Proprietary algorithm that predicts future criminals okayed by Wisconsin court | Hacker News Reader

Proprietary algorithm that predicts future criminals okayed by Wisconsin court

fusion.net

27 pointsnonprofiteer10 comments

I can't think of any reason why anyone with a decent grasp of technology or even logic and rational decision making in general would consider the use of secret, proprietary algorithms permissable in the due course of justice. So why is this deemed acceptable here? Is it a lack of understanding how a decision making tool differs from, say, the brand of microphones used in the court room?

Mind, I don't see how anyone can condone privatized penitentiary institutions either. Perverse incentives and all.

This stuff has been in use in the probation and parole system for decades now.

Whether it's a "sophisticated" algorithm or a sheet where a parole officer writes down numbers and adds them up, there's this constant idea that it can be used to determine how likely an individual is to re-offend.

Rejecting the use of such a system would call into question the methodology used in deciding whether or not to revoke someone's parole/probation - this is incredibly significant, because, to take the example of Wisconsin, over half of the state prison inmates are in on revocation.

I'm just saying this by way of presenting an alternative idea of why the court might have rejected the appeal. It's also important to note that in Wisconsin we have a very conservative supreme court.

I find that the more I learn about the criminal justice system, the more shocked I am that we allow it to continue. It's not a law and order vs. wishy-washy liberalism thing - our criminal justice system demonstrably increases recidivism.

The idea of an excel spreadsheet and points is at least understandable.

A cynic might think that the value of a complicated algorithm is you can, as Pinboard said in a similar context, money launder your bias.

Hell, I do ml on a daily basis, and for anything but a regression or a decision tree, it's complex and/or impossible to even explain why an algorithm picked what it did for a specific example. Let alone to evaluate what an algorithm is picking up on in general.

And that doesn't even get into the mess that is highly correlated variables (eg ethnicity, SES, income, peer income, parental income, education, arrest rate, housing location) most of which are largely synonyms for each other. And the bias in the data themselves -- being poor or minority increases the detection rate of criminality.

> money launder your bias

I like this phrase. I'm stealing it :-)

I stole it first!

source: https://twitter.com/Pinboard/status/744595961217835008

Thank god. It's about time we finally use technology for good, and governments can finally do their jobs effectively without the guesswork.

Of all the weird things of Cyberpunk culture, the one I never thought I'd see in my lifetime is Psycho-Pass. And here we are.

It would be insanely unfair if this algorithm was based on statistics around economic (or racial) status.

From what I have read elsewhere, the data sets used for this type of decision making is biased towards race. That is an area of research currently.

https://www.fordfoundation.org/ideas/equals-change-blog/post...

Is this how AI will begin to segregate our futures and our lives?

Reply on news.ycombinator.com