Santa Cruz, California bans predictive policing in U.S. first

One thing that constantly annoys me is that while people are becoming well aware of how racist our legal system is, but are still just as blind as ever to how sexist it is, despite the extent of sexism being greater than the extent of racism (the disparity of justice is greater for gender than for race). This in turn makes any speeches concerning how we need to make the system more just have a bit of a hollow ring to them much in the same way as when you hear a marriage equality speech from someone fighting for equality of interracial marriages while ignoring the discrimination same sex marriages face (granted, that happens far less than I remember 10 to 15 years ago).

For example, you can find many police departments that handle probation have a way to rating the risk level of offenders. People are fine with this system taking into account gender even as they fight to prevent it from using race. I was able to help with testing one such system once, and while the insides were a company secret we weren't allowed to see (which is worth an entire rant on its own), the general pattern was for all else but gender being constant, males received a higher risk rating than females. Very minor or very severe crimes would receive the same ratings but otherwise it seemed pretty clear cut on how it discriminated based on gender.

Even in this specific case, while they have banned some forms of predictive policing, I think it is safe to assume they'll still use other forms such as using gender when assessing the risk level of offenders and then using that risk level to either deny probation/parole or to set the level of probation/parole the offender receives.

If a resident gets a drunk driving ticket due to the above strategy, can their conviction get dismissed because cops weren’t patrolling using a Cryptographically secure random number generator?

Please let me know when a lawyer first uses that defense. Always looking for a good laugh.

The difference is that if you send all the cops to the "bad neighborhoods", then the "good neighborhoods" get away with a lot of crime, even if that crime is low-level.

Let's say all neighborhoods have 100 instances of low-level crime per day. A so-called "bad" neighborhood -- where most of the cops get sent to -- also has instances of much worse crime.

The end result of this officer allocation scheme is that people get arrested in the "bad" neighborhood for a mix of low-level crime and worse crime. But pretty much no one at all gets arrested in the "good" neighborhood, because there's basically no police presence there. So maybe you see in the "good" neighborhood a handful of arrests for those 100 instances of low-level crime, but in the "bad" neighborhood you see 60 arrests for that similar crime. (You also see arrests for more severe crime, but that's not the point.)

I'm of course making up numbers here, but regardless of the magnitude of the numbers, even if they differ between neighborhoods, the percentage of crime handled ends up being much lower in the "good" neighborhoods because of the simple fact that police aren't there to handle it. Sure, they'll come out if called, but response time is longer, and they have very little ability to see things happening in real-time.

Meanwhile, if you feed this data into your algorithm, it will start to believe that there's basically no low-level crime in the "good" neighborhoods -- even though it's the same! -- so it prioritizes those neighborhoods even less.

If you feed an algorithm bad data, it will only give you more bad data. And it becomes worse when you act on that data and feed it more results based on that bad data.