Are frameworks and models useful?

I have always deeply hated the oft-quoted saying “All models are wrong, but some are useful.” It is not that it is wrong exactly, but to my ear, it has a negative, defeatist tone. (Just like that quote about democracy being the least worst form of government.) I was chatting about this recently with a couple of super smart colleagues, Mike Fang and Roger Williams, in the context of world models for AI. Here are some thoughts for your delectation. Probably only suitable reading material for nerds/ geeks. They are rich with references that make me and this post seem pretentious. It is left as an exercise for the reader to decide how much that is true. 1. It is true that a model is never the actual thing being modelled, it is only ever a representation of the thing. Reference: René Magritte’s picture “The Treachery of Images” which includes the line: “Ceci n’est pas une pipe” (this is not a pipe), written on a picture of a pipe. 2. It is also true that a model is always abstract, where the actual thing being modelled is concrete. Reference: philosopher Georg Hegel described the word abstract as meaning something less than the whole. And a model is always less than the whole. 3. One of the reasons a model of something is useful is that it is simpler to understand and use than the real thing. Even if we could make a model of something that captured the whole thing, it would be the same as the real thing and hence be useless. (Similar to Turing’s Halting Problem, which described the fact you can’t write a program that takes another program as input and tells you whether the program will terminate or not, because the only way to be sure is to run that program, and if it never terminates, it can’t come back and tell you that.) 4. A model is often useful for a particular set of purposes, but useless for other purposes. 5. Some models are crap (for all purposes). 6. Some bad models are more designed to be useful for the analysts/ advisors who create/ use them. Good model are designed to be useful for the owner/ user of the thing being modelled. 7. All models are biased. This was pointed out by Jacques Derrida’s statement that ‘deconstruction is not neutral’. The way we break things down for a model feels neutral, but actually imposes some meaning. Imagine breaking costs down into fixed/ variable vs direct/ indirect. Each reveals different insights. 8. Despite all this, it is admirable to model something to reflect its reality as much as possible, rather than just be useful in one way. This makes models more robust. There is an analogy to writing computer programs that just happen produce the right outputbrather than model the world they are simulating. Object-oriented coding helps a lot with this. Hope you found this interesting, and weren’t too harsh in judgements of its/ my pretentiousness. #unask #unthink #unlearn