Created: April 05, 2020
Modified: January 24, 2022
Modified: January 24, 2022
choicefulness
This page is from my personal notes, and has not been specifically reviewed for public consumption. It might be incomplete, wrong, outdated, or stupid. Caveat lector.- I got this concept from NameRedacted.
- People overestimate what they can do in a week, but underestimate what they can do in ten years. To do big things over that time frame, you need a plan. (even if it's constantly revised) (TODO maybe the whole successive section should be about goals rather than plans---I think choosing goals vs non-goals makes more sense than plans vs non-plans. Though in some sense goals are just steps of a higher-level plan, ad infinitum).
- You can view a plan as picking out the actions you're going to do, but an exponentially larger view of a plan is the actions you're not going to do. Focusing on one thing means not focusing on another (seems relevant to attention). There must be real trade-offs.
- If you want to become a doctor, you can't also become a lawyer and a scientist and a politician. If you try to do all of those at once, you'll end up with none of them.
- It might be possible, with a realistic plan, to become these things in sequence. Anthony Fauci is, in various senses, a doctor and a scientist and a politician. But it's not going to happen all at once, and it's going to require focused effort in each area.
- When making a plan, we should ask: are we comfortable saying we're not going to do the things that aren't in this plan? (unless circumstances really intervene). If not---if we genuinely think those other things are more important---then we should revise the plan.
- This seems especially important on a personal level. I need to be honest with myself about paths to my goals and disciplined in the actions I take to pursue them.
- It's also valuable in a workplace context---we need plans, a strategy, goals, and to mean anything, they need to be choiceful. But it's harder to be the dictator of a workplace---saying we're not going to do this, even if people want to---than it is to be the dictator of yourself. (which is not to say that dictating oneself is always easy).
- Question: what can I say about choicefulness under uncertainty? There's some level of total chaotic uncertainty where it makes no sense to have plans. That's probably true if we're talking about uncertainty over world state, but it's definitely true if we're talking uncertainty over goals.
- But it's also not totally true even there. First, there are universal instrumental values that we would aim for even in the absence of goals. This is what gives curiosity-driven learning its juice. Second, humans are not just MDPs; we learn our self-models from our own behavior. Picking a goal arbitrarily, and working on it, can eventually convince us that we actually care about the goal. That's a different kind of choicefulness, although we could also view it as plain sequencing. If we have a uniform distribution over goals, then in expectation we want to achieve all goals: to do this, it's better to choose a path to run down, than to try to follow all paths at once.
- Don't average over actions (the parable of the Bayesian in the woods). You're following the criminal and come to a fork where they went down one of two paths; you're unsure which. The correct Bayesian thing to do is to average over utilities: since you have no information about which path the criminal followed, you find that they have equal expected utility, so you should choose one of them arbitrarily (though in reality maybe try to gather information?). The incorrect thing to do is to average over actions. Don't take a step to the left, then a step to the right, then the left, etc. and wind up in the middle of the woods. Choosing one path means not going down the other path, at least right now.
- Takeaways: When I write down my own goals and plans, also write down non-goals, and non-plans.
- Observation: A natural way to be choiceful in life is to choose the people you're around. I might personally be uncertain whether I want a career as a programmer, or a musician, or a farmer, but I probably can't exist in all of those communities at once. Surrounding myself with programmers will tend to push me towards becoming one, since conspiracy is a thing for a reason.