Study reveals 'support-adoption gap' for socially intelligent AI agents
200 US adults surveyed: people want AI that reads the room—for others, not themselves.
Researchers from Carnegie Mellon, University of Washington, and other institutions surveyed 200 US adults to understand public perceptions of social intelligence in AI agents (Social-AI). The study, titled 'When Should AI Read the Room?', reveals that most participants have already interacted with AI they consider socially intelligent. Crucially, participants base these judgments on observable behaviors (e.g., tone, responsiveness) rather than abstract beliefs about AI having intentions or consciousness. This behavioral focus has direct implications for how developers design and deploy socially-aware systems.
A standout finding is the 'support-adoption gap': participants broadly supported the existence of Social-AI agents—such as empathetic chatbots or socially-aware robots—for society at large, but were significantly less willing to use them personally. This gap suggests that while the concept of socially intelligent AI is appealing in theory, real-world trust and comfort levels remain low. The paper also highlights concerns about context, role appropriateness, and risks to end users, offering actionable insights for AI governance. The work underscores that technical progress alone won't drive adoption—public perception and trust are equally critical.
- Survey of 200 US adults shows most have encountered AI they perceive as socially intelligent.
- Participants judge AI social intelligence based on observable behavior, not beliefs about agency or intent.
- A 'support-adoption gap' exists: people support Social-AI for others but resist using it themselves.
Why It Matters
As AI agents become more socially aware, trust gaps and adoption hesitancy will shape regulatory and design priorities.