{"language":[{"iso":"eng"}],"project":[{"name":"TRR 318 - C1: TRR 318 - Subproject C1 - Gesundes Misstrauen in Erklärungen","_id":"124"}],"doi":"10.3389/fpsyg.2025.1574809","_id":"59755","abstract":[{"lang":"eng","text":"Due to the application of Artificial Intelligence (AI) in high-risk domains like law or medicine,\r\ntrustworthy AI and trust in AI are of increasing scientific and public relevance. A typical conception,\r\nfor example in the context of medical diagnosis, is that a knowledgeable user receives AIgenerated\r\nclassification as advice. Research to improve such interactions often aims to foster the\r\nuser’s trust, which in turn should improve the combined human-AI performance. Given that AI\r\nmodels can err, we argue that the possibility to critically review, thus to distrust, an AI decision is\r\nan equally interesting target of research.\r\nWe created two image classification scenarios in which the participants received mock-up\r\nAI advice. The quality of the advice decreases for a phase of the experiment. We studied the\r\ntask performance, trust and distrust of the participants, and tested whether an instruction to\r\nremain skeptical and review each piece of advice led to a better performance compared to a\r\nneutral condition. Our results indicate that this instruction does not improve but rather worsens\r\nthe participants’ performance. Repeated single-item self-report of trust and distrust shows an\r\nincrease in trust and a decrease in distrust after the drop in the AI’s classification quality, with no\r\ndifference between the two instructions. Furthermore, via a Bayesian Signal Detection Theory\r\nanalysis, we provide a procedure to assess appropriate reliance in detail, by quantifying whether\r\nthe problems of under- and over-reliance have been mitigated. We discuss implications of our\r\nresults for the usage of disclaimers before interacting with AI, as prominently used in current\r\nLLM-based chatbots, and for trust and distrust research."}],"article_type":"original","status":"public","department":[{"_id":"424"},{"_id":"660"}],"date_updated":"2025-05-02T09:31:45Z","keyword":["trust in AI","trust","distrust","human-AI interaction","Signal Detection Theory","Bayesian parameter estimation","image classification"],"year":"2025","volume":16,"type":"journal_article","citation":{"ama":"Peters TM, Scharlau I. Interacting with fallible AI: Is distrust helpful when receiving AI misclassifications? <i>Frontiers in Psychology</i>. 16. doi:<a href=\"https://doi.org/10.3389/fpsyg.2025.1574809\">10.3389/fpsyg.2025.1574809</a>","bibtex":"@article{Peters_Scharlau, title={Interacting with fallible AI: Is distrust helpful when receiving AI misclassifications?}, volume={16}, DOI={<a href=\"https://doi.org/10.3389/fpsyg.2025.1574809\">10.3389/fpsyg.2025.1574809</a>}, journal={Frontiers in Psychology}, author={Peters, Tobias Martin and Scharlau, Ingrid} }","ieee":"T. M. Peters and I. Scharlau, “Interacting with fallible AI: Is distrust helpful when receiving AI misclassifications?,” <i>Frontiers in Psychology</i>, vol. 16, doi: <a href=\"https://doi.org/10.3389/fpsyg.2025.1574809\">10.3389/fpsyg.2025.1574809</a>.","short":"T.M. Peters, I. Scharlau, Frontiers in Psychology 16 (n.d.).","apa":"Peters, T. M., &#38; Scharlau, I. (n.d.). Interacting with fallible AI: Is distrust helpful when receiving AI misclassifications? <i>Frontiers in Psychology</i>, <i>16</i>. <a href=\"https://doi.org/10.3389/fpsyg.2025.1574809\">https://doi.org/10.3389/fpsyg.2025.1574809</a>","mla":"Peters, Tobias Martin, and Ingrid Scharlau. “Interacting with Fallible AI: Is Distrust Helpful When Receiving AI Misclassifications?” <i>Frontiers in Psychology</i>, vol. 16, doi:<a href=\"https://doi.org/10.3389/fpsyg.2025.1574809\">10.3389/fpsyg.2025.1574809</a>.","chicago":"Peters, Tobias Martin, and Ingrid Scharlau. “Interacting with Fallible AI: Is Distrust Helpful When Receiving AI Misclassifications?” <i>Frontiers in Psychology</i> 16 (n.d.). <a href=\"https://doi.org/10.3389/fpsyg.2025.1574809\">https://doi.org/10.3389/fpsyg.2025.1574809</a>."},"intvolume":"        16","date_created":"2025-05-02T09:22:39Z","title":"Interacting with fallible AI: Is distrust helpful when receiving AI misclassifications?","publication_status":"inpress","author":[{"full_name":"Peters, Tobias Martin","last_name":"Peters","id":"92810","first_name":"Tobias Martin","orcid":"0009-0008-5193-6243"},{"orcid":"0000-0003-2364-9489","first_name":"Ingrid","id":"451","last_name":"Scharlau","full_name":"Scharlau, Ingrid"}],"user_id":"92810","publication":"Frontiers in Psychology"}