|
> "low-income, ethnic minority neighbourhoods have historically been over-policed so the data shows them as crime hotspots, leading to the deployment of more police to those areas." This is wrong. They've been under-policed. And the unwillingness to not lie about that, let alone understand why that is, is why "predictive policing" is disliked. It's also why the rhetorical trick of "systemic racism" is necessary to elide the fact that racism is rare today and justify why politicians are attacking & dismantling functioning institutions rather than fix the underlying issue. Does anyone seriously claim that suburbs like Beachwood, OH and Palo Alto, CA and Everett, WA would discover a bunch of undetected rapes, gun violence, robberies, and murders if they were "over-policed" like low-income black neighborhoods? No, they don't. Because that would be too silly of a claim to make directly to anyone who has lived in such a place. There's a humane chain of logic we must follow here. Do we think slavery, red-lining, drugs & drug-enforcement, and the collapse of rising working class wages have hurt African-American well-being? Yes. Does poverty correlate with crime? Yes. Does poverty correlate with African-American neighborhoods? Yes. Ok, so we can admit that African Americans account for a hugely disproportionate amount of violent crime in the United States without disrespecting how black communities got to this point or worrying that this simple, obvious conclusion is driven by racism. This is trivially produced from well-done statistical research. If you control for either poverty rates or crime rates, you recover the same rate of police use of lethal force for both whites and blacks. "Oh, well, the data sets from the past may be biased with racism" 13% of the population accounts for between 40-50% of the homicides, year over year. Alright - How large do you think purely unfounded biases account for that? Half? Even then, you have both 1) A LE agencies either falsifying or mishandling data on a massive scale, a monumental scandal in itself and 2)a population accounting for 2x rather than 3-4x their expected homicide rate. Not a huge improvement. I don't feel like the claims of "warped data" deserve this much respect because the people who make them never give the same thought to the much more obvious conclusion - The data is true, but I do. ML is disliked here because it doesn't lie. Politicians cannot torture or cajole logistic regression to lie because it doesn't have a career, it doesn't need a promotion, it just graphs the truth. But more than this, they dislike it because it indicts them for not doing more, and more meaningful and honest things, to actually improve the lot of African-Americans in this country. The first step to doing that is being honest. |