In “Extracting the Truth From Conflicting Eyewitness Reports,” Waubert de Puiseau et al. (2012) examine how researchers can reconstruct an event from several imperfect accounts. Thirty participants watched one of two versions of a video that provided different amounts of information about an argument. They then answered questions about the event. The researchers compared majority voting with a model based on cultural consensus theory, which estimates differences in witnesses’ knowledge from their response patterns. The model reconstructed the event more accurately than majority voting in the tested conditions. However, its assumptions become problematic when people give similar wrong answers because of shared misinformation or expectations. This distinction matters: agreement can reflect knowledge, but it can also reflect a common mistake.
This article helps me clarify the research question behind Trust and Know. In my web experience, participants will encounter a magic reveal and explain how they think the guide predicted their choice. Several participants might offer the same explanation because the performance directs their attention toward similar clues. I want to examine whether AI can recover the actual mechanism from these accounts, especially when their interpretations are misleading. To make this test useful, I plan to ask participants separately what they noticed and how they explain it. Someone could misunderstand the mechanism while still reporting an observation that helps identify it. I also want to distinguish trust in the guide from confidence in an explanation and actual understanding. Giving the AI the same accounts with and without these ratings would help test whether the ratings improve its inference or distract it from useful evidence. The study therefore gives me a starting point, while my project extends the question to open-ended explanations of a deliberately misleading experience.
The feedback I received on my preproduction binder was basicly positive, with no specific revisions suggested. Preparing the binder helped me organize the experience into a clear sequence and connect the storyboard to my research questions. My next step is to turn the storyboard into a working demo and check whether the drawing, calibration, and magic reveal fit together smoothly. I also need to refine the questions that distinguish participants’ observations from their explanations, and their trust in the guide from their understanding of its mechanism. I will discuss these choices with my committee as I prepare for the pilot.
Reference
Waubert de Puiseau, B., Aßfalg, A., Erdfelder, E., & Bernstein, D. M. (2012). Extracting the truth from conflicting eyewitness reports: A formal modeling approach. Journal of Experimental Psychology: Applied, 18(4), 390–403. https://doi.org/10.1037/a0029801