Metrics Not Thresholds: How to Regulate AI and Other Expert Methods for Use in Criminal Investigations and Proceedings

Key Points

• The central aim is to evaluate whether performance thresholds should govern AI tools in criminal law.
• Analyzed U.S. evidence law to provide guidance on regulatory practices for AI tools.
• Examined the implications of performance thresholds versus metrics for reliability.
• Proposed alternatives for regulating AI based on the communication of reliability metrics.
• Found that performance thresholds may hinder the effective use of AI in criminal proceedings.
• Recommended that regulators focus on clear metrics of reliability instead of set thresholds.
• Indicated that improving communication of these metrics could empower decision-makers.