Top-level takeaway of tonight's catch-up reading on ML... I still find it hard to take AI Risk seriously as a special problem here. There are definite risks but they don't seem qualitatively different from other kinds of engineering risk.
So maybe I need to add #3 "cannot be contained within a specific substrate or environment by it's creator" to avoid the comparison with virus. The meme point is just too abstract for me to get my head around right now.