Kerrigan comments on why assume AGIs will optimize for fixed goals?

[Kerrigan]

Is this trivializing the concept of a Utility Function?