The starting point
Software can help organize a discussion, but it cannot by itself establish a safe therapeutic relationship. A group needs clear roles, consent, a human response to distress, and realistic privacy expectations. Do not ask participants to disclose sensitive experiences simply to personalize an algorithm.
What is hypothetical
This discussion of over-reliance on ai moderators includes a design scenario from the manuscript. Read proposed capabilities conditionally. A real product would need evidence for its exact use, a clear account of errors and limitations, and meaningful control for the person using it. An engaging demonstration alone would not establish a health or learning benefit.
Over-Reliance on AI Moderators
Another risk is the potential over-reliance on AI moderators. While AI can be effective at guiding discussions and providing support, it lacks the emotional depth and nuance of a human facilitator. There is a risk that people may come to depend solely on AI, rather than seeking out human connection and support.
Imagine someone who only attends AI-facilitated therapy sessions and starts to feel disconnected from in-person relationships. The risk is that they may miss out on the depth of connection that comes from sharing physical space with others—the warmth of a hug, the comfort of seeing a compassionate face.
Human relationships are irreplaceable, and technology should enhance, not replace, our natural bonds.
A concrete way to think about it
A group also needs a human route for concerns. Decide who can pause a discussion, correct an unfair interpretation, respond to distress, and arrange appropriate help. Software should not be the only place a participant can appeal a decision.
Sources & further reading
- NIMH: Psychotherapieswww.nimh.nih.gov
- NIST: Generative artificial intelligence risk profilenvlpubs.nist.gov