This post is also available in:
עברית (Hebrew)
Eyewitness testimony is a cornerstone of criminal investigations, but research shows that human biases often affect the accuracy of statements. A key issue is the featural justification effect, where witnesses who provide highly detailed descriptions—such as clothing or physical traits—are often viewed as more credible, even if their recollection is less accurate. This phenomenon can lead to wrongful conclusions and mistakes. However, a recent study suggests that artificial intelligence (AI) could be the solution to reducing these biases and improving decision-making in law enforcement.
The study, co-authored by David Dobolyi, assistant professor at the Leeds School of Business, explores how AI, particularly natural language processing (NLP), can be used to analyze eyewitness statements objectively. Unlike humans, AI can assess language from a neutral perspective, free from the biases that often cloud human judgment. The system can even assign numeric scores to statements based on the likelihood of accuracy, helping investigators prioritize the most reliable accounts, according to TechXplore.
Dobolyi explains that while traditional methods of eyewitness analysis have focused on basic word counts, advanced AI can provide deeper insights. By examining the language and confidence behind eyewitness statements, AI can help assess the true reliability of a statement. “Just because someone says they’re confident doesn’t mean they’re right,” Dobolyi said, pointing out that some of the most significant mistakes in law enforcement come from witnesses who are overly confident but inaccurate.
The study tested the effectiveness of AI assistance with 1,010 participants, who were asked to evaluate eyewitness identifications. Some participants received AI support, which included predictions on the accuracy of statements and visual explanations. The results were promising: participants who found the AI helpful were less influenced by the featural justification bias and were able to evaluate both detailed and general descriptions more accurately.
Dobolyi cautioned against blindly trusting AI, but emphasized its potential to improve decision-making in high-stakes situations like criminal investigations. “We want tools that can help people make better, less biased decisions,” he said. As AI continues to evolve, Dobolyi stressed the importance of transparency in AI recommendations, especially in legal contexts where the stakes are high.
Incorporating AI into the analysis of eyewitness testimony could lead to more reliable conclusions, reducing human error and bias in law enforcement.
