Embedded Files
Building trust through confident & justified claims
Building trust through confident & justified claims
An analysis of how people judge sources who are wrong for the right reasons or right for the wrong ones.
An analysis of how people judge sources who are wrong for the right reasons or right for the wrong ones.
The challenge
The challenge
How do people evaluate the credibility of confident but fallible sources?
In an era of overconfident communicators, this research informs models of how human and artificial systems convey confidence and manage uncertainty.
My role & methods
My role & methods
At the Institute of Communication and Cognition Sciences, I designed and implemented two online experiments (Qualtrics), recruited participants (Prolific), ran data collection, and analyzed the results.
Participants judged the testimony of two fictional witnesses: one confident, one unconfident. The scenarios varied whether the confident speaker was (1) inaccurate but well-justified, or (2) accurate but weakly justified.
Key insight: Confidence builds trust only when backed by evidence.
Key insight: Confidence builds trust only when backed by evidence.
Confidence only backfired when it exceeded the supporting evidence, even if the information itself was accurate.
This principle illustrates how human and AI systems can signal certainty responsibly, shedding light on the mechanisms behind AI "hallucinations".
Impact & recognition
Impact & recognition
Invited speaker, TRUEDEM (Trust in European Democracies) Webinar Series (2025)
Presented at numerous conferences, among them Experiments in Linguistic Meaning (ELM) 2
Published as part of the special issue “Trust and Trustworthiness: individual and institutional dimensions” in Philosophical Psychology (2024)
Published in Ars interpretandi (2023), discussing the implications for legal testimony
Experiments & results
Experiments & results
Scenario 1: Confident, wrong but justified
Scenario 1: Confident, wrong but justified
One witness is confident, the other unconfident.
One witness is confident, the other unconfident.
Both witnesses are wrong.
Both witnesses are wrong.
Both witnesses have strong evidence.
Both witnesses have strong evidence.
A well-justified error did not damage the confident witness’ credibility.
A well-justified error did not damage the confident witness’ credibility.
Scenario 2: Confident, right but unjustified
Scenario 2: Confident, right but unjustified
One witness is confident, the other unconfident.
One witness is confident, the other unconfident.
Both witnesses are right.
Both witnesses are right.
Both witnesses have weak evidence.
Both witnesses have weak evidence.
Poorly-justified accuracy damaged the confident witness’ credibility.
Poorly-justified accuracy damaged the confident witness’ credibility.
Technical skills
Technical skills
Statistical analysis (R)
Statistical analysis (R)
I used a Generalized Linear Mixed Model to analyze changes in binary choices over time, and a Cumulative Link Mixed Model to analyze changes in ordinal credibility ratings, accounting for repeated measures.
Page updated
Google Sites
Report abuse