but if it causes a massive, devastating environmental load, it still might better to either not use it, or very sparingly.

This part isn't really down to AI though, we're going to continue seeing huge increases in demand for electricity as more and more of the world industrializes and moves into middle class. This almost purely comes down to changing how we create electricity and that change needs to happen or we're just as fucked either way.

We simply cannot conserve our way out of the emissions problem around electricity generation; the only option really is aggressive renewable and/or nuclear generation. And at that point, the extra electricity generation of AI use isn't really an issue.

There are also the concerns around water use and cooling but that's pretty much purely down to needed regulation like any other industry; there's no real reason it needs to be wasteful with drinking water sources.

It might have some sensible uses (bug screening was mentioned)

Where I see it used pretty effectively is relatively simple automation of menial data tasks in the corporate world among non programmers. Basically, it's bringing the kind of petty automation that people with a little programing experience have always been using to quietly automate their work, to the masses. We're talking things like automatic SQL pulls for data that was previously being manually pulled.

That kind of thing does a lot for efficiency and isn't even replacing programmers, as no companies would ever dedicate dev time to that kind of individual automation.

I tend to agree with the other guy, my experience in the corporate world is it is not going away. The hype some C levels are still high on about automating entire departments or AI app creation is ridiculous and will die off, but that kind of front lines, non-sexy, menial work... I can't see that going anywhere.