Studies have demonstrated that lying can detrimentally impact memory. This detrimental effect depends on the type of strategy (e.g., false denials, feigned amnesia, fabrication) used to lie. So far, studies have investigated the effects of each of these strategies on memory, but no study has tested what happens when strategies are used simultaneously. The present experiment aimed to examine this issue. Participants watched a mock crime video and, then during an interview, were instructed to either tell the truth concerning the crime or lie about it by using all three strategies of false denials, feigned amnesia, and fabrication. A third group did not reply to any questions. After two days, all participants were asked to recall the crime honestly. We expected to find differences among groups, with a memory impairment for the crime in people who lied during the first session. Our results will be discussed considering their legal implications.

Students are increasingly using AI platforms to support their learning and complete assessments. In response, educational institutions are requiring students to report exactly how, where and why they used AI to complete an assessment. We explored whether students can accurately remember and report how they used generative-AI in a written task. In two studies, students imagined they were completing an assignment. They produced text in ChatGPT then attempted to improve its clarity, coherence and succinctness. Following a delay, they were asked how much they edited the ChatGPT output before “submitting” it. They also reported their attitudes towards AI-use in educational settings and completed a self-deception scale. Preliminary results (N = 76) suggest students systematically over-report how much they have edited AI-generated text. Based on findings around unethical amnesia and self-deception, we expect the accuracy of students’ reports to be mediated by their attitudes towards AI-use and their propensity to self-deceive.

Deception detection is a field with important theoretical and practical relevance. However, findings observed in one experiment typically fail to replicate or translate to other scenarios. This, we argue, stems from insufficient information in research programmes to make valid causal inferences and predictions. The methodologies employed in the field fail to account for important sources of variance to strongly impact deception studies, often unbeknownst to the researcher. These include variance in senders, judges, and the task itself. We employ two high-precision experiments aimed at establish guidelines for conducting deception research and making inferences. By varying the number of senders, judges, and data type to values higher than often employed, and utilizing robust (Bayesian) analyses we estimate the information needed for stable causal inferences and predictions. The results hold practical significance in forensic and legal contexts, offering insights into the reliability and risks associated with deception detection findings.