Personal AI, deception, and the problem of emotional bubbles

Abstract Personal AI is a new type of AI companion, distinct from the prevailing forms of AI companionship. Instead of playing a narrow and well-defined social role, like friend, lover, caretaker, or colleague, with a set of pre-determined responses and behaviors, Personal AI is engineered to tailor itself to the user, including learning to mirror the user’s unique emotional language and attitudes. This paper identifies two issues with Personal AI. First, like other AI companions, it is deceptive about the presence of their emotions, which undermines the moral value of companionship. Second, Personal AI leads to a distinctly new form of deception concerning the origins of its emotions. Its emotional attitudes appear to belong to it, when in fact they are only reflections of the user. This results in what I dub “emotional bubbles”—the false impression that personal emotions are externally validated—which have at least two troubling implications. First, emotional bubbles prevent us from encountering emotional attitudes that differ from our own, which is likely to cripple emotional growth and the ability to form diverse social and emotional relationships. Second, if we assume, as some philosophers claim, that shared emotions are constitutive of shared values, it follows that Personal AI subverts joint moral deliberation. Users believe their personal values are externally validated, when they are only validated by themselves. Because of the absence of technovirtues able to handle this problem, I suggest that we proceed very cautiously with the development and marketing of Personal AI.