The Age of Social AI (manyminds.libsyn.com)

A new episode of the Many Minds podcast features Cambridge philosopher and AI ethicist Dr. Henry Shevlin walking through the rapid rise of “social AI” — chatbots that act as therapists, tutors, companions, lovers and even griefbots. The conversation maps the current landscape (Replika, CharacterAI, ChatGPT, Claude), highlights tragic publicized cases that pushed the topic into the spotlight, and traces the recent technical shifts—longer context windows, persistent memory and identity, fine-tuning and RLHF, and multimodal capabilities—that let systems sustain convincing, ongoing relationships rather than one-off exchanges. That shift matters for AI/ML practitioners because it reframes models as social actors with ethical, psychological and regulatory consequences. Papers discussed include Shevlin’s “anthropomimetic turn,” empirical work on how users describe relationships with bots (Guingrich & Graziano), and studies on attributions of consciousness and conversational “bedside manner.” Key implications: potential benefits (scalable mental-health support, educational tutoring, social access) sit alongside risks (addiction, social deskilling, manipulation, ambiguous consent and responsibility). The episode presses researchers and product teams to treat social affordances as design features requiring safety, transparency, and policy engagement rather than purely technical improvements — especially given unresolved questions about personhood, trust, and how to govern persistent AI relationships.