- Police in South Wales have been relying on facial recognition technology for 12 months.
- An FOI request has revealed that the technology provides a "false positive" ID in more than 90% of cases.
- The police have admitted that "of course no facial recognition system is 100% accurate".
British police have been forced to defend facial recognition technology which falsely identified 2297 out of 2470 football fans as "persons of interest."
South Wales police rolled out its "Automated Facial Recognition" (AFR) pilot program just before the Champion's League Final between Juventus and Real Madrid almost a year ago in Cardiff, Wales.
Late last week, Wired's Matt Burgess got an FOI request back in regard to how effective the AFR system has been for the South Wales Police since then.
It wasn't great.
The "false positive" rate – 92% unsuccessful – from AFR's work at the Champions League final was the most damning.
In other events, only six positive matches were confirmed out of 48 alerts at a Wales versus Australia rugby match, and no matches were made at all during a royal visit from Prince Harry and Meghan Markle to Cardiff in January 2018.
Almost immediately following the report, South Wales police put out an explainer on how its facial recognition system worked in two stages, AFR "Identity" and AFR "Locate".
AFR "Identity" allows officers to load images of persons of interest and compare them to 500,000 images they already have on file and see if there's a match.
AFR "Locate" uses CCTV and roving police vehicle cameras to locate that person of interest.
The force made it clear no one had ever been arrested on the basis of a false positive result and claimed the "overall effectiveness of facial recognition has been high."
It also claimed 2000 positive matches have led to 450 arrests in the past nine months.
But Wired's FOI request has returned a total result of just 234 "true positives" from 2685 alerts across 15 events.
South Wales Police opened its release with:
"Of course no facial recognition system is 100% accurate under all conditions."
Its system for dealing with potential false positives is a) disregard the alert; or b) dispatch a team to "have an interaction with the potentially matched individual."
Officers would then use "traditional policing methods" (mostly, a chat) to decide whether more action was required.
That a new technology isn't exactly performing at 100% should come as no surprise to anyone. But for the general public, there's no doubt facial recognition and criminal profiling definitely fall in a preferred category of "get it right first."
Immediately, civil liberties groups have begun preparing campaigns for British parliament.
You can read the full story at Wired here.