ML for cancer-diagnosis of image-scans?
Identification of endangered species?
Translating dying languages?
I get the point you're making, and would add that some ML applications bring tangible benefits to humanity which are proportionate to their energy consumption.
You'd need a "proof of ethics" and that seems technologically dubious.
Or, you'd be using blockchain to create an ethical application, and whatever benefits are realized by the application are probably not best provided by blockchain.
Using computing to compute meaningful data is directly justified or at least directly measurable.