existence of objective function

In this post I talk about the existence of my objective function. Some related and important questions include: Can I change my objective function? and Should I change my objective function? But it’s impossible to measure diffs to my objective function without actually knowing what it’s starting out as.

Fact 1: Alek is an agent with preferences. Alek prefers some states to other states.

For the duration of the post we shall use to denote Alek’s objective function, where is “state”.

Fact 2: Alek is an agent that acts. Alek’s actions have an impact on , which in turn impacts

Fact 3: There is no such thing as a no-op action. Time increments. Entropy rises. No moment ever recurs. There are (presumably) no replays.

---

Fact 1 is probably the most intuitive, but it still has some complexity. One challenge relating to fact 1 is that I don’t have direct access to . At best, I can obtain noisy approximations to . Factors contributing to the noise I must fight when trying to determine include:

• how much I slept last night
• how much food and water I’ve been consuming
• how I have responded to, and perceived, the events of the day
• the set of cognitions that I’ve been letting be prominent in my mind
• the set of stimuli that I have exposed my brain to

But in some sense modelling my understanding of as being “I get ” is overly optimistic. A more realistic model might be that my brain runs processes like avoid_discomfort.cpp that are each allowed to shift mess with the output of my query to and they may do so adversarially for their own nefarious purposes.

Fact 2: I have responsibility.

Related to Fact 3: wrong exists, and it’s totally fair game for “status quo” to be wrong. Furthermore, procrastinating choices is a choice, and often wrong.

See the objective for a description of what my objective function actually is.