Real-world stress testing has become critical before deploying AI systems in law enforcement to carry out forensic analysis. Stress testing of AI biometrics, which the DFS conducts against actual casework scenarios, helps determine system limits and emulate real-world error rates with the technology such as degraded prints, mixed substrates, distorted images, and curved surfaces.
AI has integrated into investigatory work but Angelic Browntree, who is Unit Manager at the D.C. Department of Forensic Sciences explains the difficult problems that persist working with biometrics. For example, it becomes difficult to differentiate close non-matches to mitigate false-identification errors. Matching a print against older database enrollments is also tricky given how skin naturally degrades over time. Deep learning tools in Automated Fingerprint Identification Systems have significantly improved match probability by stabilising ridge flows and sifting through background noise.
According to Browntree, AI and biometric technology will improve efficiency, act as a triage layer, and speed up casework processing, with increased integration of statistical models. However, human examiners will remain essential for the final call due to the complex environmental variables involved in how latent prints are deposited.
