We study how people judge information and where mistakes go unnoticed.
Checking the evidence
Our work examines peer review, scientific publishing, news, and healthcare records. We develop methods to assess reliability and flag problems for closer review.
Why it matters
People rely on institutions to distinguish a well-supported claim from a plausible account. AI can help organize large amounts of information, while also making mistakes harder to notice when fluent text obscures missing evidence. Universities, publishers, and newsrooms need ways to examine both the claims they encounter and the processes used to check them.
Questions we ask
- What evidence do people use to judge a claim, and when do important gaps go unnoticed?
- How can AI help reviewers identify problems while making its own uncertainty visible?
- How can an institution trace a summary or assessment back to the evidence behind it?
Our approach
We bring language analysis and computational methods to the study of evaluation. The challenge includes defining what a check can establish, comparing automated assessments with human judgment, and understanding the costs of missed problems and false alarms. A useful assessment should make its evidence and limitations clear enough for someone else to inspect.
Support this research
Support could help students and faculty develop and evaluate methods for checking information, document their limitations, and compare how they work in different institutional settings. Conversations can begin with a concrete evaluation problem and the people who would use the results.
Discuss supporting this researchUpdated