The Use of Knowledge in Society (1945)
econlib.org
econlib.org
It's like when the curse of dimentionality meet big O notation, and you throw your hands in the air and give up - or satisfy yourself with simplest solutions, while knowing their limitations.
Note how this applies to all abstractions. They are generally helpful, but as the saying goes, the maps are not the territories.
Except in economics, economist try their hardest to ignore these limitations, even if they go against basic logic. It's hard for a man to understand something when his continuous employment is based on not understanding.
But a minimum wage will increase unemployment. There is no workaround. A price ceiling will cause shortages. There is no workaround. People get elected promising there are workaround, and people get paid by promising to deliver workaround, but it's just smoke and mirrors.
I find it funny when I see on TV they are now going to allow some compensation for organ donors to try to fix the organ shortage problem. Using insights from this paper from 1945 and basic econ, you can already say that it's an increase in the price ceiling, so it will improve shortages.
But there will always be shortages unless a market can exist. People who fight against basic logic will just have a little less blood on their hands. But if they can sleep at night and help politicians deliver fuzzy feeling to their electorate, I guess that's good enough to deserve their wage.
After all, people just want to be reassured that the dirty hands of capitalism won't be touching the beautiful world of love and kindness that medicine is, especially regarding organ donation (cue emotional music with people cuddling)
Or that’s what everyone believes. Empirically that doesn’t seem true. Just because your loss goes down, quality doesn’t necessarily go up. It’s a statistical correlation, but it’s only a loose correlation at best.
So it’s kind of doubly related here: not only is the knowledge a statistical summary, which hides the details of what’s happening, but we also don’t know a better way to do it. No one knows how to measure quality other than to show some samples to a bunch of humans, or to use the incorrect loss function.
I suppose the point is that it’s very easy to accidentally hide such complexities using math. No one intentionally sat down and said “I’m going to come up with a very imprecise loss function that gives a misleading view of how well AI models are doing.” It just happened by default. And it seems like that might be related to your point of economists ignoring limitations, for example. The limitations are inherent in the statistical summary.