I think you misunderstood my post? I'm comparing AI as a system vs written code as a system. Both systems can have flaws, but the way in which they fail is different. The danger comes when non-technical people try to apply intuitions about software failures to AI, because those intuitions are false when applied to AI.