Making, from first principles
Building up the poietic idea from things I can’t argue with
poieticism
philosophy
making
ai
In the last post I found a word, poietic, and a long line of Greek philosophers who used it before me. That’s nice, but a word with a good pedigree can still be empty. So I tried the approach I’d use on any engineering problem: forget the history, start from things I can’t really argue with, and see how far they go.
I numbered the steps, so you can tell me exactly which one is wrong.
- Making brings something into the world that wasn’t there before. That’s Diotima’s definition, and it’s much wider than art: a bug fix, a soup, a bookshelf, a startup and a kid’s Lego tractor all count.
- A made thing meets the world, and the world pushes back. A chair gets sat on, code gets unexpected input, a sentence gets misread. Until that happens you have a description of a thing, which is what I tried to say in A render is a promise.
- The push-back changes the maker. You come out knowing something you didn’t know before, most often that you were wrong about something. This is where Lubacz’s point fits, that knowing and making are intertwined in real activity.
- So making has two results: the thing, and a changed maker. We usually count only the first one, because it’s the one you can take a photo of.
- Made things keep needing to be made. They break, the world around them changes, and somebody has to keep adapting them (I wrote about this in Maintenance is a creative act).
- The ability to make is spread unevenly, and tools change how it’s spread. A tool can grow your ability or replace it, and from the outside the two look the same for quite a while.
- Nobody makes alone. Even a solo project runs on tools, materials and knowledge that other people made.
From these I get one question that I can ask about any tool, project or habit: does it leave people more able to make things, or less?
The uncomfortable part
This is where it gets uncomfortable for me, because I build AI tools. A model can now produce the first result, the thing, in seconds. Step 4 says that’s only half of the output. If the model made the thing, who got changed?
Sutton’s The Bitter Lesson gives an answer I didn’t expect. His argument is that the methods which win in the long run are the ones that search and learn. Put that next to step 3 and the learning happens in the system doing the searching. If I hand it all of the making, the system improves at making and I stay exactly where I was.
I’m not going to stop using AI over this. I use it every day and I’m building Graspable on top of it. What it changes is what I expect from a tool. A good one hands some of the push-back to me: it shows me what failed, lets me change the result myself, and leaves me understanding a bit more than before the session. A tool where I only press a button and judge what comes out trains my taste at best, and taste is the esthesic side again.
A name for it
I need a handle for this view, so I’ll call it Poieticism.
Yes, an -ism. I’m suspicious of those too, they tend to come with a manifesto, a logo, and eventually a split into two rival -isms. Before settling on the name I searched for it, and the closest thing the search engine offered was “poeticism”, which Dictionary.com defines as a poetic expression that has become hackneyed, forced or artificial. Good to know what to avoid.
For now, Poieticism is the list above and the question that comes out of it. If one of the steps turns out to be wrong, the whole thing has to change, and I’d like to hear which one it is.