The DMZ

« first day (5149 days earlier) ← previous day next day → last day (36 days later) »

Rory Alsop

16:16

@ThoriumBR The European AI Act has a good list of impact types, and pretty much says that it should not be used where its impact on people could be significant. (I'm paraphrasing a lot)

So for us, it could not be used in decision making on products for vulnerable customers, unless there was a very high level of governance over that...which might make it pointless to use AI at all for that role

ThoriumBR

16:52

the problem is when AI is used somewhere down the pipeline, and it's difficult to prove AI was used at all

insurance companies using AI to flag "risky customers" can say they manually reviewed their driving habits even when AI did the work, and it's difficult to prove it's a lie

they can see "AI is just advising but a human makes the decision" but we all know the human will just forward whatever decision AI took

2 hours later…

Fire Quacker

18:26

I think there are some issues specific to US police where AI is used irresponsibly, even though officers are instructed to verify output

ThoriumBR

it's not US-specific... in Brazil it happens all the time

Fire Quacker

Perhaps it's just AI companies trying to protect themselves from lawsuits, but when I hear about officers making false arrests based on AI, I usually also hear that the AI company's official guidance was not followed

That said, I wouldn't be extremely surprised if AI sales teams glossed over the shortcomings

ThoriumBR

It's the "trust the computer" mindset... because computers don't lie, so it must be the truth.

Fire Quacker

Hopefully, if enough police departments get held responsible, they'll change their practices

But that will require significant help from justice charities, since a lot of the victims aren't people with a lot of money

ThoriumBR

police departments down here aren't held responsible for anything, so I doubt something will change for better

Fire Quacker

18:37

Depending on an individual officer's goal, they may not see AI as a bad thing... catching a suspect beats catching nobody

Keep up your statistics, make people think you're doing your job

ThoriumBR

a few years ago police shoved a man with mental issues on the back of a police car, stuffed it full of tear gas grenades and held him locked in until he died... those responsible got prosecuted but are still working, free as a cop...

Fire Quacker

That's a problem

ThoriumBR

it was filmed, circulated on social networks, got broadcast on national TV, there was a national outrage... and it was that.

the police told everyone they had to rethink their approach, and it's just the same

I went after the latest news... it only took 18 months for them to be fired, but they are still free waiting for the jury.

11 minutes locked in the back of the car, filled with tear gas and pepper spray. and they kept working as normal.

a few months later, another cop got caught making fun of the incident... he was an instructor for new cops and was teaching how to use tear gas and the back of the car for torture...

so nothing changed

2 hours later…

ThoriumBR

20:22

There are two hard things in computer science: cache invalidation, naming things, and off-by-one errors.

my script to restore a bunch of tapes failed because of an off-by-one error...

« first day (5149 days earlier) ← previous day next day → last day (36 days later) »

all rooms