Police body cams with automated "AI" report writing integrations. And that "AI" is trained on what, again?

That's right: Parsed and tagged datasets of police reports… which are themselves written after the fact and based on cops' memories of the events. Things which notoriously distort and amplify certain elements, especially concerning race, and often in a reciprocal relationship to whether a crime is charge, for whom, and of what level of severity.

Everything is going to go wrong with this because it starts off from a place of utter wrongness.

@Wolven I'm all onboard with the point you made, but just as a thought experiment...

Wouldn't it be impossibly hard for the ai to be as biased as the human colleagues? #smallsteps

@iwein @Wolven On the contrary. When you train it with biased data, you get output that's just as biased.

So not only would it not be "impossibly hard", but rather it would be the baseline expectation for it to be as biased as the human colleagues.

@LouisIngenthron @Wolven I'm still in, but an ai that's obviously biased will get destroyed (hopefully), while a human will only get a talking to (unfortunately). If there's no systemic intent on racism, the ai would learn faster under these assumptions.

Iaw, the genetic algo training ai is going to beat the whatever feedback loop is supposed to keep pigs clean now... Unless @scream

@iwein @Wolven That's assuming there would be checks in place to identify the bias to destroy in the first place. And those checks, if they're done at all, would be done by humans... probably cops with the same biases we're trying to eliminate.

Sign in to participate in the conversation
Qoto Mastodon

QOTO: Question Others to Teach Ourselves
An inclusive, Academic Freedom, instance
All cultures welcome.
Hate speech and harassment strictly forbidden.