[This is probably obvious to some of you.]
I’ve been thinking about building self contained systems lately, specifically (mostly) in the context of personal productivity. If you want a self contained system that is robust to disruption, you want it to incorporate control systems.
My approach to building a self contained system for my own effort-less efficiency has been to identify the inputs and intermediate states to the psychological states that I’m aiming for, and then build control systems around those inputs and intermediaries.
For instance, if you know that subjective mental energy is one important input to flow, and when you don’t have mental energy, you become draggy, and concentration is elusive, then you want to set up an automatic system so that whenever your mental energy is low, you automatically take actions to recover it.
The problem of systems that depend on the inputs they control
However, this has a problem. If you have control systems that depend on the relevant input themselves, then they can’t really function as control systems.
For instance, speaking of mental energy again, suppose you know that if you go to the gym and physically exert yourself, you’ll experience a gain in subjective mental energy. But, unfortunately, going to the gym itself has some activation energy, and so if you are low on mental energy, you’re unlikely to do it. In this situation, you’re stuck in a less than optimal attractor, where feeling draggy prevents you from doing things that would help you not feel draggy.
(There’s also an attractor on the other side of the hill, where having energy makes it easy to do the things that help you maintain energy. But that attractor is less stable, because if anything disrupts any part of the virtuous cycle, the whole thing grinds to a stop.)
My solution to this, in practice, in most cases, is to find workarounds that have minimal activation energy, so that falling into the preferred attractor is easy. But, I’ve also assumed that there are some places where I would just have to bite the bullet and be satisfied with non-responsive systems. That is, you just rigidly make sure to exercise every day, because you know that it supports everything else.
This solution is maybe fine, but it is also pretty fragile.
Leading indicators save the day
Actually though, this is already a solved problem. Living organisms are (or are made of) homeostatic control systems that regulate their own inputs.
An animal needs calories in order to function and it spends calories to control the level of calories it has stored.
And the key thing is that there is a long lag in the system. An animal doesn’t wait until its cells are starved to go hunt. It goes to hunt, or at least goes to the refrigerator, on the basis of a much much earlier indicator, when it is hungry. It gets hungry much earlier than when it is starting to literally run out of calories.
Or thinking somewhat more abstractly: Suppose you have a control system that regulates the input of gasoline, but the control system itself depends on gasoline to function?
If such a system was constructed with very little lag, it would fail with the first sizeable shock. But if the system had look-ahead (perhaps because gas flows through a reservoir that fuels the regulator, even when there is 0 gas flowing through the regulator at the moment).
So it seems that I should be looking out for very early leading indicators of unwanted phenomenological states, and hooking up control systems to those, as opposed to building systems that track the phenomenological states of some input starting to fall apart. I may find that more things can be control systems than I had previously thought.