KMWorld 2024 Is Nov. 18-21 in Washington, DC. Register now for Super Early Bird Savings!

The twisted case of facial recognition

Article Featured Image

Two areas strike me as emblematic of the value being generated by our current wave of AI innovations. These would be the belated arrival of automatic machine translation and the perhaps equally belated arrival of widespread image recognition, particularly facial recognition. Both have been subjects of research and development since the earliest days of AI. And both problems proved surprisingly to be much more approachable recently due to the availability of commodity compute power and the reinvention of practical analytics based on deep neural networks.

The risks of facial recognition

Machine translation continues to make strides forward due primarily to Google’s continuing interest in this research and the company’s unmatched access to multi-language text data from around the world. Most users I believe would have been shocked as recently as 5 years ago to be told that they could click a button and see the text of a webpage rendered in their native language (or the reverse). Our world’s lingua franca has become the lingua Google, and the reviews are virtually all positive.

Facial recognition has most definitely entered the twilight zone. Consider the case of Robert Julian-Borchak Williams, described as a large African-American male whose 2019 arrest for shoplifting in a Shinola boutique store in Detroit’s gentrifying midtown was documented extensively in a recent New York Times report. The subtitle of this report sums up the problematic fringe of the application of facial recognition: “In what may be the first known case of its kind, a faulty facial recognition match led to a Michigan man’s arrest for a crime he did not commit.”

Surveillance camera images from the 2018 Shinola shoplifting incident showed a large black male removing watches from a display. (Full disclosure: I saw those watches in that store and nearly bought one.) Eventually, these images were posted in a law enforcement database. From there, the Michigan State Police scanned the database aided by facial recognition software from DataWorks Plus, a South Carolina company founded to perform mug-shot analysis in 2000, which now integrates facial recognition analytics from Japanese tech giant NEC and Colorado-based Rank One Computing.

The state police put out a report identifying Williams’ driver’s license photo as being a potential match for the suspect. So roughly a year after the incident, without further substantiation, the Detroit police arrested Williams and held him in jail for more than 30 hours. It was soon obvious to the officers and eventually to the courts that there was no resemblance between Williams and the shoplifter in the surveillance video. In the end, exonerated from any wrongdoing, Williams received an apology from the Wayne County prosecutor’s office and an assurance that all records of his case and his data would be expunged.

KMWorld Covers
for qualified subscribers
Subscribe Now Current Issue Past Issues