My point 3 was, somewhat snarkily, that AI will be built by humans on a foundation of crappy software, riddled with bugs, and that therefore it would very likely wind up crashing itself.

I am not a techno-optimist.