A Facial Recognition Program Used By British Police Yielded Thousands Of False Positives

Thanks for sharing. We are looking into this.

Here is the original source from the South Wales Police and their own announcement.

There are 2 competing spins to this, the one you presented (i.e., lots of false positives) and the counter which is that 90%+ of those were from the first event, after which the police say the accuracy improved greatly.

Here is their chart:

We are going to submit the following questions to the police:

• The report explains why the false positive alerts decreased after UCL, but why did the true positive alerts decrease as well? Was the sensitivity of the system adjusted such that fewer true positive alerts were generated as well or?
• "If an incorrect match has been made officers will explain to the individual what has happened and invite them to see the equipment along with providing them with a Fair Processing Notice." Can they share an example Fair Processing Notice? How many total Fair Processing Notices were sent out as part of the facial recognition usage?
• The press release https://www.south-wales.police.uk/en/newsroom/2000-positive-matches-reached-with-our-identify-facial-recognition-technology-in-past-9-months-with-over-450-arrests/says 2,000 positive matches but the event recap https://www.south-wales.police.uk/en/advice/facial-recognition-technology/ shows only 200 - 300 true positive alerts? Where were the other positive matches captured and how many false positive alerts were generated elsewhere?
• "500,000 custody images" What is a 'custody' image and how does one get into this category?
• How much did the program cost? How much was NEC paid?
• "Successful convictions so far include 6 years in prison for robbery and 4.5 years imprisonment for burglary." What other convictions have been made? What specific arrests have been made?

If anyone has any other questions or things for us to check on, let us know.

Btw, funny comment from Wired on this:

Automatic systems that scan people's faces in public and try to make matches are at an early stage in the UK. In China, systems are more advanced with a BBC News reporter being located, during a stunt, within just seven minutes.

So Wired acknowledges the BBC (Dahua) one was a stunt but still concludes it is better than the UK (NEC) one. I doubt that as the NEC is being held to much greater scrutiny and public disclosure to a BBC reporter letting himself get tricked by the PRC.