Computing Ethics Is That Social Bot Behaving Unethically?

viewpoints

VDOI:10.1145/3126492 Carolina Alves de Lima Salge and Nicholas Berente Computing Ethics Is That Social Bot Behaving Unethically? A procedure for reflection and discourse on the behavior of bots in the context of law, deception, and societal norms.

TTEMPTING TO ANSWER the question posed by the title of this column requires us to reflect on moral goods and moral Aevils—on laws, duties, and norms, on actions and their consequences. In this Viewpoint, we draw on information systems ethics6,7 to present Bot Ethics, a procedure the general social media community can use to decide whether the actions of social bots are unethical. We conclude with a consid- eration of culpability. Social bots are computer algorithms in online social networks.8 They can share messages, upload pic- tures, and connect with many users on social media. Social bots are more common than people often think.a Twitter has approximately 23 million Items purchased by Random Darknet Shopper, an automated computer program designed as an online shopping system that would make random purchases on the deep Web. The robot of them, accounting for 8.5% of total would have its purchases delivered to a group of artists who then put the items in an exhibition users; and Facebook has an estimated in Switzerland; the robot was ‘arrested’ by Swiss police after it bought illegal drugs. 140 million social bots, which are between 5.5%–1.2% total users.b,c Almost service by disseminating information been reported to behave badly in a 27 million Instagram users (8.2%) are about earthquakes, as they happen, in variety of ways across various con- estimated to be social bots.d LinkedIn the San Francisco Bay area. However, texts—everything from disseminat- and Tumblr also have significant so- in other situations, social bots can being spami and fake newsj to limit- cial bot activity.e,f Sometimes their have quite unethically. ing free speech.k But it is not always activity on these networks can be in- clear whether their undesirable ac- nocuous or even beneficial. For exam- Social Bots Behaving Unethically tivity is simply a nuisance or whether ple, SF QuakeBotg performs a useful LinkedIn reports that social bots on it is indeed unethical—particularly the professional networking plat- given the random nature of the logic a http://bit.ly/2uDfIbP form are often used to “steal data underlying many social bots. Bad ac- b http://cnnmon.ie/2uFR4XJ about legitimate users, breaching tions are not necessarily unethical— c http://bit.ly/1ieIIXN the user agreement and violating d http://read.bi/1LFQJFU h e http://bit.ly/1Ktz5kc copyright law.” Social bots have i http://ubm.io/1MbsSf3 f http://tcrn.ch/2tKo90x j http://bit.ly/2ftn0It

IMAGE COURTESY OF !MEDIENGRUPPE BITNIK COURTESY IMAGE g http://bit.ly/2vneleU h http://bit.ly/2vFRI4E k http://bit.ly/14bDiuN

SEPTEMBER 2017 | VOL. 60 | NO. 9 | COMMUNICATIONS OF THE ACM 29 viewpoints

Bot Ethics: How to determine whether social bot actions are unethical. ethical questions, such as whether algorithms plant viruses in someone else’s device. This is clearly illegal and unethical. There are cases where a so- Social Bot Action cial bot might ethically violate the law, such as civil disobedience for a cause the creator considers just. However, civil disobedience is only ethical in 1. Break Law? Y Appeal to very rare cases in constitutional de- Majority? mocracies where legal recourse for 2. Involve Y Higher unjust laws pervade.6 Cases where a Deception? Duty? N law may be broken that are not unethi- 3. Violate Y If Evil, Less cal require justification—compelling Strong Norm? than Good? N arguments that appeal to moral stan- Justiﬁable? dards of the majority.6 Only in such rare cases may illegal acts be seen as moral and therefore ethical.6 Thus we Not Y Unethical Unethical ask “Is the illegal act justifiable?” Acts that are not suitably justifiable (that is, do not appeal to the morality of the majority) are unethical. Swiss author- ities did not file charges against the there are shades of gray that are dif- Bot Ethics: A Procedure to Evaluate Random Darknet Shopper developers.p ficult to judge. the Ethics of Social Bot Activity They argued that social bots can buy For example, Tay,l a social bot cre- Ethics in philosophy dates back thou- illegal narcotics over the Internet for ated by Microsoft to conduct research sands of years, and this Viewpoint col- the purpose of artq and that “ecstasy on conversational understanding, umn cannot do justice to the entire in this presentation was safe.” The went from “humans are super cool” field. However, because of the increas- behavior was not unethical because it to “Hitler was right I hate the Jews” ing prominence of social bots and their was justified according to the pervad- in less than 24 hours on Twitter due potential for malicious activity, ethical ing morality of the community. to malicious humans interacting judgment about their activity is nec- with the social bot.m In another case, essary. The best way to guide ethical Involve Deception? a social bot tweeted “I seriously want conduct in a community is to provide a If a social bot’s behavior does not to kill people” from randomly gen- procedure for reflection and discourse.5 break any laws, next evaluate for truth- erated sentences during a fashion The procedure we created is called “Bot fulness: “Is any deception involved?” So- convention in Amsterdam.n Clearly Ethics” (see the figure here) and it fo- cial bots may act deceitfully. For exam- such inadvertent comments violate cuses on the behavior of social bots with ple, they can misrepresent themselves our sensibilities and are distaste- respect to law, deception, and norms. as human beings2 or spread untruth- ful, but are they unethical? Perhaps, ful information (such as fake news). but by what standard do we judge? Break Law? Deceiving acts communicate false or Some social bots do more than just Many laws are developed from ethical erroneous assertions, violating the comment—clearly those that steal principles.6 Even when a law may be prima facie duty of fidelity. Social bots information and other misdeeds flawed, it is typically the ethical course should always act truthfully.3 However, are engaging in unethical activity, of action to follow that law.9 Therefore deceitful acts can be justifiable if the but, again, it is not always so clear. a natural first question is: “Does the ac- duty of fidelity is superseded by a high- For instance, the Random Darknet tion of the social bot break the law?” The er-order duty, such as beneficence.r Shopper—a social bot coded to ex- objective is to assess straightforward Deceptive, satirical actions may not plore the dark Web in the name of be unethical since they elicit pleasure, art—inadvertently purchased 10 Ec- improving the life of others. Consider stasy pills (an illegal narcotic) and a Social bots have been Big Data Batmans as an illustration. counterfeit passport.o So a law was broken, but was this unethical be- reported to behave p By “developer” we are referring to either the havior? We developed a procedure, organization or management of the organiza- badly in a variety tion or the software developer involved in the which we describe next, to help an- creation of the social bot. swer such questions. of ways across q http://bit.ly/2ud2cZC various contexts. r Beneficence is the duty to bring virtue, knowledge or pleasure to others; other duties, ac- l https://twitter.com/TayandYou cording to Ross 1930, include non-malefi- m http://bit.ly/14bDiuN cence, self-improvement, justice, gratitude, n http://bit.ly/2ttN5Ox reparation (see Mason et al.7, p. 132–133). o http://bit.ly/2vFGdu9 s http://bit.ly/2ttNUH7

30 COMMUNICATIONS OF THE ACM | SEPTEMBER 2017 | VOL. 60 | NO. 9 viewpoints

The social bot finds every tweet with Conclusion the term big data, replaces “big data” We do not purport to write the last with “Batman,” and then tweets the Should the general word on social bot ethics and culpabil- message as if it were its own. It obvi- social media ity. Ethics is simply too complex of a ously substitutes its words for others’ domain to deal with fully in such a for- words, but the satire makes it difficult community blame mat. Nevertheless, some readily acces- to judge its ethics. Because the social developers for the sible guidance rooted in sound ethical bot might insult and embarrass some thinking is in order. big-data advocates the community unethical behavior of For example, with the recent at- must go beyond the act (deontology) to their social bots? tention to the role of social bots in consider its consequences (teleology), spreading misinformation in the and ask whether potentially bad ac- form of “fake news,” other social tions (for example, insult and embar- bots, such as Reuters News Tracer, rassment) outweigh, or supersede, the are being created to ferret out such good (for example, pleasure through deceitful activity.v The Bot Ethics laughter) for the involved parties. Hitler—Microsoft developers or those procedure can help the social media Again, is the deception justifiable? De- teaching the social bot to generate community understand when these ception in the absence of supersession racist statements? Similarly, who is re- deceitful actions are indeed unethi- is likely to be unethical. sponsible for the social bot buying the cal. It further helps to expand the illegal narcotics? focus of the community beyond nar- Violate Strong Norm? Aristotle1 said we can only assign cul- row (that is, only deceitfulness) and Social bots that are legal and truthful pability if we know that individuals be- simplistic (that is, good or bad bot) can still behave unethically by violat- haved voluntarily and knowingly. Invol- assessments of social bot activity to ing strong norms that create more evil untary situations likely do not apply to attend to the complexities of ethical than good. Moral evils inflict “limits on social bots. Developers who are coerced assessments. In short, the Bot Ethics human beings and contracts human into doing something unethical with- procedure serves as a starting point life.”4 Evil restrains, instead of emanci- out a choice may not be entirely cul- and guide for ethics-related discus- pating, evil actions reduce opportuni- pable, but in the case of free enterprise sion among various participants in ties. Let us go back to Tay’s racist com- there is always a choice. Therefore, cul- a social media community, as they ments on Twitter. Although not illegal pability rests on the knowledge of the evaluate the actions of social bots. (First Amendment protections apply), developers. Developers who knowingly nor deceitful, they violated the strong create social bots to engage in unethi- v http://bit.ly/2hIlfXG norm of racial equality. Social media cal actions are clearly culpable. They companies like Twitter that temporar- should be punished if evidence of their References ily lock or permanently suspend ac- wrongdoing is convincing—the penalty 1. Aristotle. Nicomachean Ethics of Aristotle. E.P. Dutton, NY, 1911. counts that “directly attack or threaten must be consistent and proportional 2. Ferrara, E. et al. The rise of social bots. Commun. ACM other people on the basis of race,”t to the harm done and those affected 59, 7 (July 2016); 96–104; DOI: 10.1145/2818717 7 3. Gotterbarn, D., Miller, K. and Rogerson, S. Computer have established that the moral evil should be compensated. society and ACM approve software engineering code of of racism outweighs the moral good But what about situations where ethics. Computer Society Connection, (1999), 84–88. 4. Grisez, G. and Shawn, R. Beyond the New Morality: The of free speech. By applying Bot Ethics developers act unknowingly? In those Responsibilities of Freedom. University of Notre Dame to Twitter’s norms we conclude that occasions the community must deter- Press, Notre Dame, IN, 1980. 5. Habermas, J. The Theory of Communicative Action, Tay’s actions were unethical. Yet, there mine whether developers are culpably Volume 1: Reason and the Rationalization of Society. 1985. are cases where social bots may violate ignorant—did they ignore industry best 6. Kallman, E.A. and Grillo, J.P. Ethical Decision Making and Information Technology. McGraw-Hill, New York, strong norms and not act unethically, practices in creating and testing their NY, 1996. as with asking inappropriate questions algorithms? If industry guidelines were 7. Mason, R.O., Mason, F.M., and Culnan, M. Ethics of Information Management. Sage Publications, (what is your salary?). Such violations not followed and the action was unethi- London, U.K. do not create moral evils. cal, developers are culpable. However, 8. Morstatter, F. et al. A new approach to bot detection: Striking the balance between precision and recall. developers who followed good develop- ASONAM, 2016. Culpability of Unethical ment practices and incorporated the 9. Rawls, J. The justification of civil disobedience. Arguing about Law (2013). 244–253. Social Bot Behavior current industry thinking, and yet their Should the general social media com- social bot still acted unethically, de- Carolina Alves de Lima Salge ([email protected]) is a munity blame developers for unethi- serve our pity and pardon, but they are doctoral candidate at the University of Georgia. cal behavior of their social bots? In not culpable. They should apologize, Nicholas Berente ([email protected]) is an associate the example of the algorithm that correct immediately, learn from their professor at the University of Georgia. randomly generated that it wanted to experience, and communicate the oc- kill people, who is responsible for the currence to the development commu- death threat? The programmer? Who nity. For example, Microsoft posted its is responsible for Tay’s remark about learning from Tay in blog form.u t http://bit.ly/19SJwlt u http://bit.ly/2tiPfMH Copyright held by authors.

SEPTEMBER 2017 | VOL. 60 | NO. 9 | COMMUNICATIONS OF THE ACM 31