时事分析 | 创新及科技发展 | 2017-10-09 | 《星岛日报》

你玩A.I. A.I.玩你?



智能科技日新月异,电脑程式击败围棋大师[1]、胜过电子游戏顶尖高手[2],已不是天方夜谭。但人类总是不愿轻易认输,会尝试以各种方法骗过程式,赚取甜头,甚至重夺主导权。人机斗法,荣辱未必是重点,胜负对社会产生的影响,才是惹人关注的话题。

钻程式空子 某个人利益

去年微软在Twitter推出一个能够透过与人类对话学习的人工智能程式[3],结果却被一撮Twitter用户「教坏」,贴出带有性别及种族歧视色彩、否认纳粹大屠杀,以及认同白人至上的「推文」,结果程式仅面世16小时,就被微软关掉。[4]这个例子正好说明,在互联网世界「恶搞」程式,受影响的不会只是恶搞者和相关企业,因为「变坏」的程式可能会伤害到不同群体的感情,以至为一些不当行为摇旗呐喊。

为了某些目的钻程式空子,其实相当普遍。现时许多网页作者、程式开发人员,也会因应网络搜寻器的排列准则,扭尽六壬,务求令自己的产品占据搜寻结果前列。[5]部分政治群体也会借此技俩宣扬政治主张,例如指称气候变化是骗局、同性恋有罪,以及散播反犹太人思想。[6]

如今共享经济成为潮流,针对程式设定的行为也出现于相关平台。英国华威大学(University of Warwick)和美国纽约大学(New York University)早前联合发表的研究,便指有Uber司机在网上讨论区探讨如何针对程式的设定「搵着数」,例如借鼓动大量同行登出Uber程式,减少司机供应量,以提升车资。另外,针对不受司机欢迎的Uber「拼车」服务,有司机在讨论区宣称,只要在接载首程乘客后登出程式,就无需再接第二程乘客,占尽Uber鼓励司机做「拼车」生意的好处——只需向Uber付一成佣金,较一般情况下的三成为低。[7]

跟程式斗智 不只为金钱

与程式斗法,不一定是为了名利,也有时候是纯粹「为啖气」,宣泄被程式操控的不满。上述有关Uber的研究,就发现有司机抱怨Uber的系统欠缺透明,令司机不知如何计算收入,难以争取最高收益。[8]此外,司机「拼车」时,系统会为他们指定行车路线,但不论司机及乘客,很多都觉得所指定的线路莫名奇妙,乘客可能因此而怪罪司机,给司机「负评」。[9]司机的另一不满点,是当他们向Uber公司求助时,往往难以接触真人,要与程式打交道。[10]

种种不满,促使了部分司机聚首网上讨论区,探讨Uber系统的运算方式,继而衍生出各种阴谋论,包括认为Uber占司机便宜,于是要作出反抗,提出各种突破程式限制的方法。[11]

人类操弄程式 影响可大可小

无论是基于什么原因,有人与程式斗法,对利用程式展开业务的企业,总会构成或多或少的威胁。以网络搜寻器为例,若其演算法能被轻易参透,人们便会对其失去信心,转投其他服务提供者。其他服务提供者,也会针对完善演算法不力的公司,抢占市场份额。例如Dogpile就声称能网罗及分析Google及雅虎等主要服务提供者的搜寻结果,提供最合乎用家需要的搜寻答案。[12]

程式被看穿,还可能演变成公关灾难。被个别政治群组试图操控搜寻结果的Google,便因此成为众矢之的,包括指其提供含有仇恨、颠倒黑白内容的搜寻结果,形同图书馆职员向查询犹太教资讯的用家,提供仇恨犹太教的书籍。[13]由此可见,若科网巨企无法控制各种虚假资讯、仇恨思想的传播,对社会和企业本身,后果也可大可小。

采取主动 促使程式改变

当然,企业不会坐以待毙,会不时更新演算法,令寻找程式漏洞的人疲于奔命。 Google现时用以判断网页排名的因素,就超过200个。[14]另外,上述Uber司机的占便宜点子,也仅是司机在网上讨论区「吹水」[15],未知是否可行,Uber也声称有技术防止司机推高车资。[16]

演算法要改进,用家也可以集众人之力,防止恶意破坏。计算社会学家J. Nathan Matias,曾与讨论区Reddit的版主合作进行实验,在一些新闻帖文内贴出讯息,鼓励网民核实相关帖文的内容,结果发现Reddit的演算法在计算帖文的排名时,会给予这类帖文较低分数[17],反映Reddit的演算法会因应网民的反应,减少某些资讯的传播。[18]

在讯息能够轻易广传的世界,如何处理有害资讯,是一个棘手问题。肆意干预企业运作,只容许某些消息曝光,可能会沦为言论审查;任由谣言和鼓吹他人作恶的讯息散播,后果同样不堪设想。除了寄望政府和企业监察,用家也需运用选择权,共同建立良好的网上文化。再广泛一点看,用家的行为固然可以影响程式的判断,然而程式背后的原理,却未必公开。当人类的日常生活愈来愈受程式的决定左右,社会是否应该要求开发程式者公开程式的运作原理?公开后又会否令程式更易被恶意操控?这些问题,在未来都会成为争议焦点。

1 "AlphaGo," Deepmind, https://deepmind.com/research/alphago, accessed August 24, 2017.
2 James Vincent, "Did Elon Musk’s AI champ destroy humans at video games? It's complicated," The Verge, August 14, 2017, https://www.theverge.com/2017/8/14/16143392/dota-ai-openai-bot-win-elon-musk.
3 Peter Lee, "Learning from Tay's introduction," Official Microsoft Blog, https://blogs.microsoft.com/blog/2016/03/25/learning-tays-introduction, last modified March 25, 2016; Alex Hern, "Microsoft scrambles to limit PR damage over abusive AI bot Tay," The Guardian, March 24, 2016, https://www.theguardian.com/technology/2016/mar/24/microsoft-scrambles-limit-pr-damage-over-abusive-ai-bot-tay.
4 Peter Lee, "Learning from Tay's introduction," Official Microsoft Blog, https://blogs.microsoft.com/blog/2016/03/25/learning-tays-introduction, last modified March 25, 2016;"Microsoft 'deeply sorry' for racist and sexist tweets by AI chatbot," The Guardian, March 26, 2016, https://www.theguardian.com/technology/2016/mar/26/microsoft-deeply-sorry-for-offensive-tweets-by-ai-chatbot; Rob Price, "Microsoft is deleting its AI chatbot's incredibly racist tweets," Business Insider, March 24, 2016, http://uk.businessinsider.com/microsoft-deletes-racist-genocidal-tweets-from-ai-chatbot-tay-2016-3.
5 "More guidance on building high-quality sites," Google Webmaster Central Blog, https://webmasters.googleblog.com/2011/05/more-guidance-on-building-high-quality.html, last modified May 6, 2011; R.L. Adams, "SEO 2017: 8 Crucial Rules For Dominating Google's Search Results," Forbes, November 29, 2016, https://www.forbes.com/sites/robertadams/2016/11/29/seo-2017-8-crucial-rules-for-dominating-googles-search-results/#4f4dcb767415; Alex Chris, "How to Improve Your Page Ranking for a Specific Keyword," Content Marketing Institute, http://contentmarketinginstitute.com/2016/01/improve-page-ranking-keyword, last modified January 6, 2016; Sarah Perez, "Report: Recent App Store Algorithm Change Points To Crackdown On 'Keyword Stuffing',” TechCrunch, https://techcrunch.com/2015/07/28/report-recent-app-store-algorithm-change-points-to-crackdown-on-keyword-stuffing, last modified July 28, 2015.
6 Olivia Solon and Sam Levin, "How Google's search algorithm spreads false information with a rightwing bias," The Guardian, December 16, 2016, https://www.theguardian.com/technology/2016/dec/16/google-autocomplete-rightwing-bias-algorithm-political-propaganda; Carole Cadwalladr, "Google, democracy and the truth about internet search," The Guardian, December 4, 2016, https://www.theguardian.com/technology/2016/dec/04/google-democracy-truth-internet-search-facebook.
7 Mareike Möhlmann, Lior Zalmanson and Ola Henfridsson, "Hands on the Wheel: Navigating Algorithmic Management and Drivers' Autonomy," May 2017, p. 12. 英国华威大学经智经询问后提供上述研究文章。
8 同7,第9页。
9 同7,第9页。
10 同7,第8页。
11 同7,第10至13页。
12 "About Dogpile," Dogpile, http://www.dogpile.com/support/aboutus, accessed September 26, 2017.
13 Carole Cadwalladr, "Google, democracy and the truth about internet search," The Guardian, December 4, 2016, https://www.theguardian.com/technology/2016/dec/04/google-democracy-truth-internet-search-facebook.
14 "Penguin is now part of our core algorithm," Google Webmaster Central Blog, https://webmasters.googleblog.com/2016/09/penguin-is-now-part-of-our-core.html, last modified September 23, 2016.
15 同7,第12页。
16 Ben Chapman, "Uber drivers work together to create price surge and charge customers more, researchers find," The Independent, August 2, 2017, http://www.independent.co.uk/news/business/news/uber-drivers-work-together-price-surge-go-offline-charge-customers-more-game-app-supply-demand-a7872871.html.
17 J. Nathan Matias, "Persuading Algorithms With an AI Nudge," CivilServant, https://civilservant.io/persuading_ais_preserving_liberties_r_worldnews.html, last modified February 1, 2017.
18 同17。