Facial recognition tests at BCLC 'disappointing': documents

Technical issues plague BCLC's facial recognition plan Documents obtained by CTV News show that the system has been hampered by problems for more than a decade.

Facial recognition systems touted by the B.C. Lottery Corporation as part of a solution that could keep criminals and problem gamblers out of casinos aren't yet ready for operation, according to documents obtained by CTV News.

Though the technology has improved dramatically since testing began a decade ago, internal reviews described the results as "disappointing" with a rate of false positives "too high for the required buy-in from BCLC and service provider staff."

"These systems are complicated. It was overpromised in the first place. Surveillance is something that is selling a sense of security and a sense of control. And that's always partial. It's always incomplete," said Mike Larsen of the Freedom of Information and Privacy Association.

Privacy advocates have long expressed concerns about facial recognition, which has reached the point that tech products like Amazon's Rekognition and Facebook's DeepFace can match a name to a face using databases of millions of uploaded photos with higher accuracy.

One former gambler, Nik Galego, said he signed up for the self-exclusion program, believing that it had facial recognition which could help detect him. He said he was very disappointed that he was allowed into a casino without being detected.

"It's part of basic security, right?" he told CTV News on Thursday, adding that a simple solution would be to check for ID at the door. "You can't have a facility like that and have open access for everyone."

About a decade ago, facial recognition technology was pitched as a way to weed out undesirable patrons from casinos without deterring customers from bringing their money into casinos and interfering with their revenue.

The CEO of BCLC at the time told reporters it was part of the company's solution to a series of lawsuits alleging the company hadn't done enough to keep out gambling addicts, as well as other concerns about security.

"There's not one thing that's going to be effective. It's going to be a multitude of things. And I think facial recognition is going to help with that," he said.

Internally, the agency was working on a solution: "There is an urgency to do a better job at identifying self-excluded, banned and prohibited subjects attempting to gain access to gaming establishments," one document, from 2011, says.

But reviews of the technology over the years show that after a $400,000 pilot, "there were many false positives generated and included a number of matches being made." Of 3647 alerts, 3255 were rejected, and 387 were accepted by the software. But of those, only 26 were confirmed by staff, the review says.

The company next tried using employees, motivating them to participate with an iPad and a TV as prizes. In that trial, there was a 62 per cent hit rate, but one out of ten identified photos were mismatched.

Another business plan in 2015 declared a set-up cost of $1.8m. The following pilot in 2017 provided a hit rate of 91.5 per cent – but false positives were still high at 5 per cent, or one of every twenty photos identified were mismatched.

"System can provide strong results with use of controlled VSE databases with clear standards for enrollment process and governance, but will otherwise provide too high a false positive rate for the required buy-in from BCLC and service provider staff…the resulting costs will likely be too high to justify the return," the pilot study concludes.

BCLC says it's not giving up. Tests that began at one casino in January are set to continue through to 2020, the company says.

"Technology continues to advance in this field, hence our ongoing efforts," a company spokesperson wrote in a statement.