Tay is you. The attribution of responsibility in the algorithmic culture.

Abstract

Social media have changed the communication practices by creating an acute need for continuous interaction. The use of social chatbots is growing as an effective way to communicate with publics. Bots have become social actors and then, someone must account for their actions. Since responsibility is bounded to agency and rationality, it cannot be directly attributed to bots. Who should be held responsible for non-human beings' actions, particularly when the consequences of these actions are negative? We address this controversy from both theoretical and empirical perspectives. Firstly, we discuss the adequacy of the notions of moral responsibility and accountability regarding non-human artificial agents, as they are ruled by complex, intentionally opaque and unpredictable interactions and processes. We do it from the two approaches currently predominant: context-dependent and structuralist. Secondly, we draw on the assumption that the failure of a computer system is an opportunity to gain knowledge about the interested powers behind its design and functioning. Then, taking the concept of media frame as an implicit way of spotting the agent of the story, we perform an exploratory analysis on how responsibility was attributed by the media in the paradigmatic case of the transformation of Tay, a chatbot launched by Microsoft in 2016, turned into a racist, Nazi and homophobic hate speaker. Our results illustrate the difficulties media experienced in consistently attributing the responsibility for the chatbots' malfunction. Results show the discourse is, in general, simplistic, non-critical and misleading, and tends to depict a reality that favors business's interests. We conclude that, while all the actors interacting with the chatbot share the responsibility of its actions, it is only Microsoft who must account for these actions, both retrospectively and prospectively.

Artificial Intelligence, Responsibility Attribution, and a Relational Justification of Explainability

“…For example, in the case of Microsoft’s 2016 Twitter chatbot Tay (a controversial chatbot which, after interaction with users, started to produce racist and misogynistic comments and had to be shut down), there were many parties involved: developers, designers, but also the company and those interacting with the chatbot. But in their analysis of the case, Suárez-Gonzalo et al ( 2019 ) point mainly to the designers, developers, and those who decided to insert the bot, rather than the Twitter users who interacted with the bot. While it is not entirely clear to me why the latter would be a lot less responsible, it is clear that distributed responsibility does not imply that responsibility is and should always be distributed equally.…”

“…Second, responsibility attribution may be difficult because one or more parties may have an interest in misrepresenting their contribution and in trying to evade responsibility. Suárez-Gonzalo et al ( 2019 ) rightly emphasize the social character of what happened and identify some of the interests at play and the role of the media. In this case, they argue, the media discourse served the interests of Microsoft (Suárez-Gonzalo et al 2019 , 8).…”

“…Suárez-Gonzalo et al ( 2019 ) rightly emphasize the social character of what happened and identify some of the interests at play and the role of the media. In this case, they argue, the media discourse served the interests of Microsoft (Suárez-Gonzalo et al 2019 , 8). Similarly, in the transportation sector there are also many parties involved and different parties may have different interests when it comes to presenting their part in, for instance, an accident with an airplane or a self-driving car.…”

Modelling a socialised chatbot using trust development in children: lessons learnt from Tay

“…Although it is debatable whether the consequences of Tay's manipulation were severe [9] or particularly harmful, the issue highlighted the susceptibility of machine learning algorithms to malicious manipulation and the limitation of autonomous systems currently. If placed in an even more consequential situation, there is a need to constrain the learning so that it only operates within a 'safe' domain.…”

Twitter Engagement in Media Organizations: The Case of the Greek National Broadcasting Corporation

“…Social media is not only a channel of communication with the public (Khan et al 2021), it also constitutes an integral part of online information with both the production and the dissemination of news, fundamentally altering the standards of human communication and networking (Zhang et al 2016). Their widespread use has changed communication practices, creating a deep-seated need for continuous interactivity (Suárez-Gonzalo et al 2019) and at the same time creating a new relationship between news outlets and their audience (Reinhold and Alt 2012). It is also evident from the international literature that journalists are increasingly embracing the characteristics of social media, looking for new ways of approaching their audiences (Broersma and Eldridge 2019).…”