There was a post yesterday giving an AI “answer” to a complex question.
The AI acknowledged its sources—and one of those was the Anti-Defamation League (which is itself the master of defamation).
AIs have the same problems humans do—they are only as good as their sources.
If they are programmed to blindly accept things without doing detailed research then we just get garbage like we do now.
That’s true.
If AI were to be used in a justice system context, I would want it on a closed network. Don’t pull opinions from reddit and treat them as useful input. Just a closed network with access to US laws, legal decisions, evidence and testimony. Add good logic circuits, and within that narrow data field determine if a crime has been committed at all, and if it was, is the defendant a guilty party?
I think we will eventually get there. Our current experience with judges and juries makes me think that we really need to do better.
AI does not think; it aggregates “information” from the Internet.
In addition to: Garbage In, Garbage Out.
We now also have: Garbage Algorithm, Garbage Out.
(See also “global warming”.)