AI error jails innocent grandmother for months in North Dakota fraud case

[AZJonnie]

22. Agreed

I admit I use AI quite a bit to confirm the accuracy of news stories and get a better understanding of events, and the absolute #1 thing AI gets wrong is overstating the certainty level about things that are inherently uncertain. Basically it tries to sound more authoritative than it should, fairly often. It will also try to tell you what it thinks you want to hear, although ChatGPT is by far and away the WORST in this regard, they all have a tendency to judge your intent.

To combat these issues, I prompt with this specifically in mind with things like "I see some recent news stories that say X and others that seem to say the opposite. Please provide a detailed assessment of the available evidence for each of the assertions and show me your work in determining which is more likely to be accurate". This way it doesn't "think" for lack of a better word, I want a definitive answer for the question "is x true?", nor that I have a "preferred" answer. Give it permission to be non-authoritative and unbiased, tell it you just want to see the evidence/arguments on both sides, and it will do better things.

It may still not find (and hence consider) evidence that you'd hope it would, however. There's limits to the scope of searches it will do so that's good to keep in mind as well.

Bottom-line, this all makes it pretty damn sketchy to use for decisions that can really impact people's lives