Look, elegance and beauty can be used at the search stage, when looking for a better model describing reality: - Can I make this equation simpler?
- Can I make this model more general?
- Can I make the formulas easier to deal with (and so the theory more useful)?
These are all valid search paths when developing the theory. But the ultimate test is how well the theory matches the experiment.In fact, programmers know this very well. We avoid spaghetti or giant-ball-of-mud code, we try to express our ideas clearly so we can be more certain that the program works as expected, and it does what it's supposed to do. We use elegance and beauty as a guide when writing and improving out code. However, the ultimate test is whether the program works as expected or not. No matter how elegant a wrong program is, it is still wrong. Wrong but elegant code can be an intermediate step to the final correct code, but it's not the final goal. The same with the scientific method. You can guide your research with elegance in mind. (At the end of the day, simpler more elegant models are nicer to work with when developing your theory and when applying it in practice!) But the ultimate test is the experiment. And also, if your theory is elegant but untestable (unfalsifiable), the theory is not scientific. It might be completely wrong. As useful as it is, even as a search heuristic elegance might be misleading. Prof. Sabine Hossenfelder ( https://backreaction.blogspot.com/ ) wrote a book on how the search for beauty and mathematical elegance led modern fundamental Physics to its currently rather unfruitful state. |
As for the underlying issue, I do think it points to something important, but only at certain stages, as you say.
A model inconsistent with empirical data is not likely to gain traction in the absence of further data. However, there are times when a set of models are equally consistent with the data, and it's not even entirely clear how to empirically distinguish between them. I think in those cases, aesthetic considerations do often come into play. Also, when the evidence slightly favors one model over others, and is also more aesthetically appealing, it tends to amplify perceptions of that evidence even further, in a way that's not logically warranted.
The deeper question is what does it mean for a theory to be aesthetically beautiful? Lots of times, as you're pointing to, it means the theory is simpler or more parsimonious, or something along those lines. I don't actually think that's negligible as a consideration, as it probably means the theory is more powerful, in an information-theoretic sense. If you have two theories, and both match the data well, but one is very baroque and convoluted, and the other is extremely simple, it could be that the convoluted one is correct, but the simpler one is doing so with fewer contortions, it has more degrees of freedom.
I also agree there's a deeper issue about meta-science, along the lines of Feyerabend, that was missed as well in the article. Scientists tend to be blind to their own role in things, as if they're neutral perfectly objective entities. Although this is a good ideal, it never works out in practice (or almost never so), and ignoring it often causes more problems than it solves. I think what Feyerabend was pointing to in part is a lack of meta-scientific evidence for anything in particular, and an awareness that "all models are wrong" applies at the meta-scientific level as well.