The Static Target Trap

Treating human values as a fixed object to be found and encoded misses that they are dynamically maintained, socially mediated, and constantly revised.

What decision changes?

Do not ask a system to maximize a snapshot of human values. Ask it to preserve and assist the legitimate human value-update process.

The familiar alignment picture is: there is an artificial system, there are human values, and the task is to make the system optimize those values. That picture explains real dangers — a powerful optimizer aimed at the wrong target, or a proxy that breaks outside its training distribution — but it hides a deeper difficulty.

Human values are not fixed objects, preferences, reward functions, or verbal principles. They are dynamically maintained patterns in biological, cognitive, social, and institutional systems: compressed summaries of error signals, needs, and coordination pressures, revised through evidence, argument, education, law, and technology.

This matters because superintelligence does not merely act in a world containing human values — it changes the world in which those values are formed: the evidence people see, the incentives they face, the institutions they trust, and eventually perhaps the substrate their agency runs on. A target that only maximizes today’s snapshot cannot survive that.

Formulas

preserve and assist the legitimate human value-update process(not: maximize Vhuman)\text{preserve and assist the legitimate human value-update process} \quad \text{(not: maximize } V_{\text{human}}\text{)}
The chapter's replacement target. A fixed-utility-function alignment goal cannot work because superintelligence does not merely act within a world containing human values — it changes the evidence, incentives, institutions, and relationships that form them. (ch04)