Comments by "EebstertheGreat" (@EebstertheGreat) on "Sherlock Holmes NEVER 'Deduced' Anything" video.

This is potentially an issue in the philosophy of science. In practice, most scientific reasoning appears to follow the logic of the fallacy of affirming the consequent. This fallacy has the form "If A then B, B, therefore A." It is clearly fallacious. For instance, if it floods, then the ground will be wet. This morning the ground is wet. But it did not flood this morning. Because the ground was wet for a different reason, in this case rain. But most scientific reasoning really does seem to work like this. A lot of weight is given to models which produce novel predictions. If those predictions end up according with new observations, we say that the theory has been corroborated. But in what sense? Certainly not the deductive one. The most famous 20th century treatments of this question come from Karl Popper and Thomas Kuhn. Popper claims that assigning epistemic certainty to scientific conclusions is just too much to ask, but that this is still an accurate description of how science works. Popper focuses on falsifiability as the demarcating criterion for science. A scientific theory asserts some conclusion that can be proved false. If you prove it false, then the theory is discarded. (This is not a fallacy but a proper application of modus tollens.) For instance, "if the Newtonian model of gravity is correct, then the perhelion of Mercury will precess only as much as is calculated. It actually precesses 0.43" per year more than that. Therefore the Newtonian model of gravity is not correct." If you fail to prove a theory false, then the theory is corroborated, in the sense that it has survived a possible attack. The theory is probably still wrong, but it has shown itself to be more useful than other theories checked so far. So in that respect, it is a good theory. This is even reflected in the focus on p-values in experiments. I can say "this result is significant with p < 0.01" if and only if the following is true: If the hypothesis is in fact false (i.e. the null hypothesis is true), and I repeat the experiment many times, then I should expect to get such an extreme result less than 1 time in 100. But you can't get rid of that "if the null hypothesis is true" part. It does not say how probable the conclusion is overall. If I conduct an experiment that convincingly shows something very implausible, say that the Moon is made of cheese, that thing is still probably not true; it's more likely that something went wrong with my experiment. The low p-value is still much higher than the prior probability of the Moon being made of cheese. So experiments really only ever test models against each other, not against any sort of ground truth. Thomas Kuhn does not accept Popper's falsificationism. He points out that the way most scientists work in practice does not resemble the idealized version of science Popper presents. Kuhn asserts that the nature of a scientific model depends on the cultural assumptions present when it is being created. In particular, Kuhn claims that science proceeds in conceptual leaps, where an older paradigm is discarded in favor of a newer one. Kuhn says that these "paradigm shifts" are basically non-rational, and that there is no good way to compare paradigms against each other. But within a single paradigm, two theories can be compared on their merits. So to Kuhn, Copernicus's model was correctly rejected in its time, and accepting a heliocentric model required a non-rational paradigm shift to a new kind of thinking. I haven't read much of Kuhn's work, so I might not be explaining his ideas all that well. A lot of people have cited Kuhn as an inspiration who don't represent his opinions very well, so take this witha grain of salt. At any rate, this is a serious philosophical quandary. All our conclusions about the real world are merely models which provide plausible explanations of what we experience. Induction is a problem on its own, but even if you can resolve that, the reliance on models is inescapable. Science doesn't really prove the world is round. It just proves that a model containing a round world survives potential attacks better than any effective model with a flat Earth yet created.
1