There’s many different kinds of things to do

There’s a bunch of things I want or need to do, and they have different shapes.

• Some can be finite sequences of actions. Thing like this are for example “move to a new apartment”. You have actions like “get moving boxes”, “look into moving options”, “pack these things into this box”. You have dependencies like “you can only unpack if you have already moved the boxes to the new place”, or “you need to order the boxes online before packing them up”. When you finish all the tasks, you are done, and you can forget about the problem forever.
• Some are basic needs, like the need to sleep or the need to get food. If I don’t get sleep, I am slow, and will eventually fall asleep no matter how much I try to stay awake. Unlike moving to a new apartment, I will always need to sleep. I can’t, say, sleep for 24 hours and then go awake for a whole week.
• Some things take a long time and are open-ended, like learning a language. There will be a time when I will be competent enough, but it will take a lot of practice, and cannot be practically modelled as a finite sequence of steps.
• And more advanced needs, like “I want to do something fun” or “I want to hang out with friends”.

Productivity methodologies I know about are way too narrow

There’s a bunch of methodologies for productivity, but I feel they often only model a small part of planning. For example, (my interpretation of) GTD puts everything in a framing of “you have Projects and projects have Actions”. But I don’t think that’s a good framing for, say, learning a language.

The model of “cost of time” is also wrong. It is a useful heuristic, but the value you assign to any activity is going to assume perfect elasticity. For example, you cannot just walk up to your employer and say “I want to work 120 hours per week”. No (reasonable) employer will say “yes” to that. Also, the many things a person wants cannot be converted into a one-dimensional value. (Footnote: Yeah I know about von Neumann-Morgenstern theorem. But people are not rational agents, and von Neumann-Morgenstern does not say anything about how practical will the resulting utility function be to evaluate.)

What I tend to do is some sort of intuitive “higher-level planning” which is sometimes a bit reflective, but not very often. When I’m in “work mode” (i.e., in my actual job), I have a few ways I try to figure out what to do on any particular day.

But the process by which I decide, say, “enough work today, let’s go get some sleep”, or maybe “I’m a bit tired but let’s go walk on the treadmill for a while”, is very intuitive.

I don’t know about any formal methodology which would basically take an input like:

• I need to work so I get money and feel good;
• I also need to sleep roughly 8 hours daily;
• I need to have some fun;
• I also want to learn German; and
• I need to do a bunch of multi-step things before their deadline,

and which would output a plan of what I should be doing at any given time.

And I would like the plan to be reasonable. The methodology should not, for example, assume I can do without sleep, or without a regular sleep schedule.

And that’s what I mean by a “realistic planning system”.

Maybe good old fashioned AI-style planning could be adapted?

In uni, I studied a bunch of planning and scheduling, which is mostly used for cases like “this is how you construct a submarine; you can’t screw in screw B217 until you screw in screws B210-B216; make a schedule which takes as little time as possible”.

These algorithms can be extended to work in more general environments, like:

• Some people have working hours and when a worker is not working, you can’t plan jobs for them.
• There’s limited resources, like “you need a drill to drill a hole and there are only 10 drills; you can’t have more than 10 workers drilling at the same time”.

I wonder if you could make a realistic and formal planning system from some extension of that setup. Say something like:

• There is a “sleep meter”. If sleep meter goes to 0, agent must sleep for 8 hours. “Sleep meter” slowly goes down during the day.
• There is some sort of penalty on effectiveness when underslept or when sleep is irregular.
• Different types of actions deplete other types of meters. Light socialization depletes “introvert points” (for me). Snuggles increase “oxytocin meter” (which is a dimension of “happiness”).
• And there could be some modelling of “if you get too tired you have to sleep”. Perhaps the algorithm would compete for control with “drive to sleep”, and the lower the sleep counter is, the higher the likelihood the person just falls asleep wherever they are.
• And I think it would be nice if the algorithm treated all “needs” symmetrically. According to internal family systems, people have many relatively smart parts good at different things and wanting different things. If the system is cooperative and does not place any particular part at a “command” level or at a “subordinate” level, it would hopefully make it easy for parts to agree to collective decisions.

Meh, too hard. I got other stuff to do.

I guess getting any model halfway realistic would be too complicated, and I probably will keep using my bunch of ad-hoc heuristics to make decisions. I have way too many things that I want to do to spend the first couple years developing a planning algorithm rooted in psychological theory.

A thing I used to do that might be useful to start doing again is having some time in which I try to optimize what I’m doing. It could bootstrap into more conscious/mindful action.