[loup-vaillant]

No human should ever be root. But we might just trust a Friendly AI. Well, provided we manage to make the AI actually Friendly (as in, does exactly what's good for us instead of whatever imperfect idea of what's good for us we might be tempted to program).

[Filligree]

And if we don't, we all die (at best), but that's nothing new. Nor is it avoidable by other means than FAI.

The route to unfriendly AI is revenue-positive right up until it kills us.