It doesn't say one group commits more crime. It says the system is more correct on identifying black people on a list. So black suspects were being identified better than other groups.
but the system was more likely to correctly identify men than women and it was “statistically significantly more likely to correctly identify black participants than participants from other ethnic groups"
With it being better at IDing men than women too there's an implication that it's just better at IDing demographics it has the most experience searching for. Most of these AI systems just keep improving with more data and feedback, so if it's constantly finding black criminals and then getting feedback on if it was right, it's going to improve at finding them faster.
There's also just an argument that the black criminal demographic in britain is just the most visually distinctive. You can often tell a Jamaican from a Somalian from a Sudanese just at first glance, and all of those are pretty likely options. Whereas most the white criminals are going to be white British and therefore most of the time it's working with one big homogeneous group where look-a-likes are far more likely to occur.
Possible reasons for the latest issue with LFR include overtraining of the algorithm on the faces of black people. Experts believe it could be rectified by adjusting system settings.
What's really crazy here is the implication (issue, overtraining) that they should "adjust" settings to, what, not correctly identify Black people as much?
Black people aren't the victim here; Black criminals are being correctly identified. It's Whites that are being accused of crimes they didn't commit. But, of course, the Guardian frames it as though Blacks are being persecuted, when they're just being prosecuted.
Which doesn't make any sense because all the facial recognition literature up to this point has shown the opposite outcome; Whites are more easily identified than Blacks.
It doesn't say one group commits more crime. It says the system is more correct on identifying black people on a list. So black suspects were being identified better than other groups.
With it being better at IDing men than women too there's an implication that it's just better at IDing demographics it has the most experience searching for. Most of these AI systems just keep improving with more data and feedback, so if it's constantly finding black criminals and then getting feedback on if it was right, it's going to improve at finding them faster.
There's also just an argument that the black criminal demographic in britain is just the most visually distinctive. You can often tell a Jamaican from a Somalian from a Sudanese just at first glance, and all of those are pretty likely options. Whereas most the white criminals are going to be white British and therefore most of the time it's working with one big homogeneous group where look-a-likes are far more likely to occur.
What's really crazy here is the implication (issue, overtraining) that they should "adjust" settings to, what, not correctly identify Black people as much?
Black people aren't the victim here; Black criminals are being correctly identified. It's Whites that are being accused of crimes they didn't commit. But, of course, the Guardian frames it as though Blacks are being persecuted, when they're just being prosecuted.
Which doesn't make any sense because all the facial recognition literature up to this point has shown the opposite outcome; Whites are more easily identified than Blacks.