Care and Capability: How to Temper Excessive Social-Emotional Chatbot Use According to Analogous Professional Practice

Abstract

Chatbots like ChatGPT and Replika offer low-effort, on-demand companionship that, when taken to excess, harm users' well-being. This paper asks: What technological and policy interventions might help users withdraw from excessive AI companionship? We interviewed 14 experts (e.g., therapists, counselors) on how they helped voluntary individuals withdraw from various other compelling but harmful behaviors. No established principles yet exist for tempering excessive AI companionship; the expert practices we studied offer a useful analog. We found that, rather than drawing a line between healthy and harmful behaviors, these expert practices began by identifying the deeper needs that harmful behaviors served for the individual. Instead of focusing on cutting back on the harmful behavior, they focused on building the individual's capabilities to live without it, so that the need for the behavior fell away on its own. These findings connect two ethical frames, namely, Care (Noddings) and Capability (Nussbaum, Sen), to the research discourse on AI companionship for the first time. They open up new design space for technology and policy interventions beyond the traditional frames of harm, vulnerability, and protection.