A big biometric security company in the UK, Facewatch, is in hot water after their facial recognition system caused a major snafu - the system wrongly identified a 19-year-old girl as a shoplifter.
A big biometric security company in the UK, Facewatch, is in hot water after their facial recognition system caused a major snafu - the system wrongly identified a 19-year-old girl as a shoplifter.
Developers were probably the first people to say that it isn’t ready. Blame the sales people that will say anything for money.
It’s impossible to have a 0% false positive rate, it will never be ready and innocent people will always be affected. The only way to have a 0% false positive rate is with the following algorithm:
def is_shoplifter(face_scan):
return False
line 2 return False ^^^^^^ IndentationError: expected an indented block after function definition on line 1
In their defense, it didn’t return a false positive
Weird, for me the indentation renders correctly. Maybe because I used Jerboa and single ticks instead of triple ticks?
Interesting. This is certainly not the first time there have been markdown parsing inconsistencies between clients on Lemmy, the most obvious example being subscript and superscript, especially when ~multiple words~ ^get used^ or you use ^reddit ^style ^(superscript text).
But yeah, checking just now on Jerboa you’re right, it does display correctly the way you did it. I first saw it on the web in lemmy-ui, which doesn’t display it properly, unless you use the triple backticks.
They worked on it, they knew what could happen. I could face criminal charges if I do certain things at work that harm the public.
I have no idea where Facewatch got their software from. The developers of this software could’ve been told their software will be used to find missing kids. Not really fair to blame developers. Blame the people on top.
It says right on their webpage what they are about.
Developers don’t always work directly for companies. Companies pivot.