Artificial Intelligence, Trust, and Perceptions of Agency

The literature on trust among humans requires that the trustee is seen to act with agency, else trust is undefined. In contrast, the literature on confidence in technology does not require that the technology we make ourselves vulnerable to is perceived to have any volition. Modern artificial intelligence (AI) technologies are distinctive in that they are often perceived as agentic to varying degrees, typically more agentic than other technologies but less than humans. We theorize how different levels of perceived agency of the AI affects human trust in AI through three mechanisms. First, a more agentic seeming AI as well as its designer appear more able, and therefore more trustworthy. Second, the trustworthiness perceptions about the AI become more important relative to those about its designer, if the AI is seen as more agentic. Third, the anticipated psychological cost of the AI violating trust increases if the AI is seen as more agentic because of betrayal aversion. These mechanisms imply that making an AI appear more agentic may increase or decrease the trust that humans place in it, and also explain why some interventions to improve human trust in AI are likely to be more robust than others.