1. L. Sweeney, “Discrimination in Online Ad Delivery,” Communications of the ACM 56, no. 5 (2013): 44–54, https://dataprivacylab.org/projects/onlineads/.
2. 同上。
3. “Racism Is Poisoning Online Ad Delivery, Says Harvard Professor,” MIT Technology Review, February 4, 2013, https://www.technologyreview.com/s/510646/racism-is-poisoning-online-ad-delivery-says-harvard-professor/.
4. Anja Lambrecht and Catherine Tucker, “Algorithmic Bias· An Empirical Study into Apparent Gender-Based Discrimination in the Display of STEM Career Ads” (paper presented at the NBER Summer Institute, July 2017).
5. Diane Cardwell and Libby Nelson, “The Fire Dept. Tests That Were Found to Discriminate,” New York Times, July 23, 2009, https://cityroom.blogs.nytimes.com/2009/07/23/the-fire-dept- tests-that-were-found-to-discriminate/·mcubz=0&_r=0; US v. City of New York (FDNY), https://www.justice.gov/archives/crt-fdny/overview.
6. Paul Voosen, “How AI Detectives Are Cracking Open the Black Box of Deep Learning,” Science, July 6, 2017, http://www.sciencemag.org/news/2017/07/how-ai-detectives-are-cracking-open-black-box-deep-learning.
7. T. Blake, C. Nosko, and S. Tadelis, “Consumer Heterogeneity and Paid Search Effectiveness: A Large- Scale Field Experiment,” Econometrica 83 (2015): 155–174.
8. Hossein Hosseini, Baicen Xiao, and Radha Poovendran, “Deceiving Google’s Cloud Video Intelligence API Built for Summarizing Videos” (paper presented at CVPR Workshops, March 31, 2017), https://arxiv.org/pdf/1703.09793.pdf; see also “Artificial Intelligence Used by Google to Scan Videos Could Easily Be Tricked by a Picture of Noodles,” Quartz, April 4, 2017, https://qz.com/948870/the-ai-used-by-google-to-scan-videos-could-easily-be-tricked-by-a-picture-of-noodles/.
9. 可见C. S. Elton, The Ecology of Invasions by Animals and Plants (New York: John Wiley, 1958)引用的数千个案例。
10. 根据2016年11月20日滑铁卢大学（University of Waterloo）校长佩尔．苏利文（Pearl Sullivan）、教授亚历山大．王（Alexander Wong）及其他滑铁卢大学教授的讨论。
11. 在装置上预测还有第四个优点：有时候为了实际目的而有必要这样做。比方说，Google眼镜必须能够判断眼皮的动作到底是眨眼（非故意）还是挤眼睛（故意），因为后者是用来控制装置的方法。因为做出判断的速度需求，将数据送到云端再等答案并不实际。预测机器必须置于装置之中。
12. Ryan Singel, “Google Catches Bing Copying; Microsoft Says ‘So What·’” Wired, February 1, 2011, https://www.wired.com/2011/02/bing-copiesgoogle/.
13. 为什么不可以接受的讨论，请参考Shane Greenstein “Bing Imitates Google: Their Conduct Crosses a Line,” Virulent Word of Mouse (blog), February 2, 2011, https://virulentwordofmouse.wordpress.com/2011/02/02/bing-imitates-google-their-conduct-crosses-a-line/；而相反的论点可参考Ben Edelman “In Accusing Microsoft, Google Doth Protest Too Much,” hbr.org, February 3, 2011, https://hbr.org/2011/02/in-accusing-microsoft-google.html.
14. 有趣的是，Google企图操纵微软的机器学习，效果并不是很好。Google进行的100个实验中，只有7到9个出现在Bing的搜寻结果中。请参考Joshua Gans, “The Consequences of Hiybbprqag’ing,” Digitopoly, February 8, 2011; https://digitopoly.org/2011/02/08/the-consequences-of-hiybbprqaging/.
15. Florian Tramer, Fan Zhang, Ari Juels, Michael K. Reiter, and Thomas Ristenpart, “Stealing Machine Learning Models via Prediction APIs” (paper presented at the Proceedings of the 25th USENIX Security Symposium, Austin, TX, August 10–12, 2016), https://regmedia.co.uk/2016/09/30/sec16_paper_tramer.pdf.
16. James Vincent, “Twitter Taught Microsoft’s AI Chatbot to Be a Racist Asshole in Less Than a Day,” The Verge, March 24, 2016, https://www.theverge.com/2016/3/24/11297050/tay-microsoft-chatbot-racist.
17. Rob Price, “Microsoft Is Deleting Its Chatbot’s Incredibly Racist Tweets,” Business Insider, March 24, 2016, http://www.businessinsider.com/microsoft-deletes-racist-genocidal-tweets-from-ai-chatbot-tay-2016-3·r=UK&IR=T.
