Hacker News Reader

Comment by akerro | Hacker News Reader

2 years ago facial recognition was used in Cardiff during UEFA final, the result was >2000 false positives

https://www.walesonline.co.uk/news/wales-news/facial-recogni...

This is a classic example of twisting statistics to fit a narrative. That match had ~65k attendance with ~2k false positives for sake of argument. thats < 3%. Not great, but then consider the system is going to scan a vast portion of those in attendance which a human eye alone could not do. Even with the false positive rate, in terms of making human jobs easier, its surely a benefit to scan 2000 false positives than some large portion of 65,000 when checking for unsavoury characters.

From the article: "But of these 92% – a total if 2,297 – were incorrect, with just 173 providing ‘true positive alerts’."

Those are frightening odds when you consider the impact such system could have on those "false positives", eg. possible detainment until your identity is confirmed.

The police may have had the resources to check each case in person during the trial but you can imagine what could happen once there is a push for more wide adoption of these systems as a way to save on police resources.

I would also love to see the proportion of people with dark skin tones who were flagged as false positives.

Then what happens to the pictures of non-confirmed positives and negatives? Are they purged from the system automatically or are they kept around for a while? What recourse have the false-positive to not be harassed in the future?

Let's face it, this genie is not meant to stay in the bottle and it's a policing dream to be able to automatically identify potential undesirables, wherever they are. Why not at the entrance of banks and other sensitive places? Like sports clubs and parks and swimming pools. At every corner of the streets maybe?

The face recognition vans were standing on busy streets and sidewalks, scanning everyone, not just match attendees.

> Even with the false positive rate, in terms of making human jobs easier, its surely a benefit to scan 2000 false positives than some large portion of 65,000 when checking for unsavoury characters.

who says it sensible to check them at all? Basic probability. Let's say the false positive rate is 1%, and there's a criminal among every thousand. That means on chance alone you'll get ten false positives for each actual hit. (the low base rate makes the false positives much worse than they sound, same reason you shouldn't panic if you get a positive test for a rare disease).

That is a ton of work for nothing, a lot of people being stopped without reason, when the saner alternative may have been to not pester anyone at all. Discovering criminality comes at a cost to people who are falsely dragged into it, and creates opportunity cost for the police.

Arresting 10x people to get 1x guilty is terrible. Talking to 10x is great.

Letting crime happen investigated is not a good solution.

Reply on news.ycombinator.com