Hallucinations are not the same as errors

I asked my AI about some obsolete card games, and it wrote a 1,000 word essay about Lansquenet. It made up the rules, the strategies, the betting techniques, all of it.

Saying that it was raining on July 14th last year is an error. Inventing an entirely new set of rules is an hallucination.

An AI can’t lie, because it doesn’t know it’s lying. In that sense, it’s sort of pathological, careening forward without a conscience. Humans anthropomorphize things, because it makes it easier for us to process and predict. And AI is tempting us to do that more than we might with an ordinary machine.

That’s probably not the most helpful strategy, and it’s setting us up for real trouble.

AI is a tool, and judgment, for the foreseeable future, remains our job. It doesn’t matter how cool your hammer is, it’s still on you to decide which nails need hammering. And to be responsible for what happens when you use it.

Source link