I think the main idea is ceding control to a system to make the kill, don't kill decision. This is the "so" part of the article and the comment.

With wmds, there are humans who make the choice at the end of the day. The loss of life scale is not relevant as much since we could make ai powered wmds as well, the fundamental question is "can a computer kill?" And maybe more importantly "who is responsible for it making mistakes? L"