imported>Cewbot：清理跨語言連結人工智能伦理成為內部連結:編輯摘要的紅色連結經繁簡轉換後存在，非bot錯誤編輯 (本次機械人作業已完成18.3%)

2026-05-26T12:41:17Z

清理跨語言連結人工智能伦理成為內部連結:編輯摘要的紅色連結經繁簡轉換後存在，非bot錯誤編輯 (本次機械人作業已完成18.3%)

新页面

{{expand|time=2016-01-29T03:20:29+00:00}}
{{TA
|G1=IT
}}{{Infobox company
| name = Google DeepMind
| logo = DeepMind new logo.svg
| logo_size = 250px
| image =
| image_caption = 位于伦敦国王十字区的总部
| former_name = DeepMind科技 (2010–2023)
| founded = {{start date and age|df=y|2010|9|23}}<ref>{{cite web|url=https://beta.companieshouse.gov.uk/company/07386350|title=DeepMind Technologies Limited – Overview (free company information from Companies House)|work=[[Companies House]]|access-date=13 March 2016|archive-date=2016-03-14|archive-url=https://web.archive.org/web/20160314013741/https://beta.companieshouse.gov.uk/company/07386350|dead-url=no}}</ref>
| location = [[英国]][[伦敦]]<ref>{{Cite web |title=King's Cross - S2 Building - SES Engineering Services |url=https://www.ses-ltd.co.uk/case-study/kings-cross-s2-building/ |access-date=2022-07-14 |website=www.ses-ltd.co.uk |language=en |archive-date=2021-06-20 |archive-url=https://web.archive.org/web/20210620093626/https://www.ses-ltd.co.uk/case-study/kings-cross-s2-building/ |dead-url=no }}</ref>
| founders = {{plain list|
*[[杰米斯·哈萨比斯]]
*{{le|謝恩·列格|Shane Legg}}、
*{{le|穆斯塔法·蘇萊曼|Mustafa Suleyman}}}}
| key_people = {{Unbulleted list
|[[杰米斯·哈萨比斯]] ([[首席执行官|CEO]])
|{{le|莱拉·易卜拉欣|Lila Ibrahim}}}} ([[首席运营官|COO]])
| industry = [[人工智能]]
| parent = {{Unbulleted list|class=nowrap|[[Google]] (2014–2015, 2023–至今)<ref>{{Cite web |title=Announcing Google DeepMind |url=https://deepmind.google/discover/blog/announcing-google-deepmind/ |website=Google DeepMind |date=2024-12-17 |language=en |access-date=2025-01-14 |archive-date=2025-01-30 |archive-url=https://web.archive.org/web/20250130071513/https://deepmind.google/discover/blog/announcing-google-deepmind/ |dead-url=no }}</ref>|[[Alphabet]] (2015–2023)}}
| type = [[子公司]]
| num_employees = 1,000+ (2022)<ref>{{cite web |last1=Kahn |first1=Jeremy |title=Former DeepMind employees say A.I. company mishandled serious sexual assault complaints |url=https://fortune.com/2022/03/30/deepmind-sexual-harassment-allegations-mishandled-ft/ |website=[[Fortune (magazine)|Fortune]] |date=30 March 2022 |archive-url=https://web.archive.org/web/20220624000021/https://fortune.com/2022/03/30/deepmind-sexual-harassment-allegations-mishandled-ft/ |archive-date=24 June 2022 |url-status=live}}</ref>
| products = [[AlphaGo]], [[AlphaStar]], [[AlphaFold]], [[AlphaZero]]
| website = {{URL|https://deepmind.google/}}
}}

'''Google DeepMind'''（{{langx|zh|深智科技}}），前称'''DeepMind科技'''（DeepMind Technologies Limited）'''，'''是一家[[英国]]的[[人工智能]]公司。公司创建于2010年，在2014年被[[Google公司收購列表|Google收购]]。

2022年5月12日，DeepMind發布[[通用人工智慧]]模型{{le|Gato|Gato (DeepMind)}}<ref>{{Cite web|title=超越人類智慧水準的 AI 即將來臨，DeepMind 推出全新通用 AI 代理程式 Gato|url=https://technews.tw/2022/05/23/deepmind-gato/|access-date=2022-05-24|work=TechNews 科技新報|language=zh-TW|archive-date=2022-06-24|archive-url=https://web.archive.org/web/20220624194828/https://technews.tw/2022/05/23/deepmind-gato/|dead-url=no}}</ref><ref>{{Cite web|title=A Generalist Agent|url=https://www.deepmind.com/publications/a-generalist-agent|access-date=2022-05-24|work=www.deepmind.com|language=en|archive-date=2022-08-02|archive-url=https://web.archive.org/web/20220802000307/https://www.deepmind.com/publications/a-generalist-agent|dead-url=no}}</ref>。

== 历史 ==

=== 2010年到2014年 ===
在2010年由[[杰米斯·哈萨比斯]]，{{le|謝恩·列格|Shane Legg}}和{{le|穆斯塔法·蘇萊曼|Mustafa Suleyman}}成立[[创业公司]]。<ref>{{cite news | url=http://www.bloomberg.com/news/2014-01-27/google-buys-u-k-artificial-intelligence-company-deepmind.html | work=Bloomberg | title=Google Buys U.K. Artificial Intelligence Company DeepMind | date=27 January 2014 | accessdate=13 November 2014 | archive-date=2015-01-11 | archive-url=https://web.archive.org/web/20150111012232/http://www.bloomberg.com/news/2014-01-27/google-buys-u-k-artificial-intelligence-company-deepmind.html | dead-url=no }}</ref><ref>{{cite news | url=http://www.ft.com/cms/s/0/dfedc62e-874e-11e3-9c5c-00144feab7de.html | title=Google makes £400m move in quest for artificial intelligence | newspaper=Financial Times | date=27 January 2014 | accessdate=13 November 2014 | archive-date=2016-05-14 | archive-url=https://web.archive.org/web/20160514064825/http://www.ft.com/cms/s/0/dfedc62e-874e-11e3-9c5c-00144feab7de.html | dead-url=no }}</ref>Hassabis和Legg最初在[[伦敦大学学院]]的{{link-en|生命科学系|UCL Faculty of Life Sciences}}的盖茨比计算神经科学小组里相识。<ref>{{cite web|title=Demis Hassabis: 15 facts about the DeepMind Technologies founder|url=http://www.theguardian.com/technology/shortcuts/2014/jan/28/demis-hassabis-15-facts-deepmind-technologies-founder-google|publisher=The Guardian|accessdate=12 October 2014|archive-date=2016-02-23|archive-url=https://web.archive.org/web/20160223051427/http://www.theguardian.com/technology/shortcuts/2014/jan/28/demis-hassabis-15-facts-deepmind-technologies-founder-google|dead-url=no}}</ref>

之后[[維港投資]]和{{le|Founders Fund|Founders Fund}}两家[[风险投资|风险投资公司]]对DeepMind进行了投资<ref>{{Cite news|title=DeepMind buy heralds rise of the machines|url=https://www.ft.com/content/b09dbd40-876a-11e3-9c5c-00144feab7de|newspaper=Financial Times|date=2014-01-27|accessdate=2025-01-14|first=Robert|last=Cookson|archive-date=2018-02-16|archive-url=https://web.archive.org/web/20180216084619/https://www.ft.com/content/b09dbd40-876a-11e3-9c5c-00144feab7de|dead-url=no}}</ref>，除此之外还有企业家{{le|史考特·班尼斯特|Scott Banister}}和[[伊隆·马斯克]]的投资。<ref>{{cite web|title=DeepMind Technologies Investors|url=https://angel.co/deepmind-technologies-limited|accessdate=12 October 2014|archive-date=2015-03-17|archive-url=https://web.archive.org/web/20150317024831/https://angel.co/deepmind-technologies-limited|dead-url=no}}</ref><ref>{{cite web|url=http://www.ibtimes.co.uk/elon-musk-artificial-intelligence-potentially-more-dangerous-nukes-1459710|title=Elon Musk: Artificial Intelligence 'Potentially More Dangerous Than Nukes'|first=Anthony|last=Cuthbertson|date=|work=International Business Times UK|accessdate=2016-01-29|archive-date=2016-02-03|archive-url=https://web.archive.org/web/20160203082226/http://www.ibtimes.co.uk/elon-musk-artificial-intelligence-potentially-more-dangerous-nukes-1459710|dead-url=no}}</ref>{{le|Jaan Tallinn|Jaan Tallinn}}是公司早期的投资者和顾问。<ref>{{cite web|title=Recode.net – DeepMind Technologies Acquisition|url=http://recode.net/2014/01/26/exclusive-google-to-buy-artificial-intelligence-startup-deepmind-for-400m/|accessdate=27 January 2014|archive-date=2016-01-27|archive-url=https://web.archive.org/web/20160127082941/http://recode.net/2014/01/26/exclusive-google-to-buy-artificial-intelligence-startup-deepmind-for-400m/|dead-url=no}}</ref>在2014年，DeepMind荣获了[[剑桥大学]]计算机实验室的“年度公司”奖项。<ref>{{cite web|title=Hall of Fame Awards: To celebrate the success of companies founded by Computer Laboratory graduates.|url=https://www.cl.cam.ac.uk/ring/awards.html|publisher=University of Cambridge|accessdate=12 October 2014|archive-date=2016-02-03|archive-url=https://web.archive.org/web/20160203112759/https://www.cl.cam.ac.uk/ring/awards.html|dead-url=no}}</ref>

DeepMind创造了一个以人类的方式学习如何玩[[电子游戏]]的[[人工神经网络]]<ref name="arxiv medium">{{cite web|title=The Last AI Breakthrough DeepMind Made Before Google Bought It|publisher=The Physics [[arXiv]] Blog|url=https://medium.com/the-physics-arxiv-blog/the-last-ai-breakthrough-deepmind-made-before-google-bought-it-for-400m-7952031ee5e1|accessdate=12 October 2014|archive-date=2019-05-01|archive-url=https://web.archive.org/web/20190501121110/https://medium.com/the-physics-arxiv-blog/the-last-ai-breakthrough-deepmind-made-before-google-bought-it-for-400m-7952031ee5e1|dead-url=no}}</ref>，并且这个神经网络可以接入一个外部的存储器，就像一个传统的[[图灵机]]一样，使得一台电脑可以模拟人类的[[短期记忆]]。<ref>{{Cite web |title=Best of 2014: Google's Secretive DeepMind Startup Unveils a "Neural Turing Machine" |url=http://www.technologyreview.com/view/533741/best-of-2014-googles-secretive-deepmind-startup-unveils-a-neural-turing-machine/ |website=MIT Technology Review |url-status=dead |archive-url=https://web.archive.org/web/20151204081728/http://www.technologyreview.com/view/533741/best-of-2014-googles-secretive-deepmind-startup-unveils-a-neural-turing-machine/ |archive-date=2015-12-04 |access-date=2025-01-14}}</ref>

=== 被Google收购 ===
2014年1月26日，[[Google]]宣布已经同意收购DeepMind科技。<ref>{{cite web|title=Google to buy artificial intelligence company DeepMind|url=http://www.reuters.com/article/2014/01/27/google-deepmind-idUSL2N0L102A20140127|date=26 January 2014|publisher=Reuters|accessdate=12 October 2014|archive-date=2015-09-24|archive-url=https://web.archive.org/web/20150924192908/http://www.reuters.com/article/2014/01/27/google-deepmind-idUSL2N0L102A20140127|dead-url=no}}</ref>这次收购发生在2013年[[Facebook]]与DeepMind科技结束谈判之后。<ref>{{cite web |title=Google beats Facebook for Acquisition of DeepMind Technologies |url=https://www.theinformation.com/Google-beat-Facebook-For-DeepMind-Creates-Ethics-Board |accessdate=27 January 2014 |archive-date=2014-01-31 |archive-url=https://web.archive.org/web/20140131121719/https://www.theinformation.com/Google-beat-Facebook-For-DeepMind-Creates-Ethics-Board |dead-url=no }}</ref>

这次收购的价格估计在4亿[[美元]]<ref name="econ2015"/>到5亿[[欧元]]之间。<ref>{{cite news |title=Google Acquires UK AI startup Deepmind |url=http://www.theguardian.com/technology/2014/jan/27/google-acquires-uk-artificial-intelligence-startup-deepmind |newspaper=The Guardian |accessdate=27 January 2014 |archive-date=2016-02-05 |archive-url=https://web.archive.org/web/20160205012245/http://www.theguardian.com/technology/2014/jan/27/google-acquires-uk-artificial-intelligence-startup-deepmind |dead-url=no }}</ref><ref>{{cite web|title=Report of Acquisition, TechCrunch|url=http://techcrunch.com/2014/01/26/google-deepmind/|publisher=TechCrunch|accessdate=27 January 2014|archive-date=2014-01-27|archive-url=https://web.archive.org/web/20140127114318/http://techcrunch.com/2014/01/26/google-deepmind/|dead-url=no}}</ref><ref>{{cite web |title=Reuters Report |last=Oreskovic |first=Alexei |url=http://www.reuters.com/article/2014/01/27/us-google-deepmind-idUSBREA0Q03220140127 |publisher=Reuters |accessdate=27 January 2014 |archive-date=2014-01-27 |archive-url=https://web.archive.org/web/20140127042513/http://www.reuters.com/article/2014/01/27/us-google-deepmind-idUSBREA0Q03220140127 |dead-url=no }}</ref><ref>{{cite news|title=Google Acquires Artificial Intelligence Start-Up DeepMind|url=http://www.theverge.com/2014/1/26/5348640/google-deepmind-acquisition-robotics-ai|work=The Verge|accessdate=27 January 2014|archive-date=2017-01-04|archive-url=https://web.archive.org/web/20170104105853/http://www.theverge.com/2014/1/26/5348640/google-deepmind-acquisition-robotics-ai|dead-url=no}}</ref><ref>{{cite web|title=Google acquires AI pioneer DeepMind Technologies|url=http://arstechnica.com/business/2014/01/google-acquires-ai-pioneer-deepmind-technologies/|work=Ars Technica|accessdate=27 January 2014|archive-date=2014-01-30|archive-url=https://web.archive.org/web/20140130040807/http://arstechnica.com/business/2014/01/google-acquires-ai-pioneer-deepmind-technologies/|dead-url=no}}</ref>

其中DeepMind要求谷歌的一个条件是他们将成立一个[[人工智能伦理]]理事会。<ref>{{cite magazine|title=Inside Google's Mysterious Ethics Board|url=http://www.forbes.com/sites/privacynotice/2014/02/03/inside-googles-mysterious-ethics-board/|magazine=Forbes|accessdate=12 October 2014|date=3 February 2014|archive-date=2016-03-04|archive-url=https://web.archive.org/web/20160304075801/http://www.forbes.com/sites/privacynotice/2014/02/03/inside-googles-mysterious-ethics-board/|dead-url=no}}</ref>

2023年4月21日，Alphabet表示将合并DeepMind和谷歌大脑。<ref name="新浪财经_手机新浪网 2023">{{cite web | title=应对微软和OpenAI威胁：谷歌将合并DeepMind和谷歌大脑两大AI研究部门 | website=新浪财经_手机新浪网 | date=2023-04-21 | url=https://finance.sina.cn/usstock/mggd/2023-04-21/detail-imyrawer9250878.d.html | language=zh | access-date=2023-04-21 | archive-date=2023-04-25 | archive-url=https://web.archive.org/web/20230425170501/https://finance.sina.cn/usstock/mggd/2023-04-21/detail-imyrawer9250878.d.html | dead-url=no }}</ref>

== AlphaGo ==
{{main|AlphaGo|Master (围棋软件)|AlphaGo Zero|AlphaZero|AlphaStar}}
[[File:Google-Deep Mind headquarters in London, 6 Pancras Square.jpg|thumb|位于英国伦敦泛克拉斯广场6号的建筑入口，谷歌及其子公司DeepMind的所在地。]]
DeepMind於2014年开始开发[[人工智能]][[圍棋軟體]][[AlphaGo]]。

2015年10月，分布式版AlphaGo[[分先]]以5:0击败了欧洲[[围棋]]冠军華裔法籍职业棋士[[樊麾]][[围棋段位制|二段]]<ref name="bbcgo">{{cite news |url=http://www.bbc.com/news/technology-35420579 |title=Google achieves AI 'breakthrough' by beating Go champion |date=27 January 2016 |work=[[BBC News]] |accessdate=2019-01-26 |archive-date=2016-01-30 |archive-url=https://web.archive.org/web/20160130035623/http://www.bbc.com/news/technology-35420579 |dead-url=no }}</ref><ref name="MetzWired2016">{{Cite web|title = In Major AI Breakthrough, Google System Secretly Beats Top Player at the Ancient Game of Go|url = http://www.wired.com/2016/01/in-a-huge-breakthrough-googles-ai-beats-a-top-player-at-the-game-of-go/|website = WIRED|access-date = 2016-02-01|language = en-US|date = 2016-01-27|last = Metz|first = Cade|archive-date = 2016-02-01|archive-url = https://web.archive.org/web/20160201121004/http://www.wired.com/2016/01/in-a-huge-breakthrough-googles-ai-beats-a-top-player-at-the-game-of-go/|dead-url = no}}</ref><ref>{{Cite web|url = http://www.britgo.org/files/2016/deepmind/BGJ174-AlphaGo.pdf|title = Sepcial Computer Go insert covering the AlphaGo v Fan Hui match|dte = |access-date = 2016-02-01|website = |publisher = British Go Journal|last = |first = |month = |orig-year = 2017|archive-date = 2016-02-02|archive-url = https://web.archive.org/web/20160202065347/http://www.britgo.org/files/2016/deepmind/BGJ174-AlphaGo.pdf|dead-url = no}}</ref> 。这是电脑围棋程序第一次在十九路棋盘且分先的情況下击败职业围棋棋手<ref name="lemondego">{{cite news |url=http://www.lemonde.fr/pixels/article/2016/01/27/premiere-defaite-d-un-professionnel-du-go-contre-une-intelligence-artificielle_4854886_4408996.html |title=Première défaite d'un professionnel du go contre une intelligence artificielle |date=27 January 2016 |work=[[Le Monde]] |language=fr |accessdate=2019-01-26 |archive-date=2016-01-29 |archive-url=https://web.archive.org/web/20160129103942/http://www.lemonde.fr/pixels/article/2016/01/27/premiere-defaite-d-un-professionnel-du-go-contre-une-intelligence-artificielle_4854886_4408996.html |dead-url=no }}</ref>。

2016年3月，AlphaGo挑战世界冠军[[韩国]][[職業棋士|职业棋士]][[李世乭]]（{{lang|ko|이세돌}}）九段，對弈結果为AlphaGo 4:1战胜了李世乭。

2016年12月29日，一个注册名为Magister（大陆客户端显示为Magist）的账号开始在弈城围棋网与职业棋手对弈，次日用户名改为Master。12月31日取得30连胜后，2017年1月1日Master转往[[腾讯]]野狐围棋网对弈，最终Master在野狐也取得30连胜，总成绩为60场不败<ref name="WSJ Mourn">{{cite news|title=Humans Mourn Loss After Google Is Unmasked as China’s Go Master|url=https://www.wsj.com/articles/ai-program-vanquishes-human-players-of-go-in-china-1483601561|accessdate=6 January 2017|work=Wall Street Journal|date=5 January 2017|archive-date=2019-05-26|archive-url=https://web.archive.org/web/20190526022838/https://www.wsj.com/articles/ai-program-vanquishes-human-players-of-go-in-china-1483601561|dead-url=no}}</ref>，在60局结束后，DeepMind CEO[[杰米斯·哈萨比斯]]声明确认Master就是升级版AlphaGo<ref name="Master">{{cite web|author=[[杰米斯·哈萨比斯]]|url=https://twitter.com/demishassabis/status/816660463282954240|title=Demis Hassabis on Twitter: "Excited to share an update on #AlphaGo!"|publisher=Demis Hassabis's [[Twitter]] account|date=4 January 2017|accessdate=4 January 2017|archive-date=2019-05-04|archive-url=https://web.archive.org/web/20190504005638/https://twitter.com/demishassabis/status/816660463282954240|dead-url=no}}</ref><ref name="Nature-Master">{{cite web|author=Elizabeth Gibney|url=http://www.nature.com/news/google-reveals-secret-test-of-ai-bot-to-beat-top-go-players-1.21253|title=Google reveals secret test of AI bot to beat top Go players|publisher=[[Nature]]|date=4 January 2017|accessdate=4 January 2017|archive-date=2017-01-05|archive-url=https://web.archive.org/web/20170105052336/http://www.nature.com/news/google-reveals-secret-test-of-ai-bot-to-beat-top-go-players-1.21253|dead-url=no}}</ref>。

2017年10月19日，AlphaGo的团队在《[[自然 (期刊)|自然]]》杂志上发表了一篇文章，介绍了AlphaGo Zero，这是一个没有用到人类数据的版本，比以前任何击败人类的版本都要强大。<ref name="Nature2017">{{cite web|url=http://www.nature.com/nature/journal/v550/n7676/full/nature24270.html|title=Mastering the game of Go without human knowledge|publisher=[[自然 (期刊)|Nature]]|date=19 October 2017|accessdate=19 October 2017|archive-date=2017-10-19|archive-url=https://web.archive.org/web/20171019114953/http://www.nature.com/nature/journal/v550/n7676/full/nature24270.html|dead-url=no}}</ref>通过跟自己对战，AlphaGo Zero经过3天的学习，以100:0的成绩超越了[[AlphaGo李世乭五番棋|AlphaGo Lee]]的实力，21天后达到了[[Master (围棋软件)|AlphaGo Master]]的水平，并在40天内超过了所有之前的版本。<ref name="Deepmind20171018">{{cite web|url=https://deepmind.com/blog/alphago-zero-learning-scratch/|title=AlphaGo Zero: Learning from scratch|publisher=DeepMind official website|date=18 October 2017|accessdate=19 October 2017|archive-date=2017-10-19|archive-url=https://web.archive.org/web/20171019220157/https://deepmind.com/blog/alphago-zero-learning-scratch/|dead-url=no}}</ref>

2018年12月7日，DeepMind在《[[科学_(期刊)|科学]]》期刊上发表关于[[AlphaZero]]的论文，AlphaZero使用與[[AlphaGo Zero]]類似但更一般性的演算法，在不做太多改變的前提下，並將演算法從[[圍棋]]延伸到[[将棋 (日本)|将棋]]與[[國際象棋]]上。

2019年1月25日，DeepMind人工智慧AlphaStar在《[[星际争霸II]]》以10：1戰勝人類職業玩家。<ref>{{Cite web |url=https://www.engadget.com/2019/01/24/deepmind-ai-starcraft-ii-demonstration-tlo-mana/ |title=DeepMind AI AlphaStar goes 10-1 against top 'StarCraft II' pros |accessdate=2019-01-26 |archive-date=2019-01-26 |archive-url=https://web.archive.org/web/20190126123727/https://www.engadget.com/2019/01/24/deepmind-ai-starcraft-ii-demonstration-tlo-mana/ |dead-url=no }}</ref>

2020年12月23日，DeepMind公佈其AI 算法MuZero<ref>{{cite news |title=Nature收录DeepMind“四代AlphaGo”论文，有望成为广泛使用的强化学习算法｜解读 |url=https://www.163.com/dy/article/FUMC4DRF05119734.html?clickfrom=w_tech |accessdate=2020-12-26 |archive-date=2021-08-15 |archive-url=https://web.archive.org/web/20210815063629/https://www.163.com/dy/article/FUMC4DRF05119734.html?clickfrom=w_tech |dead-url=no }}</ref>。

==AlphaFold==
{{main|AlphaFold}}
2016年，DeepMind開始利用人工智能研究[[蛋白质折叠]]，蛋白質結構是[[生物科学]]重要且棘手的課題。

2018年12月，DeepMind研發的AlphaFold通过成功预测43种蛋白质中的25种最准确的结构，赢得了第 3 蛋白质结构预测技术关键评估（CASP）。“这是一个灯塔项目，我们第一次在人力和资源方面对一个基本的、非常重要的、现实世界的科学问题进行重大投资，”哈萨比斯对卫报说。2020 年，在第 14 届 CASP 中，AlphaFold 的预测达到了与实验室技术相当的准确度分数。科学评审小组之一安德烈·克里什塔福维奇博士将这一成就描述为“真正了不起的”，并表示预测蛋白质如何折叠的问题已“基本解决”。<ref name="mg21">{{Cite web |url=https://www.mg21.com/deepmind.html |title=谷歌旗下人工智能全球领导者：DeepMind Technologies |publisher=美股之家 |date=2021-07-11 |access-date=2021-07-11 |archive-date=2021-07-15 |archive-url=https://web.archive.org/web/20210715105510/https://www.mg21.com/deepmind.html |dead-url=no }}</ref>

==DeepMind AlphaStar==

2019年1月，DeepMind 推出了 AlphaStar，这是一个玩即时战略游戏星际争霸 II 的程序。 AlphaStar 使用基于人类玩家回放的强化学习，然后与自己对战以增强其技能。在演示时，AlphaStar 拥有相当于 200 年游戏时间的知识。它与两名职业选手连续 10 场比赛获胜，尽管它具有能够看到整个场地的不公平优势，不像人类选手必须手动移动镜头。<ref name="mg21"/>

2019年7月，AlphaStar 开始在公共 1v1 欧洲多人天梯上与随机人类对战。与 AlphaStar 的第一次迭代只玩了 Protoss v. Protoss 不同，这个版本的 AlphaStar 扮演了游戏中的所有种族，并且更早地修复了不公平的优势。到2019年10月，AlphaStar 在星际争霸 II 的所有三场比赛中都达到了大师级别，成为第一个在没有任何游戏限制的情况下进入广受欢迎的电子竞技顶级联赛的人工智能。<ref name="mg21"/>

== 環境生成 ==
2024年3月，DeepMind推出了「Genie」（Generative Interactive Environments，生成式互動環境），這是一個可以根據文本描述、圖像或草圖生成類遊戲、動作可控虛擬世界的AI模型。Genie作為一個自迴歸潛在擴散模型（latent diffusion model）構建，無需標記的動作數據進行訓練，即可實現逐影格（frame-by-frame）的互動性。其後繼者Genie2於2024年12月發布，擴展了這些功能，以生成多樣化且具互動性的3D環境。<ref>{{Cite web |last=Orland |first=Kyle |date=2024-12-06 |title=Google's Genie 2 "world model" reveal leaves more questions than answers |url=https://arstechnica.com/ai/2024/12/googles-genie-2-world-model-reveal-leaves-more-questions-than-answers/ |access-date=2024-12-21 |website=Ars Technica |language=en-US |archive-date=13 December 2024 |archive-url=https://web.archive.org/web/20241213143646/https://arstechnica.com/ai/2024/12/googles-genie-2-world-model-reveal-leaves-more-questions-than-answers/ |url-status=live }}</ref>Genie3於2025年8月發布，具備更高解析度的世界生成能力，並能維持長達數分鐘的視覺一致性。<ref>{{Cite web |last=Whitwam |first=Ryan |date=2025-08-05 |title=DeepMind reveals Genie 3 "world model" that creates real-time interactive simulations |url=https://arstechnica.com/ai/2025/08/deepmind-reveals-genie-3-world-model-that-creates-real-time-interactive-simulations/ |access-date=2025-11-26 |website=Ars Technica |language=en}}</ref> 2026年1月29日，DeepMind向AI Ultra訂閱者正式發布了Project Genie。<ref>{{Cite web|title=Google Project Genie lets you create interactive worlds from a photo or prompt|url=https://arstechnica.com/google/2026/01/google-project-genie-lets-you-create-interactive-worlds-from-a-photo-or-prompt/|website=Ars Technica|date=2026-01-29|access-date=2026-01-31|language=en|first=Ryan|last=Whitwam}}</ref>

== 研究 ==

=== 深度强化学习 ===
繼[[AlphaGo]]之後，Google DeepMind執行長Demis Hassabis曾表示將研究用人工智-{zh-cn:能;zh-tw:慧;}-與人類玩其他遊戲，例如[[即時戰略遊戲]]《[[星海爭霸II|-{zh-cn:星际争霸II;zh-hk:星際爭霸II;zh-tw:星海爭霸II;}-]]》（{{lang|en|''StarCraft II''}}）<ref>{{Cite web |url=http://news.ltn.com.tw/news/life/breakingnews/1627739 |title=AlphaGo挑戰人類圍棋後傳可能再戰《星海爭霸》 |accessdate=2016-03-10 |archive-date=2016-03-10 |archive-url=https://web.archive.org/web/20160310202451/http://news.ltn.com.tw/news/life/breakingnews/1627739 |dead-url=no }}</ref><ref>{{Citation|last=DeepMind|title=StarCraft II DeepMind feature layer API|date=2016-11-04|url=https://www.youtube.com/watch?v=5iZlrBqDYPM|accessdate=2016-11-23|archive-date=2016-11-22|archive-url=https://web.archive.org/web/20161122210606/https://www.youtube.com/watch?v=5iZlrBqDYPM|dead-url=no}}</ref>。深度AI如果能直接使用在其他各種不同領域，除了未來能玩不同的遊戲外，例如[[自動駕駛]]、{{le|财务顾问|Financial adviser|投資顧問}}、音樂評論、甚至司法判決等等目前需要人腦才能處理的工作，基本上也可以直接使用相同的神經網路去學而習得與人類相同的思考力。

== 参考文献 ==
{{reflist|colwidth=32em|refs=
<ref name="econ2015">{{cite web| url= http://www.economist.com/news/science-and-technology/21645108-you-can-teach-computer-play-games-better-it-teach-itself-computers| title= Computers, gaming| date= 28 February 2015| publisher= [[The Economist]]| accessdate= 2016-01-28| archive-date= 2018-01-14| archive-url= https://web.archive.org/web/20180114183719/https://www.economist.com/news/science-and-technology/21645108-you-can-teach-computer-play-games-better-it-teach-itself-computers| dead-url= no}}</ref>
<ref name="nature2015">{{cite journal|last= Mnih1|first= Volodymyr|last2= Kavukcuoglu1|first2= Koray|last3= Silver|first3= David|date= 26 February 2015|title= Human-level control through deep reinforcement learning|url= http://www.nature.com/nature/journal/v518/n7540/full/nature14236.html|journal= Nature|publisher= |volume= 518|issue= |pages= 529–33|doi= 10.1038/nature14236|access-date= 25 February 2015|pmid= 25719670|archive-date= 2017-09-25|archive-url= https://web.archive.org/web/20170925141022/http://www.nature.com/nature/journal/v518/n7540/full/nature14236.html|dead-url= no}}</ref>
}}

== 外部链接 ==
* {{official website|https://deepmind.google/}}
* [https://deepmind.google/research/publications/ Google DeepMind - publications]
* [https://www.youtube.com/channel/UCP7jMXSY2xbc3KCAE0MHQ-A DeepMind在YouTube上的官方频道]{{Wayback|url=https://www.youtube.com/channel/UCP7jMXSY2xbc3KCAE0MHQ-A |date=20161115111006 }}

{{Alphabet Inc.}}
{{Google公司}}
{{Google AI}}
{{全球主要資訊科技公司}}
{{Differentiable computing}}
{{生成式人工智能}}
{{通用人工智能的存在風險}}

[[Category:Google收购]]
[[Category:英国软件公司]]
[[Category:2010年成立的公司]]
[[Category:机器学习]]
[[Category:人工智能]]
[[Category:Alphabet]]
[[Category:人工智能实验室]]
[[Category:Google DeepMind]]

Google DeepMind - 版本历史

imported>Cewbot：​清理跨語言連結人工智能伦理成為內部連結:編輯摘要的紅色連結經繁簡轉換後存在，非bot錯誤編輯 (本次機械人作業已完成18.3%)

imported>Cewbot：清理跨語言連結人工智能伦理成為內部連結:編輯摘要的紅色連結經繁簡轉換後存在，非bot錯誤編輯 (本次機械人作業已完成18.3%)