Using Artificial Intelligence to Assess Eyewitness Identification Accuracy

Drawing on recent experimental research, we propose three ways in which artificial intelligence (AI) can be used to assess eyewitness identification accuracy. First, we suggest that AI could be used as an outside evaluator of eyewitness accuracy. AI could be exposed to identification evidence and left to make its own classification decisions. Second, AI could be used as a tool to inform human evaluators on the likely accuracy of an identification decision. This could range from clarifying witness information to providing evaluators with numeric estimates of accuracy. Third, AI could be used as a model of how human evaluators should go about assessing identification accuracy. Researchers could examine how successful AI models discriminate between accurate and inaccurate identification decisions and train human evaluators to use a similar algorithm. We contend that transparency and interpretability will be key ingredients in determining how AI impacts the legal system.