Study Reveals Benefits of Human-Machine Collaboration in Face Identification
Face identification is most accurate when human analysts and machines work together, a recent multidisciplinary study uncovered.
According to a report published in the Proceedings of the National Academy of Sciences, three scientists at the National Institute of Standards and Technology (NIST) found that facial identification performance is optimal when humans and machines collaborate.
Specifically, the researchers observed the best results after pairing a single professional facial examiner with a top-performing algorithm. This analytical setup was more accurate than arrangements involving multiple examiners or algorithms.
Face Identification Most Effective When Humans and Machines Collaborate
For the study, NIST analyzed the face identification performance of 184 participants, 87 of which were professional face examiners who analyze faces for legal purposes. There were also 13 “super recognizers” who are naturally gifted at comparing faces.
The control group of 84 participants involved 53 fingerprint examiners and 31 undergraduate students, none of whom had received training in facial examinations.
NIST asked each participant to compare 20 face image pairs that were deemed extremely challenging based on their characteristics. The scientists then conducted the same analyses using four machine learning facial recognition technology algorithms developed between 2015 and 2017.
Professional face examiners were more accurate than individuals in the untrained group. The NIST researchers asserted that these results should help scientifically substantiate experts’ testimony in court.
The algorithms also performed well, but their accuracy was even better when combined with the insights of an expert examiner. This human-machine collaboration even beat out groups of professional face examiners.
Results Reveal a Need for Further Research
“This is the first study to measure face identification accuracy for professional forensic facial examiners, working under circumstances that apply in real-world casework,” said NIST electronic engineer P. Jonathon Phillips. “Our deeper goal was to find better ways to increase the accuracy of forensic facial comparisons.”
The NIST researchers noted that current real-world forensic frameworks don’t encourage the combination of human examiners and artificial intelligence (AI). They asserted that the study’s findings can serve as a road map for improving face identification in the future and raised questions about what distinguishes human approaches from AI-based methods.