Should AI Help Police with Investigative Interviews? Benefits, Pitfalls and the Need for Scientific Standards

Summary

Investigative interviews are central to solving crimes, but how interviews are conducted shapes reliability and later legal outcomes. This article examines whether AI should assist police in investigative interviewing, highlighting memory distortion, misinformation, and automation bias as key concerns, while noting potential benefits in information gathering and interviewer training. It also situates AI's role within the UK policy landscape, including the launch of a national AI centre for policing in 2026 and calls for rigorous scientific standards in AI-assisted interviewing.

• AI could help in the initial information gathering stage with chatbots and automated transcription at scale.
• Real-time AI prompts and new lines of inquiry could guide interviewers toward open questions and broader investigations.
• There are significant risks, including post-event misinformation, memory contamination, and automation bias in decision making.
• Policy context in the UK emphasizes responsible AI adoption and the need for evidence-based practice in investigative interviewing.

Introduction: Investigative interviewing and AI in policing

Investigative interviewing—the process of obtaining accurate accounts from victims, witnesses and suspects—lies at the heart of the criminal justice system. How police interview can shape reliability, completeness, and credibility of evidence is well established. Historically, accusatory, non-evidence-based interrogation methods increased the risk of false confessions and distorted witness accounts. Classic studies, including a 1992 evaluation of video-taped interviews in England and Wales, showed that officers often relied on assumptions and confirmation-seeking rather than open-ended information gathering. More recent research highlights how questions and post-event information, including media coverage, can distort memory, leading to the misinformation effect. The authors emphasize that memory is reconstructive, not a perfect recording, and that memory can be contaminated by external input from other witnesses and subsequent discussions.

AI in investigative interviews: potential roles

The article outlines three practical domains where AI could theoretically assist policing tasks around investigative interviewing: information gathering, interview honing, and officer training. In the early stages after a suspected crime, conversational AI agents could conduct initial interviews at scale, complementing human investigators and enabling automated transcription tools to accelerate processing. As an investigation narrows, AI could provide real-time prompts to help interviewers pose open, non-leading questions and to generate fresh lines of inquiry. AI-enabled avatars, such as EchoMind and Innsikt, are discussed as tools for training officers in best-practice interviewing techniques. The UK government’s interest in AI for policing is underscored by the June 2026 launch of a national AI centre for policing, aimed at accelerating responsible AI use across 43 forces in England and Wales.

Key challenges and the evidence gap

While AI promises efficiency, the authors stress significant uncertainties. AI systems can introduce post-event misinformation into witness memories, and exposure to AI systems that validate and elaborate accounts can decrease recall accuracy. The concerns extend to vulnerable witnesses, including children and individuals with cognitive disabilities, and to how AI contamination might propagate across multiple stages of an investigative workflow. A documented risk of automation bias—where decision-makers rely too heavily on algorithmic outputs—exists when AI outputs appear confident and well-presented. The authors note that live AI interactions in interviews may display behaviours that are difficult to observe in lab settings, with courtroom outcomes only revealing effects long after interviews occur.

To address these risks, the article previews a systematic taxonomy (playbook) of potential AI large language model errors in law enforcement. This taxonomy would categorize errors as factual (untrue statements), faithfulness (inaccurate reflection of information given), and task-specific errors (failure at a task despite correct facts). It also discusses how such errors could propagate through decision systems, reinforcing the need for rigorous evaluation and safeguards before widespread deployment.

Why science-based interviewing matters

The piece reminds readers that extensive, science-based investigative interviewing has developed over decades, with England and Wales’ Achieving Best Evidence guidance serving as a cornerstone. It argues that AI adoption is occurring rapidly, often without the same level of empirical validation that traditional interviewing practices have required. The authors frame their concerns as a call for AI use to be aligned with the long-established scientific standards of investigative interviewing, emphasizing robust evidence and careful assessment of AI-assisted tools before they influence the investigation or courtroom outcomes.

Policy context and ethical considerations

Beyond technical capabilities, the article emphasizes governance, training, and policy alignment. It acknowledges the public interest in innovation while pushing for precautionary use of AI that adheres to established practices and ethical norms. The authors do not oppose innovation; they advocate for AI applications that are evidence-based, transparent, and subject to ongoing evaluation. The piece also notes that AI’s rapid integration into investigative work should be matched by rigorous standards and ongoing research to understand how AI interacts with memory, cognition, and juror reasoning in legal contexts.

Conclusion

AI could offer meaningful gains in information gathering, interview preparation, and training, but the potential pitfalls are substantial. The authors call for AI adoption that respects scientific standards, with a clear emphasis on safeguarding memory accuracy, avoiding misinformation, and mitigating automation bias. The message is one of cautious, evidence-driven innovation rather than outright fear or rejection of technology, seeking to ensure AI complements rather than compromises the integrity of investigative interviewing and the justice process.