AI risk arguments often gesture at smarter AIs being "closer to a perfect utility maximizer" (and hence be more dangerous) but what does this mean, concretely? Almost anything can be modeled as a maximizer of some utility function.

The only way I can see to salvage this line of reasoning is to restrict the class of utility functions one can have such that the agent's best-fit utility function cannot be maximized until it gets very capable. The restriction may be justified on the basis of which kind of agents are unstable under real-world conditions/will get outcompeted by other agents.

What do we mean when we say a person is more or less of a perfect utility maximizer/is more or less of a "rational agent"?

With people, you can appeal to the notion of reasonable vs. unreasonable utility functions, and hence look at their divergence from a maximizer of [...]

---

Outline:

(00:48) What do we mean when we say a person is more or less of a perfect utility maximizer/is more or less of a rational agent?

(01:55) Unsatisfactory answers Ive seen

(01:59) A1: Its about being able to cause the universe to look more like the way you want it to

(02:24) A2: Its more rational if the implied utility function is simpler

(02:43) A3: Its the degree to which you satisfy the VNM axioms

(02:56) The most promising answers Ive seen are ways to formalize the reasonableness restriction

(03:02) A4: Its the degree to which your implied preferences are coherent over time

(03:40) A5: Its the degree to which your implied preferences are robust to arbitrary-seeming perturbations

---

First published:

Source:

---

Narrated by TYPE III AUDIO.