Notes

In “Against Crucial Experiments,” Duhem argues that we can never truly test a theory or a hypothesis in isolation from other ones. Later, Quine put this more broadly in his “Two Dogmas of Empiricism” and presented an argument that whenever we try to confirm or infirm a statement, we are testing a whole body of all knowledge. Their main reason for this thesis is that any hypothesis relies on outside knowledge that is not necessarily true. In this paper, I argue that they are only partially correct in their view because if a test of a new hypothesis fails, it’s much more likely that it’s this new hypothesis that is incorrect than some part of our previous knowledge.
Both Quine and Duhem present a similar reason to believe that we can never identify whether something is true or false in isolation. Duhem explains this in very specific terms by stating that in order to prove a hypothesis wrong by an experiment, “it would be necessary to enumerate completely all various hypotheses which may cover a determinate group of phenomena; but the physicist is never sure he has exhausted all the imaginable assumptions.” In other words, in order to be sure that an experiment failed because of some hypothesis, we need to test all other pieces of knowledge that are assumed to be true. But testing any such piece of knowledge relies on some other knowledge that we cannot be completely sure about. For example, even something as fundamental as our understanding of time has undergone a revision: time does not pass at a constant rate, as it was assumed before the 20th century, but can speed up or slow down depending on other variables such as mass or velocity.
This argument is very convincing because there were many such occurrences in the history of science when a theory we held for granted and assumed to be true turned out to be wrong. Moreover, this is a very important part of the nature of scientific progress by which we come up with better theories and get rid of the worse ones. Therefore, denying this argument is to forget the history of science. As Duhem warns, we have to be prepared to revise an old theory that is universally accepted if warranted by experimental contradictions instead of constantly shifting the blame to the new theories.
However convincing this thesis is, the real issue comes when Duhem and Quine present their ultimate conclusion that it is impossible to test new ideas in isolation. I agree that it’s ultimately impossible, but in philosophy of science, do we only need to care about the question of possibility? I would argue that we should be more concerned with the question of likelihood because it’s likelihood that is useful for science, not possibility. Firstly, I will show how the question likelihood in science is more important than that of possibility. And secondly, I explain why if an experiment of a hypothesis fails, it’s extremely likely that it’s this hypothesis that really failed and not some other part of related theories.
The reason why should we be concerned with likelihood more than with possibility is that in science there are many theories that would make it completely useless if viewed only through a lens of “what is possible.” For example, according to quantum mechanics, it is perfectly possible that completely unimaginable events can occur, such as going through walls or suddenly waking up on Pluto. However, nobody uses this fact as a guiding principle because even though it’s possible, it’s very improbable. If we did, then we wouldn’t be able to make any useful predictions; this is why we use probabilities. We always try to quantify the degrees of possibility, which is exactly what likelihood is.
One might respond that in order to be confident when building on top of the old theories, we need complete certainty. That is, we can’t rely just on degrees of probabilities. After all, if some theory among the whole body of our knowledge has a non-negative probability of being incorrect, then one day, it is almost certain that we will prove or disprove a theory based on wrong assumptions. However, this line of argumentation does not solve the problem. The reason is that even if some theory in our body of knowledge is wrong, there is no better way to at least try to find out which one than to continually build on top of existing theories and test them (even if we can’t be 100% sure that we are really testing them). The other option is to be afraid of proposing new ideas on top of old ones and just cycle through the old theories and try to falsify them. And this option does not escape this problem of attribution of an experiment’s failure. Therefore, we cannot disregard degrees of probabilities as it is our only