From Observation to Intervention: Identifiability of Latent User Bias in Human-AI Interaction
3.60T1 sourcearXiv cs.HC
Source record
Published by arXiv cs.HC (T1 source). The original is at https://arxiv.org/abs/2610.13185.
Pipeline notes
The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.
SummaryTheoretical paper proving that passive observation cannot distinguish a cognitively biased user from an unbiased one in a transformed environment, and that active adversarial interventions of finite length are required to make their action distributions separable.
Why it mattersFormal non-identifiability result challenges the common assumption that user behavior reliably reflects preferences, with implications for alignment and feedback-loop design.
Cited by
No citations on record.
