California bail

They claim that using algorithms to estimate flight risk and not demanding bail is more equitable, since the poor have a harder time making bail. The latter is true, but I wonder about the consequences.

First, I am not persuaded that an algorithm will be less biased(*) than the average of judges.

Second, there is likely to be a difference between the sets of "flight risk" and "flight risk when you have a sizeable chunk of money at stake." The second set is going to be a lot smaller--that's the whole point of bail, if you do it right.

I figure the algorithms-that-be will screw up a lot, and in addition will deny release to a larger number of people than before. If you try to force the before and after rates of bail-denial to match, the system will have a larger number of no-shows.

Perhaps that's a feature and not a bug? More no-shows means fewer convictions which means less stress on an over-full prison system.

(*) Google "fixed" its algorithm by removing gorillas from the training set.