zlacker

>>pseudo+(OP)
People are boiling this down to "less crime = good". Like if a mathematician can design a deployment system such that the percentage of crimes where an arrest is made increases from 15% to 20%, how can that be a bad thing?

What if every single one of those additional arrests is a Black person? Well that might require an additional look. It's possible that it's all Black people doing those crimes. It's also possible that the mathematician built an algorithm anchored on years of arrest data from racist cops doing racial profiling.

Will the mathematician be able to say, "hey cops can we take a second look?". No. Can the police be trusted to dig into the nuance themselves? Also no. Is racism active and documented in police departments across the country? Yes.

Should everyone who has any ability to do so stand up and say, we won't support this shit, which is what these mathematicians are doing? I think yes, but that's obviously my opinion.

>>alexil+Pv
Statisticians will be the best of all the disciplines at teasing out that sort of bias. If the police are going to be racially biased they don't need statisticians to help them. The addition of statisticians is only going to increase the number of people who are uncertain about the model. There are two broad paths here:

(1) Statisticians predict where crimes are likely to happen for the police using data.

(2) Police predict where the crimes are likely to happen using highly biased guesswork.

The statisticians are being irresponsible pushing the police towards (2). Option (1) can be de-biased, can be debated, its effects can be assessed and its parameters can be tweaked over time.

This is a choice between two options and the people signing on to this letter are arguing for the worse one. They link a bunch of news articles, but the academic complaint seems to be that the model says "assume crimes happen in high crime areas". That isn't a very scary model, and if anyone has an alternative they should be pushing it towards the police, not holding it back.

>>roenxi+uU
I think this is a good point, however over time I believe less and less of option (2) will be used. Option (1) will become ubiquitous, and it is important it is done well. The researchers signing the letter should be exactly the ones helping improve the models used, as they obviously care enough to consider the inherent biases and societal effects.

Just because they choose to "opt out", it doesn't mean everyone else will. It leaves the task to a smaller pool of competent people (best case), or unscrupulous people out to make a quick buck or push an agenda (worst case).