Source record

Published by arXiv cs.HC (T1 source). The original is at https://arxiv.org/abs/2608.20828.

Pipeline notes

The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.

Analyzes a 240-participant human-AI decision-making dataset and finds that 67.3% of trial-level belief changes are exactly zero, with 76.4% under five percentage points. Proposes a 'belief update gate' decomposition separating whether reported beliefs change at all from how they change conditionally.

A measurement-aware decomposition worth noting for researchers designing human-AI interaction studies or interpreting belief-updating data, though the findings are methodological rather than directly applicable to day-to-day agentic workflows.

Cited by

No citations on record.