I am not necessarily against this as long as there is a human override option.
In either direction?
Yes.
When I worked with Drug Courts we used risk assessment tools and they were extremely useful. Our success rate improved a good clip after we implemented it. But we did have a system that our counselors could override, depending on the behavior of the client.
Of course, if you find yourself overriding a lot, then the question might be, can the behavior (or whatever) on which you’re basing your override something you can build into the model?
I have some skepticism of the algorithms because they can often include biases, but the cash bail system isn’t working.
The algorithms themselves can’t have biases, only the people that created them could.
Using data is almost always better than using humans.
People build algorithms. People with biases. Maybe not intentional ones, but they are there.
Agreed, but most of those algorithms have so much research built in that bias’ are next to nil
The article said that the algorithm didn’t change the racial disparity. Maybe the data backs up the racial disparity.
This conversation reminds me of people complaining when retailers put items behind locks that have high shrink numbers.
The algorithm looks at a SKU number but apparently it is racist
According to the article. The data was flawed.
I’d have to see the algorithm to be certain.
I assume they’re matching data to prior arrests, etc. such as someone with a crack charge is less likely to return to court.
We need to stop saying everything is racist and look to see if it’s valid.
People were saying target moving out of south Chicago was racist. No, they had unacceptable levels of shrinkage.
This topic was automatically closed 3 days after the last reply. New replies are no longer allowed.