AI assistance improves people's ability to distinguish correct from incorrect eyewitness lineup identifications

• 成果类型: Article
• 署名作者: Kelso, Lauren E.; Dobolyi, David G.; Grabman, Jesse H.; Dodson, Chad S.
• 署名单位: University of Virginia; University of Colorado System; University of Colorado Boulder; New Mexico State University
• 刊物名称: PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA
• ISSN/ISSBN: 0027-11963
• DOI: 10.1073/pnas.2503971122
• 发表日期: 2025-05-19
• 关键词: confidence; accuracy; reality; models; memory

摘要

Mistaken eyewitness identification is one of the leading causes of false convictions. Improving law enforcement's ability to identify correct identifications could have profound implications for criminal justice. Across two experiments, we show that AI-assistance can improve people's ability to distinguish between accurate and inaccurate eyewitness lineup identifications. Participants (Experiment 1: N = 1,092, Experiment 2: N = 1,809) saw an eyewitness's lineup identification, accompanied by the eyewitness's verbal confidence statement (e.g., I'm pretty sure) and either a featural (I remember his eyes), recognition (I remember him), or familiarity (He looks familiar) justification. They then judged the accuracy of the eyewitness's identification. AI-assistance (vs. no assistance) improved people's ability to distinguish between correct identifications and misidentifications, but only when they evaluated lineup identifications based on recognition or featural justifications. Discrimination of identifications based on familiarity justifications showed little improvement with AI-assistance. This project is a critical step in evaluating human-algorithm interactions before widespread use of AI-assistance by law enforcement.