Claims that are accompanied by semantically-related, but otherwise non-probative, photos are typically deemed more trustworthy than claims without photos—the truthiness effect. But can truthiness cause havoc in legal contexts by biasing people’s perceptions of witness evidence? In two pre-registered experiments, we used authentic witness statements and AI-generated images from the Exhibit-AI campaign (https://www.exhibitai.com.au/). Participants read two statements made by asylum-seekers who documented their experiences of being housed in Australian offshore detention centres. One statement was presented with AI-generated images illustrating the asylum-seeker’s experience and the other statement was presented alone. Participants were told that the images were AI-generated. Participants rated the witness statements on multiple dimensions, including credibility. Experiment 1 (N=200) showed that the AI-generated images did not affect perceptions of the witnesses’ statements. Data collection for Experiment 2 continues. The results will help to refine theoretical accounts of truthiness and determine the boundary conditions of the truthiness effect.

During the conflict in Northern Ireland of the late 20th Century, audio recordings of 10 controversial political groups were banned by the British government from being broadcast on British airwaves. Interestingly, video recordings of the people in question were instead overdubbed by paid actors implying that the political figures’ original voices were more convincing, trustworthy, or believable than those of the actors. This study uses AI-generated audio to examine the efficacy of this type of censorship. We carried out a between-subjects experimental study (n = 636), where we tested the effects of three different types of vocal manipulations in comparison with unedited footage of two well-known Northern Irish politicians. We measure the perceived credibility and legitimacy of the statements as well as how charismatic the politicians are viewed. There are numerous implications for this work in future cognitive research on perceptions of translation, particularly translated and overdubbed political coverage.

People use their smartphones in everyday life to film unfolding live events as they experience them. What happens when the event they film involves a crime? This study examined whether videorecording scenes in which a crime occurs helps or hurts peoples’ memory for what they witnessed. Participants witnessed a scene involving a crime either while filming the scene as it unfolded with a smartphone or viewing it unimpeded, without a smartphone. Later they had to report what they remembered. Results showed that people remembered less about the crime itself and about the suspect when they videotaped the scene with a smartphone than when just observed the scene without using a smartphone. Recording the scene resulted in fewer correct identifications of the suspect in a photo lineup and in lower confidence in their memory accuracy. Memory for details unrelated to the crime was not negatively impacted by phone usage.