Constitution]]

== References ==
<references>

<ref name="nature2015">Mnih, Volodymyr. "Human-level control through deep reinforcement learning". ''Nature''. 26 February 2015.</ref>

</references>

== External links ==
* {{Official website}}
* [https://github.com/google-deepmind GitHub Repositories] 

{{Google AI}}
{{Google LLC}}
{{Generative AI}}
{{Existential risk from artificial intelligence}}
{{authority control}}
{{Subject bar|auto=yes|portal1=Companies|portal2=Technology}}

[[Category:2010 establishments in England]]
[[Category:Artificial intelligence laboratories]]
[[Category:British companies established in 2010]]
[[Category:Deep learning]]
[[Category:Game artificial intelligence]]
[[Category:Google acquisitions]]
[[Category:Applied machine learning]]
[[Category:British subsidiaries of foreign companies]]
[[Category:Alphabet Inc. subsidiaries]]
[[Category:2014 mergers and acquisitions]]
[[Category:Google DeepMind| ]]
[[Category:Information technology companies of the United Kingdom]]