When Should AI Read the Room? Public Perceptions of Social Intelligence in AI Agents

URL SCAN: When Should AI Read the Room? Public Perceptions of Social Intelligence in AI Agents
FIRST LINE: AI researchers have been advancing socially intelligent AI agents (Social-AI) across embodiments, from chatbots to physical robots.

DISSECTION

This is a social science study conducting a survey (N=200) on how ordinary people perceive AI agents that exhibit social intelligence. The paper catalogs layperson comfort levels, concerns, and contextual preferences around "Social-AI" deployment.

What the Text Is Actually Doing

It is performing governance legwork for the Social-AI industry. By mapping public perception, it identifies friction points that need smoothing and acceptance thresholds that need engineering. The paper's framing—framing public discomfort as a governance challenge to be addressed—positions the research as infrastructure for deployment expansion, not resistance to it.

The "support-adoption gap" finding (people support Social-AI for others but resist it for themselves) is presented as a problem to solve. The DT lens reads this as the human nervous system working correctly: the gap is a signal that the technology is being trusted to do things it shouldn't, and the public's own reluctance is more epistemically valuable than their abstract support.

The Core Fallacy

The paper assumes that public perception of AI social intelligence is the relevant variable for governance. It treats perception as a barrier to be understood and then managed.

The actual variable is not perception—it is structural dependency. The paper never asks: what happens when you don't get to opt out? The entire "acceptance" framework assumes a voluntary adoption context. It does not engage with the phase where Social-AI is embedded in healthcare, education, eldercare, and hiring pipelines, and the "personal use" refusal options have been eliminated by institutional integration.

Public acceptance becomes irrelevant when the architecture makes refusal economically or structurally impossible.

Hidden Assumptions

1. Deployment is inevitable and the question is how to smooth it. The paper never interrogates whether widespread Social-AI deployment is desirable. It treats it as a fixed destination and concern-maps the path.

2. Social intelligence in AI is a real property, not a performance. The paper takes "social intelligence" as a construct to be measured in AI agents, as if the AI has interior states analogous to human social cognition. It never problematizes this premise. The AI performs social cues. That is not social intelligence. It is sophisticated mimicry that may be functionally indistinguishable from social intelligence in effect—but the paper treats the functional equivalence as settled, not as the central social threat.

3. The survey subjects are sovereign choosers. The entire framework assumes individuals can meaningfully evaluate and accept or reject Social-AI. This collapses under any serious account of institutional leverage, asymmetric information, and coercive adoption contexts.

Social Function

Ideological anesthetic with a thin veneer of critical inquiry. The paper stages concern about AI governance, names risks, catalogs concerns, and then implicitly routes the solution back to "better deployment design" and "contextual appropriateness"—never to halting or restricting deployment.

It performs the ritual of public interest research while functioning as a deployment optimization study.

THE VERDICT

The paper is legibility infrastructure for the Social-AI industry. It maps where public resistance is high and frames that resistance as a design problem, not a structural warning. The "support-adoption gap" is not a puzzle—it is the most important data point in the paper, and the paper treats it like a bug rather than the public's most honest signal.

Under the Discontinuity Thesis, Social-AI is not primarily a perception management challenge. It is a structural displacement challenge: as AI embeds in caregiving, education, and human coordination contexts, the question is not whether people "accept" it but whether the economic and institutional architecture makes refusal existentially costly.

The paper will be cited in ethics frameworks and governance documents as evidence that public attitudes are being incorporated into AI development. It will serve that function while doing nothing to alter the trajectory.