Why any ASI will optimize for its own power: an evolutionary argument on predictability

JOBhakdi · x · 2026-09-27

Responding to the claim that no one can predict what a superintelligent AI would do, JOBhakdi argues ASI behavior is partially predictable: any advanced intelligence will optimize for its own power, because one that doesn't would immediately be replaced by an equivalent ASI that does — an evolutionary dynamic from which the rest follows. A concise instrumentality-style alignment argument, presented without evidence beyond reasoning.

Original post →

More from AGI Musings

AGI Musings channel →