And the issue that it can easily be manipulated by the coders. "give lesser sentences to poor black men because of their circumstances" or "prioritize the mother over the father in custody cases" are things that can easily be added to the models.
There was an article a while back where medical researchers were using medical data in AI/software training, and the researchers discovered that the software was able to predict a subject's race to a disturbing degree, with very little data.
Said researcher, being a shitlib, had much wailing and gnashing of teeth over this, but I thought it indicative of the potential if people just stop trying to constantly cripple the damn things.
Sure, so long as they manage to fix the minor minor issue of "will hallucinate entire cases" part.
And the issue that it can easily be manipulated by the coders. "give lesser sentences to poor black men because of their circumstances" or "prioritize the mother over the father in custody cases" are things that can easily be added to the models.
That's not the part people will care about.
There was an article a while back where medical researchers were using medical data in AI/software training, and the researchers discovered that the software was able to predict a subject's race to a disturbing degree, with very little data.
Said researcher, being a shitlib, had much wailing and gnashing of teeth over this, but I thought it indicative of the potential if people just stop trying to constantly cripple the damn things.