375
In every reported case where police mistakenly arrested someone using facial recognition, that person has been Black
(www.businessinsider.com)
this post was submitted on 07 Aug 2023
375 points (97.7% liked)
Technology
59323 readers
4666 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
This isn’t new. It’s been a known problem for a long time, because facial recognition software is trained using white people. So it gets really really good at differentiating between white people. But with black people as a tiny fraction of the sample data, it basically just learns to differentiate them with broad strokes. It’s good at telling them apart from white people, but not much else.
It's not just a training issue. Lighter (color) tones reflect. Dark tones absorb. There have been lots of issues with cameras or even just sensors having issues with people having dark skin tones because the lower reflectivity/contrast of dark tones. 3D scanners - even current models - have similar issues with objects having black parts for similar reasons. Training on more models can help, but there's still an overall technical difficulty to overcome as well (which is also a good reason that using facial recognition in this manner is just bullshit, period).
As a technological problem it could have a technological partial solution: the darker the skin, the higher the threshold to declare a match. This would also mean more false negatives (real matches not caught by the software) but not much to do about that.
I'm interested what dataset they're using because simply adding more black people to the training set seems like a pretty straightforward fix.
It seems like past mugshots would be an ideal part of the training set. Are they not using those?
Personally, I'm leaning towards "It's not the image recognition program that's the problem here."