It's far from obvious to me that an AI would be able to inspect a complex system like this for (potentially unanticipated) failures more effectively than a knowledgeable, experienced human.

And aside from the proximate cause of the fire, the thread highlights a deeper and more insidious issue, namely the diffusion of management responsibility for decisions and their consequences. It seems to me that delegating inspections to an algorithm would make that problem much worse, not better.

I doubt it would have picked up this failure. That said, I think dones + AI can reduce the cost of routine maintenance, the savings of which shoud be used for infrequent detailed inspections that would pick up a defect like this. More likely the savings will just be pocketed

I believe so. People can probably set up some physical models and simulate the wear against various elements, and possibly detect similar wear in the reality. Even without AI, human inspectors can remotely monitor the tower using drones and check more checkboxes than one.

By employing such effort and maintaining towers more responsibly, PG&E at least could partially evade the blame and liability. IMO this could be a good startup idea if founders know the right people.

Yes. I saw them doing just this in South Korea in 2015. The guys doing it were pretty chill, said I could not take pictures of their drone but they told me a bit about how it worked. Computer vision was part of it, human inspection was too. They said it was much safer than a manned helicopter, and faster.