Evaluating Face Identification Expertise: Turning Theory into Practice

January 2020

Convenors: Dr David White, A/Professor Romina Palermo, Professor Gillian Rhodes

Critical face identification decisions that underpin security, forensic and legal processes are often made by people. Border control officers compare a passport photo to a traveler, surveillance officers see a person of interest in a crowd, police officers compare mugshots to CCTV images. This need for human oversight can be problematic because many decades of research has shown that the average person is surprisingly poor at matching the identity of unfamiliar faces.

The workshop provided the unique opportunity to address this challenging and complex problem by bringing together over 20 world-leading psychologists, computer scientists and legal scholars in face identification research, with practitioners and policymakers to reach a consensus on how expertise in face identification should be defined and evaluated.

The main conclusions and recommendations of the workshop were:

  • face identification is now a mature multi-disciplinary field incorporating forensic science, cognitive psychology and artificial intelligence research
  • recent research shows that accuracy of the best algorithms and the best humans are comparable, but performance is optimized by combining decisions made by the best algorithms and humans
  • face identification ‘experts’ must consistently demonstrate superior performance on tasks representative of the claimed expertise
  • there is substantial variation in accuracy and performance between individual experts, and between individual algorithms
  • new types of expert practitioners and researchers are required to design, evaluate and oversee modern face identification systems.

Critically, the workshop participants concluded further interdisciplinary research collaboration between scientists, practitioners and policy makers should be continued to address the mounting challenges ahead, as face identification becomes increasingly prevalent in forensic, legal and identity management processes.