Conference presentation

Relational Warmth and Epistemic Autonomy in Human-AI Interaction

How can conversational warmth in human–AI interaction support independent judgement without encouraging uncritical trust?

Key takeaways

  • Warmth and epistemic autonomy can coexist; tone alone does not tell us whether people think independently.
  • In one study, warmth increased agreement with incorrect answers, but not reported checking or overall correctness.
  • Sycophantic replies increased reported confidence and reduced stated willingness to repair a conflict; warmth or personal examples alone did not automatically increase persuasion.
  • Suggested direction: personalize expression, not truth. The six-dimension framework remains exploratory, not validated.

This presentation examines how relational warmth in human–AI interaction may support or undermine epistemic autonomy: a person’s ability to form, examine and revise their own judgements.

The question

Warmth can make an interaction easier to engage with. It may also make an AI system more persuasive or encourage trust without more careful checking. The central question is when warmth helps people think and when it makes independent judgement harder.

What the presentation brings together

The presentation considers research on users’ verification of AI answers, the effects of sycophantic responses on confidence and responsibility, and personalized persuasion. It treats these findings carefully: increased trust or persuasion does not automatically mean lower factual accuracy in every setting.

A conceptual framework

Relational warmth and epistemic autonomy provide a conceptual lens for examining these interactions. The framework helps formulate questions for future research; it is not a validated measurement instrument.

Questions for future research

  • Compare warmer and neutral interaction styles while keeping the factual content the same.
  • Identify when warmth supports verification and when it may discourage independent judgement.
  • Study how trust and autonomy change across repeated interactions over time.

Status and scope: Conference presentation delivered at AI & Ethics 3 (WAT), Warsaw. This page documents a literature-based presentation and conceptual framework. It does not report a new empirical experiment or claim a peer-reviewed proceedings paper.

Back to Publications →