Peter Singer’s 1974 article in the New York Review of Books, Discovering Karl Popper is extremely favorable of Popper’s philosophy of science–except for three paragraphs in the middle, which are highly informed criticism. I’ve reproduced them below along with some limited comments in light of that criticism.
If I did not assume that because water has come out of my tap in the past when I turned the handle the same will happen today. I might equally sensibly hold my glass under the electric light. On this pragmatic issue Popper’s more recent contributions do have a little more to say, but it does not help. He says that, as a basis for action, we should prefer “the best-tested theory.” This can only mean the theory that has survived refutation in the past; but why, since Popper says that past corroboration has nothing to do with future performance, is it rational to prefer this? Popper says that it will be “rational” to do so “in the most obvious sense of the word known to me…. I do not know of anything more ‘rational’ than a well-conducted critical discussion.”
The reader familiar with Popper’s contempt for linguistic philosophy will rub his eyes at this. Popper has picked up that once trusty but now discarded weapon of linguistic philosophers, the argument from a “paradigm usage” of a word—in this case, the word “rational.” The argument proves nothing. As Popper himself has said many times, words do not matter so long as we are not misled by them. Popper’s argument is no better than Strawson’s claim that induction is valid because inductive reasoning is a paradigm of what we mean by “valid” reasoning. In fact Popper’s identification of a “well-conducted critical discussion” with the idea of rationality is doubly unhelpful, since until we know how to establish which theory is more likely to hold in the future we have not the faintest idea how to conduct a “well-conducted critical discussion” that has any bearing on the question we want answered.
More fundamental still is the question how, even in theory, we can possibly prefer one hypothesis to another, or take one as a nearer approximation to truth than the other, if past corroboration has no implications for the future. Without the inductive assumption, the fact that a theory was refuted yesterday is quite irrelevant to its truth-status today. Indeed, in the time it takes to say: “This result corroborates Einstein’s theory but not Newton’s,” all the significance of the remark vanishes, and we cannot go on to say that therefore Einstein’s theory is nearer to the truth. So jettisoning the inductive assumption makes nonsense of Popper’s own theory of the growth of scientific knowledge. While it is true that on Popper’s view induction is not a means of scientific discovery, as it was for Bacon. It remains indispensable, and the logical problem of induction is no nearer to solution than it was before Popper tackled it.
My thoughts are that Singer is half-right: preference for ‘best-tested theories’ is a mistake of Popper’s; instead, we ought to reject only theories that are incompatible with the conclusions drawn from a critical discussion. In other words, ‘critical discussion’ is an activity that occurs during the ‘context of justification,’ but rather than attempt to justify our theories, it is an intersubjective attempt to find incoherence in our knowledge, for if we accept contradictions, we become untethered.
This is only half the equation, though, for it is possible to have a truly coherent position by refining a hypothesis to rule out counterexamples. Without methodological rules, any coherent system is rational. Methodological rules are, so I conjecture, in order: If a theory is incompatible with background knowledge or observation statements, and we find no present fault with our background knowledge or observation reports, it is summarily disqualified, and can only be readmitted once we should learn that the background knowledge or basic statement does not correspond to the facts. Of course, the problem is that ‘learning’ is not justified in any way.
What, then, is ‘learning’? It is little more than the process of trial and error on the communal level: if an individual thinks a piece of background knowledge or an observation statement is false, they can do their best to show them wrong by demonstrating its self-incoherence or incompatibility with the rest of the system (trials), and if no objections to this criticism can be found (other than ad hoc adjustments to our definitions), then we learn that we are wrong (error). And yet none of these decisions are grounded.