Discussion
https://www.theguardian.com/technology/2026/mar/19...
Essex Police pausing their AI facial recognition due to racial bias.
Essex Police pausing their AI facial recognition due to racial bias.
Why switch it off, just make it better at identifying other ethnicities and women.
Catching less criminals in order to be 'fair' to male criminals and black criminals seems mental.
I bet they wouldn't turn off speed cameras if they were better at detection certain types of vehicle.
I'm a man, I have zero problem with face recognition being better at detecting men, why would I? If they can improve detection of women, great, if not it's still useful.
Catching less criminals in order to be 'fair' to male criminals and black criminals seems mental.
I bet they wouldn't turn off speed cameras if they were better at detection certain types of vehicle.
I'm a man, I have zero problem with face recognition being better at detecting men, why would I? If they can improve detection of women, great, if not it's still useful.
Edited by BikeBikeBIke on Friday 20th March 10:01
BikeBikeBIke said:
Why switch it off, just make it better at identifying other ethnicities and males.
Catching less criminals in order to be 'fair' to men and black criminals seems mental.
YesCatching less criminals in order to be 'fair' to men and black criminals seems mental.
Before clicking I assumed the problem was false identification of black people, not that it was better at identification. What a bizarre reason to stop the trial.
BikeBikeBIke said:
Why switch it off, just make it better at identifying other ethnicities and women.
Catching less criminals in order to be 'fair' to male criminals and black criminals seems mental.
I bet they wouldn't turn off speed cameras if they were better at detection certain types of vehicle.
I'm a man, I have zero problem with face recognition being better at detecting men, why would I?
If it was easy they would have already done it so the only easy solution here would be to let x% of black criminals found by this to balance it up.Catching less criminals in order to be 'fair' to male criminals and black criminals seems mental.
I bet they wouldn't turn off speed cameras if they were better at detection certain types of vehicle.
I'm a man, I have zero problem with face recognition being better at detecting men, why would I?
Edited by BikeBikeBIke on Friday 20th March 09:46
Edited by BikeBikeBIke on Friday 20th March 09:50
butchstewie said:
I've read that piece a few times and I'm not sure I understand correctly.
Is the issue it is disproportionate in singling out black people for further checks who have done nothing wrong - or it has a list of "known offenders" and it's better at detecting black offenders?
The AI is better at identifying black criminals than other ethnic groups. Is the issue it is disproportionate in singling out black people for further checks who have done nothing wrong - or it has a list of "known offenders" and it's better at detecting black offenders?
Slow.Patrol said:
The AI is better at identifying black criminals than other ethnic groups.
...and Male criminals. Weird that didn't make it into the headline. Because I'm sure men are up in arms about a slight increase in detection rates when identifying male criminals.Oh wait, we're not.
Data is data. AI is fed by data, particularly the models used would have been fed by crime data recorded over decades. Sometimes, the data presented isn't what we want to see. Doesn't mean it's wrong. That being said, I am completely against the use of large scale facial recognition technologies. Big Brother state can do one. What a crap world.
Xenoous said:
Data is data. AI is fed by data, particularly the models used would have been fed by crime data recorded over decades. Sometimes, the data presented isn't what we want to see. Doesn't mean it's wrong. That being said, I am completely against the use of large scale facial recognition technologies. Big Brother state can do one. What a crap world.
Its not that simple and plenty of systems have had biases codes in. Kodak had a feature on digital cameras that warned if people in the picture had their eyes shut, it kept incorrectly warning when taking pictures of Asian people. Automatic hand dryers were only tested on white hands and wouldnt trigger for people with darker skin. Of course its entirely possible that this system correctly identified more male and black criminals because there were just more.
Scabutz said:
Its not that simple and plenty of systems have had biases codes in. Kodak had a feature on digital cameras that warned if people in the picture had their eyes shut, it kept incorrectly warning when taking pictures of Asian people. Automatic hand dryers were only tested on white hands and wouldnt trigger for people with darker skin.
Of course its entirely possible that this system correctly identified more male and black criminals because there were just more.
Of course this is is also true. But it's extremely well documented that those from certain backgrounds commit crime at a higher rates than others. There is of course many reasons as to why this is the case that need addressing as much as anything else. There will be 'bias' inbuilt into this based on widely available statistics that the model will be trained from. Of course its entirely possible that this system correctly identified more male and black criminals because there were just more.
Scabutz said:
Its not that simple and plenty of systems have had biases codes in. Kodak had a feature on digital cameras that warned if people in the picture had their eyes shut, it kept incorrectly warning when taking pictures of Asian people. Automatic hand dryers were only tested on white hands and wouldnt trigger for people with darker skin.
Of course its entirely possible that this system correctly identified more male and black criminals because there were just more.
The system wasn't identifying criminals. The people involved were actors. It was just better at identifying males to its pictures on file and black males within that. Maybe they had more distinguishable features, who knows. Of course its entirely possible that this system correctly identified more male and black criminals because there were just more.
Males definitely have more distinctive facial features than females, and black people even more so. Just based on where people are originally from, facial features are completely different, skull shape even differs in various regions across the globe. Unless there's a genuine software flaw (which of course there could well be) that explicitly picks out males more, it just seems completely logical to me.
butchstewie said:
I've read that piece a few times and I'm not sure I understand correctly.
Is the issue it is disproportionate in singling out black people for further checks who have done nothing wrong - or it has a list of "known offenders" and it's better at detecting black offenders?
It has a low false positive rate across the board so doesn't pick up innocent people but the true positive rate is higher for black people than anyone else.Is the issue it is disproportionate in singling out black people for further checks who have done nothing wrong - or it has a list of "known offenders" and it's better at detecting black offenders?
Reading between the lines, this could be because the police don't have pictures of all criminals so they don't know what the others look like.
Mr Penguin said:
It has a low false positive rate across the board so doesn't pick up innocent people but the true positive rate is higher for black people than anyone else.
Reading between the lines, this could be because the police don't have pictures of all criminals so they don't know what the others look like.
I can see why people would want to know why it was doing that.Reading between the lines, this could be because the police don't have pictures of all criminals so they don't know what the others look like.
But I'm also not sure the fact it's doing it is reason to stop using it so long as it's correctly identifying people who it should identify rather than lots of false positives (of anyone).
JagLover said:
Scabutz said:
Its not that simple and plenty of systems have had biases codes in. Kodak had a feature on digital cameras that warned if people in the picture had their eyes shut, it kept incorrectly warning when taking pictures of Asian people. Automatic hand dryers were only tested on white hands and wouldnt trigger for people with darker skin.
Of course its entirely possible that this system correctly identified more male and black criminals because there were just more.
The system wasn't identifying criminals. The people involved were actors. It was just better at identifying males to its pictures on file and black males within that. Maybe they had more distinguishable features, who knows. Of course its entirely possible that this system correctly identified more male and black criminals because there were just more.
Essex police can't "fix" that, they just bought the system in. It's quite likely that the people they bought it from can't fix it either, because they just bought in the software, or at least the underlying face detection system.
butchstewie said:
I can see why people would want to know why it was doing that.
But I'm also not sure the fact it's doing it is reason to stop using it so long as it's correctly identifying people who it should identify rather than lots of false positives (of anyone).
It's definitely something to look at to improve (in terms of making it better at identifying the other criminals) but shouldn't be a reason to not use it IMO.But I'm also not sure the fact it's doing it is reason to stop using it so long as it's correctly identifying people who it should identify rather than lots of false positives (of anyone).
Gassing Station | News, Politics & Economics | Top of Page | What's New | My Stuff