Percepta prides itself on using Ethical AI technology to accurately detect shoplifting in retail stores. But, what is “Ethical AI?”
At Percepta, we developed Ethical AI to minimize the potential for discrimination and to protect customers’ privacy, all while performing complex behavioral analysis. In other words, our Ethical AI eliminates the possible role of demographic features, or characteristics, in shoplifting surveillance.
We accomplish this by anonymizing shoppers in camera feeds and removing the focus on demographic features, from the outset. Combining advanced action recognition, pose estimation, and gaze estimation techniques, our Ethical AI relies solely on suspicious movements and actions for detection. Analyzing these movements, which strongly correlate with shoplifting behavior, rather than irrelevant factors such as race, has proven to be more accurate in detecting shoplifting incidents.
In a study conducted by Dean A. Dabney, Richard C. Hollinger, and Laura Dugan, they found that “while demographic characteristics of the shopper were important for predicting who shoplifts, our data clearly indicate that behavioral variables are much more significant predictors.” Unlike other solutions, Percepta accurately detects shoplifting without analyzing customers’ faces, or invading their privacy.
In a world where AI can be weaponized to impede on privacy, security, and the quality of life, Percepta finds it essential that our AI be ethical. A common issue found in AI technology is bias. For example, in 2018 Amazon discovered that its AI recruiting technology was biased against female applicants. Similarly, Facebook was recently sued over their AI advertising technology that could discriminate based on gender and race.
The retail industry faces similar issues, with the growing number of racial-profiling claims, based on facial recognition AI errors. For example, many African Americans have been incorrectly arrested for shoplifting on the basis of this AI. This technology possesses bias as it “[hasn’t] learned how to recognize the differences between black faces because the images used to train it had mostly been of white faces.”
Retailers have faced public relations nightmares, with the growing number of claims. In addition to bias in facial recognition AI, Professor Pittman at Case Western Reserve University conducted research, a few years back, interviewing 55 middle-class African American shoppers in New York City, and found that “80 percent reported experiencing racial stigma and stereotypes when shopping.” Therefore, it is of utmost importance to Percepta to produce Ethical AI, that protects privacy, mitigates bias, and reduces loss.