Shame we're seeing places explicitly banning facial recognition rather than biometric recognition.
Killing one person in any kind of failed transaction is a bad enough offense that I believe you deserve at least a decade in jail. That said, I do believe in rehabiliatory justice, and that we shouldn't just throw people into an environment that's essentially some cross between a dysfunctional highschool and a networking seminar for criminals; but people who are willing to kill over $100 do not belong in public until they have proven conclusively that they have enough self control to be released.
The opiate epidemic, on the other hand, is a crime so heinous that it is nearly genocide-tier in nature. The perpetrators of such (i.e. the Sacklers among others) knew what they were doing, continually doubled down over time, and profited tremendously. Such parenthetical elites usually suffer absolutely zero punishment for their actions, nor is there any attempt to make rehabiliatory justics take on the challenge of so-called "white collar crime". At best, the perp learns how to hide their intentions and actions more, how to operate behind more intermediaries for abstraction's sake, etc.
Now, consider the topic at hand (which is somehow facial recog and not Daft Punk whom I loved dearly): what is the common factor here? It's simple: if you perform an analysis of the group memberships of the people who commit crimes in the aggregate, uncomfortable truths are revealed.
https://www.npr.org/2020/06/24/882683463/the-computer-got-it...
https://lawandcrime.com/civil-rights/third-innocent-black-ma...
People already know the root cause, and it is overpolicing and discretionary enforcement of crimes like drug possession.
The broken-windows policy has been disproven, despite disproportionately impacting "the specific groups" (black Americans) that now pollute the dataset that ML uses.
And even in the concept or drug possession and drug consumption, all groups have been shown to use them in the same distribution. For example. These kinds of things start a cycle that means the second and third minor infractions cause greater consequences in court, which further reduce opportunities that lead to the dangerous crimes being committed.
So we already know the dataset is polluted.
Pointing to the top of the iceberg and saying "well they commit the crimes no need to spend any energy on this mystery why don't we all just admit they're the problem" is a complete deflection promulgated intentionally and you should really check your peer group and media sources if this is the extent of the comfortable worldview you (or anyone passing by) have, the problem and solution is already known.
The solutions are being implemented in a patchwork and slowly, which is not incorporated in datasets that ML use, largely due to apathy and lack of awareness of engineers and the management of the tech companies involved, and lack of representation of the affected groups in engineering and management of tech companies.
No, you've only pushed the root cause back a step by doing this. It's entirely possible to understand that the way we're doing policing is wrong without resorting to an artificial, borderline-creationist mindset on aggregate group behaviour. Religious thinking is not going to help anything here.
> So we already know the dataset is polluted.
The dataset reflects something approaching reality. If anything, it reflects a version of reality that is already attempting to artificially compensate for group differences in order to quell conflict.
If you want to change reality, if you want to see less crime, if you want to see truly fair policing, then admitting to reality is an important first step. You're basically arguing for "juking the stats" in order to find fairness, when in reality such actions won't stop people from getting robbed, murdered, or having lives that offer so few opportunities for advancement that they end up turning to hard drugs to cope. ML is not to solve this problem either way, but it can actively prevent the problem from being solved if it becomes yet another mechanism to paper over the actual situation and instead point the finger for responsibility away from where it belongs: with individuals and their choices.
It's a shame it has to be like this. There's plenty of valid statistical / demographic information out there that lines up perfectly with everyone's lived experience, but the conclusions drawn from it are not pretty. All I can say is that it is possible to make inferences without making value judgements; we can point at problems without having evil intentions and without suggesting tyrannical interventions.
I could see that being a disguise