Lesson n. one from my Masters degree

AI is, indeed, dangerous. Not because it will SkyNet us into the oblivion, but because far too many malicious or, worse, incompetent actors try to use it, burning ridiculous amounts of resources and creating systems that, in the end, benefit either no one, or only small groups of people.

However, when used correctly, the AI approach yields great results. There are two main use cases: approximation of computationally demanding or noisy data, and generation of objects. The former can often be achieved via other means like "classic" ML or approximating parameters of known function. The latter, however, is almost unreasonably good in certain domains like prediction of chemical reactions in organic chemistry. It's a shame we mostly use these systems for generating low-value text and cartoonish porn. Or whatever do people use it for.