前言
(p.Ⅱ)小冰的贴心和温暖：James Risley, “Reporter John Markoff and Microsoft Research head Peter Lee talk self-drivingcars, robots taking the SAT and the AI of Hollywood,” GeekWire,October 1, 2015, www.geekwire.com/2015/reporter-john-markoff-and-microsoft-research-head-peter-lee-talkself-driving-cars-robots-taking-the-sat-and-the-ai-of-hollywood.
(p.Ⅴ)2016年，非营利杂志：Jeff Larson et al., “How We Analyzed the COMPAS Recidivism Algorithm,” ProPublica, May23, 2016, www.propublica.org/article/how-we-analyzed-the-compas-recidivism-algorithm.
(p.Ⅵ)推送虚假新闻的算法：Abby Ohlheiser, “Threedays after removing human editors, Facebook is already trendingfake news,” Washington Post, August 29, 2016, www.washingtonpost.com/news/theintersect/wp/2016/08/29/a-fake-headline-about-megyn-kelly-was-trending-onfacebook/?noredirect=on&utm_term=.9185140ef0f1.
(p.Ⅵ)招聘广告中的性别偏见：Tom Simonite, “Probing the Dark Side of Google's Ad-Targeting System,” MIT Technology Review, July 6, 2015, www.technologyreview.com/s/539021/probing-the-dark-side-of-googles-adtargeting-system.
(p.Ⅵ)自动补全算法中的反犹太主义：Carole Cadwalladr, “Google, democracy and the truth about internet search,” The Guardian, December 4, 2016, www.theguardian.com/technology/2016/dec/04/google-democracytruth-internet-search-facebook.
(p.Ⅵ)她称之为“数学型杀伤性武器”：Cathy O'Neil, Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy (New York: Crown, 2016).
(p.Ⅷ)根据美国监管机构的一份报告：Eric M. Aldrich, Joseph A. Grundfest, and Gregory Laughlin, “The Flash Crash: A New Deconstruction,” Social Science Research Network, January 26, 2016, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2721922.
(p.Ⅷ)根据一些估算，几乎1万亿：Ben Rooney, “Trading program sparked May ‘flash crash,’ ” CNN Money, October 1, 2010, http://money.cnn.com/2010/10/01/markets/SEC_CFTC_flash_crash/index.htm.
01 算法世界中的自由意志
(p.007)然而，请掂量掂量这些事实：Carlos A. Gomez-Uribe and Neil Hunt, “The Netflix Recommender System: Algorithms, Business Value, and Innovation,” ACM Transactions on Management Information Systems 5, no. 4 (January 2016): 13, https://dl.acm.org/citation.cfm?id=2843948.
(p.007)亚马逊上几乎35%的销售：Paul Lamere and Stephen Green,“Project Aura: Recommendation for the Rest of Us,” JavaOne Conference presentation, 2008, www.oracle.com/technetwork/systems/ts-5841-159144.pdf.
(p.008)这种行为非常普遍：Tristan Harris, “How better tech could protect us from distraction,” TEDxBrussels, December 2014, www.ted.com/talks/tristan_harris_how_better_tech_could_protect_us_from_distraction.
(p.009)每次用户打开Facebook应用：Lars Backstrom, “News Feed FYI: A Window Into News Feed,” Facebook Business, August 6, 2013,
www.facebook.com/business/news/News-Feed-FYI-A-Window-Into-NewsFeed.
(p.012)盖利斯报道说超过一半：David Gelles, “Inside Match.com,” Slate, July 30, 2011, www.slate.com/articles/life/ft/2011/07/inside_matchcom.html.
(p.012)2012年，Facebook开展了一次研究：Micah L. Sifry, “Facebook Wants You to Vote on Tuesday. Here's How it Messed With Your Feed in 2012,” Mother Jones, October 31, 2014, www.motherjones.com/politics/2014/10/can-voting-facebook-button-improve-voter-turnout.
02 意外后果之定律
(p.017)我很确定，就算我没有做这个功能，也会有人做出来：Liz Gannes, “Nearly a Decade Later, the Autocomplete Origin Story: Kevin Gibbs and Google Suggest,” All Things D, August 23, 2013, http://allthingsd.com/20130823/nearly-a-decade-later-the-autocomplete-origin-story-kevingibbs-and-google-suggest.
(p.018)每个猜测都必然会引发：Carole Cadwalladr, “Google,democracy and the truth about internet search,” The Guardian, December 4, 2016, www.theguardian.com/technology/2016/dec/04/google-democracytruth-internet-search-facebook.
(p.025)一位前员工揭秘：Michael Nunez, “Former Facebook Workers: We Routinely Suppressed Conservative News,”Gizmodo, May 9, 2016, https://gizmodo.com/former-facebook-workers-we-routinely-suppressedconser-1775461006.
(p.025)根据估算，前20位：Timothy B. Lee, “The top 20 fake news stories outperformed real news at the end of the 2016 campaign,” Vox, November 16, 2016, www.vox.com/new-money/2016/11/16/13659840/facebook-fake-news-chart.
03 计算机的煎蛋食谱：算法是如何编写的
(p.041)特拉维夫大学的研究团队：Eyal Carmi et al., “Is Oprah Contagious? The Depth of Diffusion of Demand Shocks in a Product Network,” MIS Quarterly 41, no. 1 (2017): 207っ21, https://misq.org/catalog/product/view/id/1795.
(p.043)我和弗莱德做的第一个研究：Daniel M. Fleder and Kartik Hosanagar, “Blockbuster Culture's Next Rise or Fall: The Impact of Recommender Systems on Sales Diversity,” Management Science 55, no. 5 (May 2009): 697っ712, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=955984.
(p.043)为了进一步验证这个理论：Dokyun Lee and Kartik Hosanagar,“How Do Recommender Systems Affect Sales Diversity?A Cross-Category Investigation via Randomized Field Experiment,” Information Systems Research, October 7, 2016, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2603361.
(p.046)我的团队的后续研究：Kartik Hosanagar et al., “Will the Global Village Fracture into Tribes: Recommender Systems and Their Effects on Consumers,” Management Science 60, no. 4 (April 2014): 805っ23, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1321962.
(p.046)以及麻省理工学院的一个团队：Erik Brynjolfsson et al., “Goodbye Pareto Principle, Hello Long Tail: The Effect of Search Costs on Concentration of Product Sales,” Management Science, January 2011, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=953587.
(p.046)意识到了协同过滤算法的偏见性：Sander Dieleman, “Recommending music on Spotify with deep learning,” Sander Dieleman [blog], August 5, 2014, http://benanne.github.io/2014/08/05/spotify-cnns.html.
(p.047)Spotify另外使用机器学习算法：Sophia Ciocca,“How Does Spotify Know You So Well?” Medium, October 10, 2017, https://medium.com/s/story/spotifys-discover-weekly-how-machine-learning-finds-yournew-music-19a41ab76efe.
04 算法的智能化：人工智能简史
(p.049)1783年5月28日，匈牙利发明家：Wolfgang Kempelen to Benjamin Franklin, May 28, 1783, Founders Online, National Archives, last modified June 13, 2018, https://founders.archives.gov/documents/Franklin/01-40-02-0041. Original source: The Papers of Benjamin Franklin,vol. 40, May 16 through September 15, 1783, ed. Ellen R. Cohn (New Haven and London: Yale University Press, 2011), 80っ81.
(p.050)那个时候，肯佩伦的机器：Nathan Ensmenger, “Is chessthe drosophila of artificial intelligence? A social history of analgorithm,” Social Studies of Science 42, no. 1 (February 2012): 5っ30,https://pdfs.semanticscholar.org/c9e7/3fc7ec81458057e6f96de1cba095e84a05c4.pdf.
(p.050)1950年，真正的科学基础：A. M. Turing, “Computing Machinery and Intelligence,” Mind, n.s., 59, no. 236 (October 1950): 433っ60,www.csee.umbc.edu/courses/471/papers/turing.pdf.
(p.050)话虽如此，基金会还是勉强赞助了：Ronald R. Kline, “Cybernetics, Automata Studies and the Dartmouth Conferenceon Artificial Intelligence,” IEEE Annals of the History of Computing 33, no. 4 (2011): 5っ16, www.semanticscholar.org/paper/Cybernetics%2C-AutomataStudies%2C-and-the-Dartmouth-on-Kline/0dad2e0f8520d81ad0080a7aff45c96e4866c541.
(p.052)在参加达特茅斯会议的前几个月：Hunter Heyck, “Defining the Computer: Herbert Simon and the Bureaucratic Mind—Part 2,”IEEE Annals of the History of Computing 30, no. 2 (AprilっJune2008): 52っ63, www.cbi.umn.edu/about/nsl/v24n1text.pdf.
(p.052)软件证明了定理：同上
(p.052)对这种工程学上的壮举：同上
(p.054)计算机史学家内森·恩斯门格：Ensmenger，“Is chessthe drosophila...?”
(p.059)虽然数据收集的手段比较原始：Ford Burkhart, “Dr.
JackMyers, 84, a Pioneer In Computer-Aided Diagnoses,” The New York Times, February 22, 1998, www.nytimes.com/1998/02/22/us/dr-jack-myers-84a-pioneer-in-computer-aided-diagnoses.html.
(p.060)使用自己的五感：Luke Oakden-Rayner, “The End of Human Doctors—Understanding Medicine,” Luke Oakden-Rayner, PhD Candidate/Radiologist [blog], April 24, 2017,https://lukeoakdenrayner.wordpress.com/2017/04/24/the-end-of-human-doctors-understanding-medicine.
05 机器学习和预测性-适应性悖论
(p.063)就算深蓝的国际象棋知识：Murray Campbell, “20 Years After Deep Blue, a New Era in Human-Machine Collaboration,” IBMTHINK Blog, May 11, 2017, www.ibm.com/blogs/think/2017/05/deep-blue.
(p.064)基于超过3000万手专业围棋棋手的走法：Cade Metz, “In a Huge Breakthrough,Google's AI Beats a Top Player at the Game of Go,” Wired, January 27, 2016, www.wired.com/2016/01/in-a-huge-breakthroughgoogles-ai-beats-a-top-player-at-the-game-of-go.
(p.064)执黑的AlphaGo落下了不同寻常的一子：“Move 37!! Lee Sedol vs AlphaGo Match 2,” YouTube, March 12, 2016, www.youtube.com/watch?v=JNrXgpSEEIE.
(p.068)话要说清楚，机器学习系统也能被操纵：Kevin Eykholt et al., “Robust Physical-World Attacks on Deep Learning Models,” arXiv.org,Cornell University Library, July 27, 2017, https://arxiv.org/abs/1707.08945.
(p.072)法律和制度必须与人类的思想齐头并进：Thomas Jefferson to H.Tompkinson (aka Samuel Kercheval), July 12, 1816, www.loc.gov/resource/mtj 1.049_0255_0262.
(p.075)2001年，微软的研究者们：Michelle Banko and Eric Brill, “Scaling to Very Very Large Corpora for Natural Language Disambiguation,” Microsoft Research, January 1, 2001, www.microsoft.com/en-us/research/publication/scaling-to-very-very-large-corpora-fornatural-language-disambiguation.
(p.076)几个星期里，仅仅凭借自学：David Silver et al., “Mastering the game of Go without Human Knowledge,” Nature 550 (October2017): 354-59, https://www.nature.com/articles/nature24270.epdf?author_access_token=VJXbVjaSHxFoctQQ4p2k4tRgN0jAjWel9jnR3ZoTv0PVW4gB86EEpGqTRDtpIz-2rmo8-KG06gqVobU5NSCFeHILHcVFUeMsbvwSlxjqQGg98faovwjxeTUgZAUMnRQ.
(p.077)所以毫不意外：Dawn Chan, “The AI That Has Nothing to Learn From Humans,” The Atlantic, October 20, 2017,www.theatlantic.com/technology/archive/2017/10/alphago-zero-the-ai-that-taught-itself-go/543450.
06 算法心理学
(p.085)在2010年：Daniel Fleder et al., “Recommender Systems and Their Effects on Consumers: The Fragmentation Debate,” EC'10: Proceedings of the 11th ACM Conference on Electronic Commerce, Cambridge, MA, June 7-11, 2010 (New York: ACM Digital Library, 2010), 229-30, https://dl.acm.org/citation.cfm?id=1807378.
(p.085)和2014年的研究中：Kartik Hosanagar et al., “Will the Global Village Fracture into Tribes: Recommender Systems and Their Effects on Consumers,” Management Science 60, no. 4 (April 2014): 805-23, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1321962.
(p.085)2015年，Facebook的研究团队：Eytan Bakshy et al., “Exposure to ideologically diverse news and opinion on Facebook,” Science 348, no. 6239 (June 2015): 1130-32, http://science.sciencemag.org/content/348/6239/1130.
(p.087)研究者发现绝大多数：Seth Flaxman et al., “Filter Bubbles, Echo Chambers, and Online News Consumption,” Public Opinion Quarterly 80, no. S1 (2016): 298-320, https://5harad.com/papers/bubbles.pdf.(p.089)研究者们发现仅仅：Kiran Garimella et al., “Reducing Controversy by Connecting Opposing Views,” WSDM'17: Proceedings of the Tenth ACM International Conference on Web Search and Data Mining, Cambridge, UK, February 6-10, 2017 (New York: ACM Digital Library, 2017), 81-90, https://dl.acm.org/authorize.cfm?key=N20012.
07 我们信仰算法
(p.094)根据某些估算：Adrienne Lafrance, “Self-Driving Cars Could Save 300 000 Lives Per Decade in America,” The Atlantic, September 29, 2015, www.theatlantic.com/technology/archive/2015/09/self-driving-carscould-save-300000-lives-per-decade-in-america/407956.
(p.094)然而2018年4月一次民意调查：Edward Graham, “Americans Less Trusting of Self-Driving Safety Following High-Profile Accidents,” Morning Consult, April 5, 2018, https://morningconsult.com/2018/04/05/americans-less-trusting-self-driving-safety-following-high-profile-accidents.
(p.095)2017年底时：Alex Eule, “As Robo-Advisors Cross $200 Billion in Assets, Schwab Leads in Performance,” Barron's, February 3, 2018, www.barrons.com/articles/as-robo-advisors-cross-200-billion-in-assets-schwableads-in-performance-1517509393.
(p.096)1996年，明尼苏达大学的两位心理学家：William M. Grove and Paul E. Meehl, “Comparative efficiency of informal (subjective,impressionistic) and formal (mechanical, algorithmic) prediction procedures: The Clinical-Statistical Controversy,” Psychology, Public Policy, and Law 2, no. 2 (January 1996): 293-323, https://experts.umn.edu/en/publications/comparative-efficiency-of-informal-subjective-impressionisticand.
(p.097)对约100万名学生进行了调查：Mark D. Alicke and Olesya Govorun, “The Better-Than-Average Effect,” in The Self in Social Judgment, ed. Mark D. Alicke, David A. Dunning, and Joachim Krueger, Studies in Self and Identity (New York: Psychology Press, 2005), 85-106, www.researchgate.net/publication/230726570_The_better-than-average_effect.
(p.097)一次对教师的调查发现：K. Patricia Cross, “Not Can, But Will College Teaching Be Improved?” New Directions for Higher Education 17 (Spring 1977): 1-15, https://onlinelibrary.wiley.com/doi/abs/10.1002/he.36919771703.
(p.097)和我们的驾驶主题相关：Ola Svenson, “Are we all lessrisky and more skillful than our fellow drivers?” Acta Psychologica 47, no. 2 (February 1981): 143-48, www.sciencedirect.com/science/article/pii/0001691881900056.
(p.098)研究者珍妮佛·洛格：Jennifer M. Logg, Julia A.Minson, and Don A. Moore, “Algorithm Appreciation: PeoplePrefer Algorithmic to Human Judgment,” Harvard Business School Working Paper 17-086, April20, 2018, www.hbs.edu/faculty/Publication%20Files/17-086_610956b6-7d91-4337-90cc-5bb5245316a8.pdf.
(p.099)另一个有趣的理论：Robyn M. Dawes, “The robust beauty of improper linear models in decision making,”American Psychologist 34, no. 7 (July 1979): 571-82, http://psycnet.apa.org/record/1979-30170-001.
(p.101)同样的调查问卷：Graham, “Americans LessTrusting.”
(p.102)小型浸润性乳腺癌的确诊数量：Siddhartha Mukherj ee, “A.I.Versus M.D.,” The New Yorker, April 3, 2017, www.newyorker.com/magazine/2017/04/03/ai-versus-md.
(p.104)这并不是说所有比较人工智能和医生的研究：Andre Esteva et al., “Dermatologistlevel classification of skin cancer with deep neural networks,”Nature 542 (February 2017): 115-18, www.nature.com/articles/nature21056.
(p.104)所以钱德拉塞卡尔和其他从业者：Varun Gulshan et al.,“Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs,”JAMA 316, no. 22 (December 2016): 2402-10, https://jamanetwork.com/journals/jama/article-abstract/2588763.
(p.105)你不可能从纽约吼到加利福尼亚：Mukherjee, “A.I. Versus M.D.”
08 哪个才是主宰——算法还是用户？
(p.110)研究者伯克利·迪特沃斯特、约瑟夫·西蒙斯和凯德·梅西：Berkeley Dietvorst, Joseph P. Simmons, and Cade Massey, “Overcoming Algorithm Aversion: People Will Use Imperfect Algorithms if They Can (Even Slightly) Modify Them,” Management Science 64, no. 3 (March 2018): 1156-70, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2616787.
(p.110)计算机科学家团队：F. Maxwell Harper et al., “Putting Users in Control of Their Recommendations,” RecSys'15: Proceedings of the 9th ACM Conference on Recommender Systems, Vienna, Austria, September 16-20, 2015 (New York: ACM Digital Library, 2015), 3-10, https://dl.acm.org/citation.cfm?id=2800179.
(p.111)“我们有些人”：Rion Thompson, “Chris Urmson Explain Google's Self-Driving Car Project,” SXSW, March 3, 2016, www.sxsw.com/interactive/2016/chris-urmson-explain-googles-self-driving-car-project.
(p.114)没有明确的答案，因为：William Langewiesche, “The Human Factor,” Vanity Fair, October 2014, www.vanityfair.com/news/business/2014/10/air-france-flight-447-crash.
(p.114)根据电梯历史学家李·格雷：Steve Henn, “The Big Red Button,” Planet Money, Episode 642, NPR, July 29, 2015, www.npr.org/templates/transcript/transcript.php?storyId=427467598.
(p.115)2013年巴西和法国的研究团队：Natalia Araujo Pacheco et al., “A Perceived-Control Based Model to Understanding the Effects of Co-production on Satisfaction,” BAR—Brazilian Administration Review 10, no. 2 (April?June 2013), www.scielo.br/scielo.php?pid=S1807-76922013000200007&script=sci_arttext.
09 黑盒之内
(p.123)研究者们已经观察到：Weiquan Wang and Izak Benbasat,“Recommendation Agents for Electronic Commerce:Effects of Explanation Facilities on Trusting Beliefs,” Journal of Management Information Systems 23, no. 4 (Spring 2007): 217-46,https://www.researchgate.net/profile/Weiquan_Wang/publication/220591412_Recommendation_Agents_for_Electronic_Commerce_Effects_of_Explanation_Facilities_on_Trusting_Beliefs/links/0f31753698f5cb6beb000000/Recommendation-Agents-for-ElectronicCommerce-Effects-of-Explanation-Facilities-on-Trusting-Beliefs.pdf.
(p.124)克孜尔切克着手开始创建一个在线版本：René F. Kizilcec, “How Much Information?: Effects of Transparency on Trust in an Algorithmic Interface,” in CHI'16: Proceedings of the 2016 CHI Conference on Human Factors in Computing Systems, San Jose, CA, May 7-12, 2016 (New York: ACM Digital Library, 2016), 2390-95, https://doi.org/10.1145/2858036.2858402.
(p.128)但根据研究：Wang and Benbasat, “Recommendation Agents for Electronic Commerce.”
(p.131)根据相关报道，最主要的障碍在于：Julia Powles, “New York City's Bold, Flawed Attempt to Make Algorithms Accountable,” The NewYorker, December 20, 2017, www.newyorker.com/tech/elements/newyork-citys-bold-flawed-attempt-to-make-algorithms-accountable.
(p.133)卡内基·梅隆大学的研究人员：Anupam Datta, Shayak Sen, and Yair Zick, “Algorithmic Transparency via Quantitative Input Influence: Theory and Experiments with Learning Systems,” in 2016 IEEE Symposium on Security and Privacy, San Jose, CA, May 22-26, 2016 (IEEE Explore Digital Library, 2016), 598-617, https://ieeexplore.ieee.org/document/7546525/#full-text-section.
(p.134)另一种方法由研究人员：Dave Gershgorn,“We Don't Understand How AI Make Most Decisions, So Now Algorithms are Explaining Themselves,” Quartz, December 20,2016, https://qz.com/865357/we-dont-understand-how-ai-make-most-decisions-so-now-algorithms-areexplaining-themselves.
(p.134)障碍之一是：Stuart J. Russell and Peter Norvig, Artificial Intelligence: A Modern Approach, 3rd ed. (Harlow, Essex, UK:Pearson Education, 2009).
(p.134)2016年，国防高级研究计划局：David Gunning,“Explainable Artificial Intelligence (XAI),” DARPA, www.darpa.mil/program/explainable-artificial-intelligence.
10 算法权利法案
(p.139)2016年10月，白宫科学技术政策办公室：Executive Office of the President, National Science and Technology Council, Committeeon Technology, “Preparing for the Future of Artificial Intelligence”(Washington, DC: Office of Science and Technology Policy, October 2016), https://obamawhitehouse.archives.gov/sites/default/files/whitehouse_files/microsites/ostp/NSTC/preparing_for_the_future_of_ai.pdf.
(p.140)由信息技术领域的教育家、研究人员和专业人士组成：ACM US Public Policy Council, “Statement on Algorithmic Transparency and Accountability,” January 12, 2017, www.acm.org/binaries/content/assets/public-policy/2017_usacm_statement_algorithms.pdf.
(p.141)纽约大学AI Now研究所：Alex Campolo et al.,“AI Now 2107Report,” AI Now Institute, 2017, https://ainowinstitute.org/AI_Now_2017_Report.pdf.
(p.141)其中一位支持者是本·施耐德曼：Thomas Macaulay, “Pioneering computer scientist calls for National Algorithm Safety Board,”TechWorld, May 31, 2017, www.techworld.com/data/pioneeringcomputer-scientist-calls-for-national-algorithms-safety-board-3659664.
(p.142)2017年12月，纽约市：Julia Powles, “New York City's Bold, Flawed Attempt to Make Algorithms Accountable,” The New Yorker, December 20, 2017, www.newyorker.com/tech/elements/new-york-citys-boldflawed-attempt-to-make-algorithms-accountable.
(p.146)为了更好地理解这两个支柱：Nicholas Diakopoulos and Michael Koliska, “Algorithmic Transparency in the NewsMedia,” Digital Journalism 5, no. 7 (July 2016): 809-28, www.nickdiakopoulos.com/wpcontent/uploads/2016/07/Algorithmic-Transparency-in-the-News-MediaFinal.pdf.
(p.149)那么摆在我们面前的：“The Russell-Einstein Manifesto,” Pugwash Conferences on Science and World Affairs, July 9, 1955, https://pugwash.org/1955/07/09/statement-manifesto.
结论：算法的游戏
(p.151)恩斯门格几年前曾指出：Nathan Ensmenger, “Is chess the drosophila of artificial intelligence? A social history of analgorithm,” Social Studies of Science 42, no. 1 (February 2012): 5-30, https://pdfs.semanticscholar.org/c9e7/3fc7ec81458057e6f96de1cba095e84a05c4.pdf.
(p.152)游戏结束时，电脑：Raman Chandrasekar,“Elementary? Question answering, IBM's Watson, and the Jeopardy! challenge,” Resonance 19, no. 3 (March 2014): 222-41, https://link.springer.com/article/10.1007/s12045-014-0029-7.
(p.153)尼尔斯·尼尔森宣称：Nils J. Nilsson, “Human-Level Artificial Intelligence? Be Serious!” AI Magazine 26, no. 4 (Winter 2005): 68-75, www.aaai.org/ojs/index.php/aimagazine/article/view/1850.
(p.154)分析表明，有将近一半：James Manyika et al., “Jobslost, jobs gained: What the future of work will mean for jobs, skills, and wages,” McKinsey Global Institute, November 2017, www.mckinsey.com/featuredinsights/future-of-organizations-and-work/jobs-lost-jobs-gained-what-thefuture-of-work-will-mean-for-jobs-skills-and-wages.
(p.154)十分之一的参与者认为：Nick Bostrom, Super intelligence: Paths, Dangers, Strategies (Oxford, UK: Oxford University Press, 2014).
(p.155)这项研究的目的是探讨：Mike Lewis et al.,“Deal or No Deal? End-to-End Learning for Negotiation Dialogues,” arXiv.org eprint, arXiv: 1706.05125, June 16, 2017, https://arxiv.org/pdf/1706.05125.pdf.
(p.157)相反，爱丽丝与鲍勃之间的不寻常交流：Paul Lilly, “Facebook kills AI that invented its own language because English was slow,” PC Gamer, July 27, 2017, www.pcgamer.com/facebook-kills-ai-thatinvented-its-own-language-because-english-was-slow.