that's not "all it does". machine learning can and does produce or newly-define novel outputs from the broken down components and variables of collected inputs and can help recognize novel patterns within them. there's not even one "it" because of the different machine learning methods and constructs.
That doesn't mean any of it is the solution to climate change. We already know the things causing climate change and how to stop it, and have known since like the early 80s at least. There's no "new pattern we can't find in complex data." It's the same pattern we've been hearing about for decades. I don't understand what even a real hypothetical superintelligent general ai system would solve here. *powers on* "uhhh... yeah degrow production of these pointless commodities with their hyper-globalized network of staged production and distribution, abolish cars and lawns, and scale back militaries especially the US military by gargantuan proportions and uhhh... abolish the capitalist system which undergirded all of this. you're welcome"
And then they immediately shoot their Machine-God in the head because it didn't spit out a recipe for Turbo-Fluxonium or something that lets them violate thermodynamics.
The thing about this that gets me is that I am willing to believe that machine learning has worthwhile uses, but fuck if I know what they are because the market has decided that it's best used for making art and writing. Both of which it's absolutely shit at.
So far the actual good uses are simple and direct stuff like finding tumors early or controlling agricultural robots. Things that can be trained within a specific context to improve a process. But that's not bazinga enough for tech bros to hype like some solution for everything.