The win is correctly predicting recidivism. It isn’t supposed to be making a decision.
-
-
-
It is erroneous to conclude that the machine is better at predicting recidivism if the judge isn’t trying to predict recidivism.
- 4 more replies
New conversation -
-
-
It’s not super clear to me that the risk scores are wrong here from a social benefits perspective
-
The mismatch between the age structure of behavior, what we know about mental maturity, and modern society's age thresholds for adulthood is a big problem that is under-discussed.
End of conversation
New conversation -
-
-
I think this is a bigger issue than this in economics. We talk about the 'efficiency' increase of lots of counterfactual. But that only masks the fact that we may not actually know the underlying objective/utility.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
It’s key to remember that machine learning requires an objective function. This is extremely hard (impossible?) to specify. Try it. Try to do it for very simple decisions. Exponentially harder for complex decisions. Like sentencing.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Universal basic income would dramatically decrease recidivism. It would also decrease incarceration rates. Or we could just continue arguing if machines or people are better at locking people up based on how we define "better."
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
And things considered this still seems like a solid performance by ML.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.
Machine learning outperforms judges at predicting recidivism, but only because judges have multiple objectives -- they don't want to incarcerate young people.
How many other machine learning "wins" are actually misspecified objective functions?