lilysea: Serious (Default)

[personal profile] lilysea 2024-05-26 11:35 am (UTC)(link)
There was a case in New Zealand about a woman being incorrectly flagged as a shoplifter at a supermarket by facial recognition

and the issue was that she was Maori, and the facial recognition is ***much less accurate*** for people who aren't white.
lilysea: Serious (Default)

[personal profile] lilysea 2024-05-28 09:56 am (UTC)(link)
I bet that if the facial recognition was ***much less accurate*** for white people it would have been fixed by now!
bens_dad: (Default)

[personal profile] bens_dad 2024-05-28 06:00 pm (UTC)(link)
How accurate is facial recognition of white faces in non white-majority parts of the world ?

Do whites in Beijing or Lagos have the problems that African Americans have in the US ?
melchar: medieval raccoon girl (Default)

[personal profile] melchar 2024-05-26 04:19 pm (UTC)(link)
I loved the article about 'Princess Mononoke', which is - depending on my mood - either pretty much my favorite anime movie, or at the very least in my favorite top 5.
movingfinger: (Default)

[personal profile] movingfinger 2024-05-26 05:08 pm (UTC)(link)
1. I do wonder when someone in the UK is going to get fed up with this crap and start implementing the Zodiac solution of sealing the waste pipes.
calimac: (Default)

[personal profile] calimac 2024-05-26 05:11 pm (UTC)(link)
5) I once had a case of that without technology, when I was accosted at a bus stop by a shopkeeper who was sure I was the person who'd come into his shop and made racist remarks. He was hopping mad and apparently had been scouring the neighborhood looking for the culprit. Meanwhile his fellow shopkeeper was tugging at his arm and saying "No, that's not the guy." I certainly wasn't. I had just gotten off the Tube from central London and had been nowhere near any local shops.
armiphlage: Ukraine (Default)

[personal profile] armiphlage 2024-05-26 05:54 pm (UTC)(link)
I work in a factory that makes electronic circuit assemblies. Green fiberglass rectangles with electronic parts soldered to them.

We have automated camera systems that look at each assembly, and identifies whether or not everything is perfect, or if there is a potential issue that a human has to look at.

Despite being nearly 2D, and despite everything being the same distance from the camera, moving at a known speed, with well-controlled lighting, our false call rate is about 10,000 PPM. That's 1%. And our electronic circuit assemblies don't wear glasses or hats, or makeup.

The police claiming that they have a false call rate of 1 in 33,000 (30 PPM) doesn't make sense to me.
bens_dad: (Default)

[personal profile] bens_dad 2024-05-27 12:17 pm (UTC)(link)
Is your false call rate 1 false positive per 100 parts or 1 false positive per 100 positive claims ?

If you have 1 false positive per 100 positive claims and the police have 1 false positive per 33000 IDs the numbers would not be incompatible.
armiphlage: Ukraine (Default)

[personal profile] armiphlage 2024-05-27 10:15 pm (UTC)(link)
1 false positive per 100 parts.
bens_dad: (Default)

[personal profile] bens_dad 2024-05-28 04:02 am (UTC)(link)
How many true positives ?

I guess if the police system doesn't claim many hits, it cannot generate many false positives.
armiphlage: Ukraine (Default)

[personal profile] armiphlage 2024-05-28 10:40 am (UTC)(link)
About 100 PPM true positives, or 0.01%.
mellowtigger: (clock spiral)

[personal profile] mellowtigger 2024-05-26 09:30 pm (UTC)(link)
#2 I wonder if that's an historical effect at other U.S. military ports besides SF? It reminds me of a story I mentioned about docks back in 2009, heard from a family member in the Navy in the 1960s.