Salvaging POSIWID

Is it ever reasonable to say the Purpose Of a System is What It Does?

Scott Alexander just put out the post Come On, Obviously The Purpose Of A System Is Not What It Does about the titular idiom / saying / adage / whatever - or as he abbreviates it, POSIWID.

One of the ways he pushes back on this is by defining:

System: New York bus system
What It Does: emit four billion tons of carbon dioxide
Purpose QED: emitting four billion tons of carbon dioxide

Now I agree that some people are just doing the facile thing, and would only really deploy POSIWID to criticise a system they already didn't like, and so would claim that the purpose of the bus system is the carbon dioxide. (If they were weirdly both against public transport and pro environmentalism... but maybe they exist.)

I agree that that seems like a silly thing to say, but the question I'm interested in investigating is "can POSIWID be salvaged for intelligent debate". That might also be a bad idea if it creates a motte-and-bailey for people to use, but oh well.

I propose the key here is "what it does". For simplicity, any time I say POSIWID from here on out, I mean "the good version of POSIWID".

People state purposes to get things off the ground - but not always the real ones, or all of them.

Scott says the "purpose of the New York bus system is to transport New Yorkers. The carbon emissions are an unintended side effect". Honestly? Probably true. But it's not necessarily the actual purpose the original architects had in mind, or in their hearts. (This has become a much more important question for, e.g., American constitutional law.)

For instance, there might be a "bootlegger and baptist" situation - maybe there were lots of people saying building the bus system is a good idea because it will transport New Yorkers (the baptists), but maybe there were also people pushing for it to be built to create opportunities for mafioso graft (the bootleggers).

So how do we weigh evidence against these hypotheses?

I think some systems seem pretty unambiguous: either ones that are very flagrantly bad, or are very simple (good or bad). Some

But many - maybe most - systems are... somewhere in the middle. What the New York bus system does is {convoluted bureaucratic flowchart}, and one of the downstream effects of that is transporting a whole lot of New Yorkers - and emitting four billion tons of carbon dioxide. It's not super clear to me that we should prefer the "transporting" to the "emitting" as the action we point to as conclusive evidence!

I think what POSIWID is really getting at requires you to identify the key action the "system" takes (and we might circle back to defining the system in the first place). So in this case we should point to the {convoluted bureaucratic flowchart}. "It" employs a bunch of people to make a lot of promises and do a lot of paperwork and employ other people and etc. etc.

Big punchline: feedback loops!

Basically all systems have feedback loops. Some are deliberately built in, some are emergent, and some are maybe even only tangentially related - imagine some system on a computer in a basement somewhere with no outputs exposed to the rest of the world; it's consuming electricity, and with nothing to show for it, it would seem that sooner or later it will be turned off. A feedback loop: time elapses, people update on evidence, evaluate their utility functions, act "on" the system.

Many human systems have self-reinforcing feedback loops, through a combination of survivorship and selection biases (maybe others; I think those are the main/right ones). Any system that blows itself up or doesn't do things that cause it to reproduce doesn't make it more than a "generation", so on average a system picked at random will have the property of "propagates over time, people, and/or resources".

Feedback loops seem like a very meaningful way to reason about and/or label the purpose of a system. Over time, I claim that many - if not all! - systems end up doing a lot of "propagating themselves", and I think this is a really compelling way to talk about their purpose! The purpose of a lot of components within the New York bus system is to make sure there's a bus system tomorrow.

So: purpose?

The purpose might be transparent, and it might not be nefarious, and we might even like it.

But sometimes this isn't true: some news story comes out about the bus system in its current form spending resources or taking steps (via its human (for now) components, of course) to continue to exist in its current form. And it turns out this actually traded off against transporting New Yorkers (and emitting carbon). And we don't like that, and someone points and says:

POSIWID.

And I think this is precisely the right case for POSIWID.

Quick-fire questions

Q: I think capitalism is doing this self-replicating thing at the cost of human wellbeing. So should we burn it all down?

A: No, I don't think in-and-of-itself POSIWID is reason to burn anything down. Maybe it's necessary but not sufficient or something - I think this is dominated by the facts of the matter, like how much of the byproducts we like, relative to the ones we don't, get created by the system/purpose.

Q: So does the purpose have to be baked in at the start for this to all hold?

A: This could even happen some time after the bus system is initially stood up. The original politicians all had totally pure intentions; it had that as its core purpose in all relevant senses for a number of generations; but eventually the graft crept in. I think it's basically reasonable to point to the current system and say

• "current POSIWI currently D"

• "previous POSIWI previously D, and this changed"

• "it could have been reasoned out that the system would, in the limit / at equilibrium, change purpose in this way"

without saying

• "and therefore the true P is and always was current POSIWI currently D"