Conversational AI uses doubt and belief updates to improve answers
Truth for Believable AI: Expressed Doubt, Provenance, and Belief Revision as an Engineerable Stance
Computation and Language
Summary
Many chatbots give answers as if they are always certain, which isn’t always true or helpful. The authors explored a way to add a behavior layer that lets AI express doubt, show where information comes from, and update its beliefs based on new information. They tested this with simulated data and a real language model, finding that some features work better than others. Their findings help make AI responses more nuanced, but more work is needed before real users can judge trust or believability. The research focuses on how the system handles truth claims internally rather than whether people find it more believable.
What this means in practice
- •For chatbot developers: Add mechanisms for chatbots to express doubt and update knowledge transparently during multi-turn conversations.
- •For customer support teams: Deploy conversational agents that indicate uncertainty and track information sources to improve response quality and accountability.
Authors
Sebastian Cochinescu
Abstract
Conversational agents often express answers in a uniformly confident register. We test whether expressed uncertainty, provenance-aware assertion, and explicit belief revision can be implemented as a behavior layer over a fixed language model; we do not test believability or trust. The layer combines three epistemic states, per-claim confidence and typed provenance, a provenance-gated expression rule, and a persistent revision store with auditable acknowledgments and partial resistance to false corrections. We evaluate it on a constructed, mechanically scored multi-session benchmark using a synthetic model and Qwen2.5-0.5B-Instruct. The synthetic instrument passes all five checks. On the real model, acknowledgment soundness, a by-construction guarantee, holds in 100% of cases, and true corrections are accepted more often than false ones (0.44 vs. 0.15 on held beliefs; 0.875 vs. 0.420 including rule-accepted corrections of unheld facts), but the pre-specified expression-fidelity, contradiction-separation, and provenance margins fail. A disclosed post hoc analysis shows that expression gated on mean answer-token probability ranks correctness below chance end to end (AUC 0.41, conversation-clustered), whereas gating on sampling consistency discriminates (AUC 0.66). A consistency-gated configuration selected from this finding and evaluated under a separately committed protocol meets the conversation-level manipulation and capability-equivalence criteria and replicates on a redrawn conversation set. The manipulation result is selection-dependent, and both criteria remain unresolved when uncertainty is clustered over the 60 facts. The supported conclusions are limited to the by-construction audit guarantee, store-dependent partial correction discrimination, and a benchmark- and model-specific failure of token-probability gating; scaling the fact base is required before human evaluation.