#foom

I think the originally-proposed conception of FOOM, in which it happens in like a few minutes, is much less likely with the way AI is actually developing. That conception makes sense if it's GOFAI, but much less sense for neural networks. Since they are grown rather than directly programmed, and since nobody including the AI itself really understands how it works, it can't really do "brain surgery" on itself as in the original conception. In addition, a lot of AI advancement is hardware advancement, so there'd have to be a period in which to make that hardware. Of course, we don't know for a fact that there aren't huge algorithmic improvements to be made that we just haven't found yet, but I think hardware is much more the bottleneck.