Optimising for Process

Do we optimise for outcome or process?

(Context: Written in response to someone in a discussion about A Rant on Motivation)

This is probably a false dichotomy

Ideally our processes are such that they will lead to the outcome we want

This was the one of the main insights I wanted to get across in the essay. Interest in something is a reasonably good measure of probability of success and reward of success of doing that thing. By the process of optimising for interest, we will generally optimise for a successful outcome. We don't necessarily need to explicitly plan for a successful outcome in order to achieve a successful outcome, which was why I brought up emergent order. I talked about emergent order in economics, but we can also see it in e.g. evolution, where complex organisms well-adapted to their environments can arise without any explicit planning, through natural selection. I also recently read an interesting post about how Facebook "moves fast and breaks things". Facebook doesn't centrally plan releases - everyone commits to a giant monorepo which is then automatically deployed to production! This works because Facebook has developed a lot of canaries/monitoring that will tell them if something has gone wrong, and they can immediately fix it in hours, which is a kind of selection effect

There's several ways the model might break (which is why it's a useful model, and not pseudoscience). I've mentioned before that it's possible for the next step to be too big, particularly in things Maths which are more "difficult". (I would argue that "difficulty" is not a fixed characteristic - complex things can be understood more easily just by having a better model i.e. we experience maths as difficult because it's generally not well-explained. But that's separate from the point I'm trying to make here.)

Another way the model could break is that we might not have adequate "tests" to check that our process is getting us closer to the outcome that we want. I don't really have a good answer for this, and it's probably a fundamental problem for learning. But one partial answer I have been trying is to talk to people about my ideas. If someone else finds them useful, then at least I'm not on the wrong track. Of course I need to find reasonably well-informed people to talk about my ideas with, like through Less Wrong meetups 😄, who will be better judges of usefulness than a random person. I've also started going to government dialogues and public policy talks (one of my main interests is in governance), again to bounce ideas off other people. Even without talking to anyone else, if fancy people with fancy titles I've never met before are saying the same things I'm thinking, or are saying things that can explain why it's wrong - that's some evidence of the correctness of my learning process, right? I've also started a Masters in Economics with the vague goal of eventually getting into public policy. Since the ultimate test can only be reality

The third way the model could fail, and I think that's what you were alluding to when referring to the tradeoff between outcome and process, is that in practice we have outcomes we need to achieve in the short term to even continue "playing" in the long term

How do we handle these cases?

We want to acknowledge reality, but I don't think we want to lose sight of the fact that the process is a good thing. The exception is just that, an exception

So we want to manage the exception without changing the process, as opposed to changing a generally effective process to an ineffective one just to handle one edge case

I planned a little ahead with my maters and got private tuition in college-level maths and stats a few months before it started

Since I know I will need to spend much more time learning something intuitively than would be given because of how much I hate memorising, and because I know a lot of materials will likely be very badly explained

So the classes so far have been very smooth sailing, and I don't need to trade off a good understanding and just memorising stuff for grades because I don't have enough time

If there appears to be a tradeoff, we might still be able to weasel our way out of it. If I can't balance getting good grades and getting a good understanding, I have lots of options like taking less modules, or being satisfied with understanding 80% instead of 90%, or throwing some money towards a private tutor to get an extra class etc. Because I deeply value the process of understanding, I'm very motivated to find ways to avoid trading it off

If a tradeoff is really unavoidable, then it just comes down to counting benefits and costs, right? How often do we have to make the tradeoff? What do I lose by focusing on process instead of short term outcome? I don't really have a good example of this in the context of learning, but it's pretty cheap to show up early for an interview and leave a great impression, even if I generally think showing up early for work is giving an employer free money for no reason. But if I'm a game developer and I'm caught in cycles of hacking together demos to sell the game because I haven't developed the game enough that it's attractive as-is, well, I probably need to step back and rethink things

There is probably also more I can talk about on how to develop processes e.g. I read a long time ago that Warren Buffett said that reading books was like investing in yourself. And the more your knowledge bank grows, the more it returns in compounding interest. I think the quote was like he read 500 pages a day

Naively, I thought that I could just force myself to read more, but I was never really that successful. Processes don't really exist by themselves, they are embedded in other processes. One important process is something that tells you what books are worth reading. So I usually pick up books off of book reviews or little snippets the author publishes to e.g. The Atlantic, where 1) you get some evidence of quality 2) you get the context you need kind of as a "seed" to start exploring the book from

It's like manufacturing, right? You know steel manufacturing is important to industrial development, but tell half a billion peasant farmers to make steel and the whole country starves - there's a whole lot of other things going on

I guess another way to think about outcome vs process is that if we optimise for different outcomes, we don't avoid processes. Instead we have to develop different processes for different outcomes, and some kind of framework for switching between different processes, which I think is overly complicated and prone to failure. If we can develop one process that achieves most of the outcomes we want with occasional adjustments, I think that's more likely to succeed at achieving more outcomes