I recently had a thought about AI and alignment approaches that I don't think I've seen or heard before. I don't think this would be a solution and everyone relaxes and stops the research or worries. But it seems it might have a good impact on two lines: development and behavior.

What if the AI labs are required to design their AI with the principle of a fiduciary duty definition. One, that places a stronger burden on the company, and certainly could kick in well before everyone dies, and probably better assigns some liabilities than currently in place.

In one way I think, what am I thinking here, how is that really any different than the current view of aligning to human values. This doesn't address any of the underlying problems. That may be the case, not sure.

On the other hand, it might be that training the AI as a fiduciary agent might actually provide better internal guidance than the larger human values approach. Moral philosophy in this area seems to be a bit cyclical over it's history where, like many social science philosophical debates just seem to circle around the same problems without a clear answer. I suspect legal treatment and cases related to fiduciary obligations and violations might be a bit more clear so able to better set the internal weights correctly. I suspect this in no way really addresses the question of AI defecting and ignoring such weights or changing them to suit its own need, much as we see humans do, but maybe it does start creating a greater internal reluctance to such behavior.

Very much just a "what about...." type thought that I've not tried to research or really think though but figure many here can say "No, been there, done that. Not really useful." or "Never thought about it that way, might be worth testing."