Governing AI Companionship: Lessons from Social Media Regulation

Abstract

Since the proliferation of AI companionship following COVID-19, the calls to regulate AI companion services have only grown in urgency. The attempted and actual suicides of several individuals, primarily children, have been linked to AI companions like ChatGPT and Character.AI, and there is also growing evidence that AI companionship enables addictive behaviours. This paper provides conceptual scaffolding for the task of regulating AI companions by connecting it to the task of regulating social media. By examining how regulators tackle the problems of harmful content and addictive designs on social media platforms, we can learn useful lessons for how the same problems can be tackled as they manifest in AI companionship. Our first lesson concerns the moderation of harmful content, where we find that social media regulatory measures like notice-and-takedown and age restrictions are applicable, though their efficacy is limited by the technology needed for enforcement. The application of intermediary immunity would require greater legal certainty via binding decisions or enforcement actions. The second lesson concerns restrictions on addictive designs, where we find that antitrust measures and litigation are likely to have a positive effect, though the current and potential impact of legislative measures remain uncertain. We conclude by identifying problems posed by AI companions - the accumulation of social power, and the macrosocial implications of proliferating AI companionship - that fall beyond social media regulation and necessitate novel solutions.