The Loss Prevention Research Council (LPRC) provides evidence-based solutions addressing retail loss, safety, and fraud prevention. Dr. Cory Lowe, LPRC senior research scientist, presented his research results October 4, 2022, during the annual IMPACT conference hosted at the University of Florida.
The Role of AI in Reducing Bias
In a presentation titled “Face Off: Examining the Role of Al in Reducing Bias and Improving Decision Making,” Lowe explained how he pitted unaided research participants against those using AI face matching technology. LPRC selected FaceFirst software for the tests. Researchers installed the software in the LPRC lab and conducted the tests independently.
Humans Alone: Correct Just 23% of the Time
Lowe showed a diverse array of fictional offender faces to 155 research participants; 78 unassisted, 77 assisted. Among the unassisted group, 77 percent misidentified the fictional subjects in a photo lineup just minutes after seeing the fictional subject image. The assisted group got it right 63 percent of the time. “The assisted group did nearly three times better,” Lowe said. “There was a 2.7 times improvement in accuracy when assisted by facial recognition.”
For context: Humans were only correct on their own 23 percent of the time, even with a small sample of faces they had been shown only minutes earlier. No technology is 100 percent accurate in the wild, but humans alone are demonstrably prone to error. Lowe noted the participants were not told of the Al’s accuracy alone (100 percent accurate in this study), so individuals may have discounted the solution’s accuracy.
FaceFirst Helps Retailers Prevent Violence
Facial recognition can be used to reduce error and bias, and it can narrow the LP focus to those individuals who are most likely to offend in retail locations.
FaceFirst considers use of AI with human oversight vital for retailers. Calculate the risks of being caught unaware when a known offender enters your store. If you knew there was a proven solution to keep your valued customers and associates safer from violent offenders, would you implement it? The real risk is answering no. FaceFirst’s solution is fast, accurate, and ethical—learn more today at facefirst.com.