Friday, October 18, 2013

Revisiting Kuhn

I am teaching a course on scientific revolutions this term, and so I've been rereading Structure. I've also been blogging a bit about it. I previously wrote about my annoyance that the new edition has different page numbers and some reflections on recent writing about Kuhn. This post is about a thread that runs through Kuhn's discussion which, I think, gets something importantly right.

[cross-posted at Footnotes on Epicycles]

On Wednesday, we talked about Kuhn's claim that different paradigms are incommensurable, and today we talked about the considerations which might convince scientists to shift from the old paradigm to a new one. Kuhn characterizes the shift as a conversion experience, but not one that is totally unmotivated by reasons. Kuhn reviews a whole range of possible reasons, including puzzle-solving power, precision, novel prediction, and simplicity.

He insists that none of these reasons are necessarily decisive, however. He writes that
paradigm debates are not really about relative problem-solving ability.... Instead, the issue is which paradigm should in the future guide research on problems many of which neither competitor can yet claim to resolve completely. A decision between alternate ways of practicing science is called for, and in the circumstances that decision must be based less on past achievement than on future promise. (p. 156)
Because a paradigm serves to guide normal science, accepting a paradigm means committing to do normal science in that way. So the choice is forward-looking, while all of the reasons are backward-looking. So, one might say, the choice of paradigm is strategic rather than simply evidential.

While discussing this passage, I realized that the conclusion does not rely on incommensurability at all. Rather, it just relies on the problem of induction: Past performance of a paradigm provides no guarantee of future results. So proceeding with one paradigm rather than the other is a kind of gamble. Reasonable people with different hunches or different tolerance for risk might disagree about which way to go.

This allows for a philosophically conservative reading of Kuhn which accepts that revolution is different than normal science, because different paradigms would guide scientific practice in substantially different ways. The conservative reading also accepts that the choice between paradigms cannot be determined by the relevant reasons, especially during the period of crisis.

The conservative reading isn't adequate as a reading of Kuhn, because it accepts those things without any appeal to incommensurability. The change between paradigms might be like a conversion experience, as Kuhn would have it, because some strategic choices are; consider choosing a career, choosing to get married, or choosing whether or not to have children. But it might instead be a self-conscious choice, like choosing between mutual funds for your retirement account.

I think that this recommends the conservative reading as a philosophical position, even if it disqualifies it as a reading of Kuhn. The description of normal science and crisis is the really insightful part of Structure, while the stuff about incommensurability is the most problematic.

It occurs to me that what I've called here the conservative reading of Kuhn, in which underdetermination comes from the problem of induction rather than incommensurability, looks a lot like Lakatos' Methodology of Scientific Research Programmes. We're doing Lakatos next week in class, so I'll see if that idea holds up.


  1. Interesting points -- the Lakatos point in particular strikes me as insightful and helpful.

    I wanted to follow up on one bit: "the conclusion does not rely on incommensurability at all." I wonder about this. If there is no incommensurability at all between the paradigms, then it seems like there would be one paradigm that is better supported by the (limited) evidence available to the community (or it could be an objective tie). Put in your framework of gambles: given a choice between two bets, very often one is unequivocally better than another, i.e. there is a unique rational action. (Lots of philosophers characterize rationality in terms of what bets to accept.) Incommensurability disrupts many of these cases of unique rational choices.

    The above is probably too abstract; let me put it in terms of a concrete analogy. If the goal is simply to maximize profits, then for any bet, we can say which side of the bet is more rational to choose (or that they are equally rational to choose). However, if it is not settled/given that maximizing profits is the only goal -- perhaps other purported goals include maximizing free time, or maximizing chocolate consumption -- then typically there will be no unique/ univocal rational answer to 'Which side of the bet should I choose?' (Just to spell out what I hope is obvious, different goals here are supposed to be analogous to different paradigms.)

    1. Greg: Given the goal of maximizing future profits, facts about past returns can never be entirely decisive. So there are periods and cases where there isn't a unique rational choice.
      The same holds for maximizing future puzzle-solving based on a history of prior puzzle-solving. When there one paradigm is in crisis and the other isn't fully developed yet, the evidence just isn't sufficient to settle which one would be the best guide going forward.

  2. "Given the goal of maximizing future profits, facts about past returns can never be entirely decisive. So there are periods and cases where there isn't a unique rational choice." But a bet can be rational or irrational even if past evidence isn't "entirely decisive" or "sufficient to settle" which side of the bet will be more successful. The rational bettor doesn't maximize future profits; rather she maximizes expected future profits.

    Maybe you meant the problem of induction point more seriously than I realized before? But that seems (to me) to prove too much: no bets about future events are rational if we cannot rationally infer anything about the future from the past...

  3. In the original post, I was thinking of the problem of induction in the way you suggest. It would mean that commitment to a paradigm would always involve faith or a gamble, even outside of crises. I suspect that this fits Lakatos, who inherits a distrust of any positive confirmation from Popper.

    In replying to your comment, I was allowing that inductive conclusions are sometimes rationally compelling. It might suffice for the conservative reading of Kuhn for there to be underdetermination in times of crisis and revolution. Although Bayesians might say that there are expected utilities even then, I don't see any prospect of them being shared and objective.

  4. Christian Strasser and I wrote an article exactly on this topic: Kuhn and the question of pursuit worthiness. Here is a link to it:
    (a draft version is available also at

  5. Thanks for the pointer. I'll take a look.

  6. I think the only reason why you can avoid incommensurability is because you misinterpret the meaning of the problem of induction in Kuhn's framework as leading only to "risk" but not to "uncertainty". What Kuhn is referring to when talking about the future is not its associated "risks" but its genuine "uncertainty" because not only future outcomes are unknown but also the standards to evaluate them. If paradigm choice would be an instance of choice under "risk", then paradigm choice would be as you put it "like choosing between mutual funds for your retirement account." But at the heart of Kuhn's conception of science is the idea that the standards for science are not given but themselves subject to scientific investigation and therefore as it were coevolve with the very theories they regulate. As a consequence it is only after all knowledge about the world has been gathered that we will know what the right standards for their evaluation should have been: "though the experience of scientists provides no philosophical justification for the values they deploy (such justification would solve the problem of induction) those values are in part learned from that experience, and they evolve with it." (Kuhn 1977, 335) In other words, it is only at the end of knowledge that paradigm choice will be reduced to choice under "risk", and until then their choice is made under "uncertainty".
    The problem with paradigms is that what is at issue is not the solutions to puzzles, but what those puzzles should be and what their solution should look like. Paradigm choice then depends not on a comparison of their (expected) utility, but on what counts as utility. Instead of maximizing (expected) utility, scientists will sometimes simply reconceive of what counts as utility. What standard to use to compare previous performance of paradigms itself depends on future results. "the choice [between competing paradigms] is not and cannot be determined merely by the evaluative procedures characteristic of normal science, for these depend in part upon a particular paradigm, and that paradigm is at issue." (Kuhn 1970, 94) There is then genuine "uncertainty" about the future because even the underlying probability distribution is unknown. It is because of this uncertainty that Kuhn can't propose an algorithm for theory choice but needs to resort to heuristics instead. It is well-established that algorithms don't work under uncertainty, but that successful action in such circumstances is still possible by using heuristics. Whereas algorithms provide solutions, heuristic only specify how to look for a solution. Similarly, Kuhn characterized scientific values as criteria that function not as rules but as "Criteria that influence decisions without specifying what those decisions must be" (1977, 330).

    Ultimately I think whether one believes induction to lead to risk vs. uncertainty hinges on whether or not the standards of science can be established independently of that scientific activity itself. At the heart of Kuhn's account is the connection between the two, this "feedback loop through which theory change affects the values which led to that change" (Kuhn 1977, 336) without which there could not be any scientific revolutions. But if both are independent, then the problem of induction would not cause uncertainty (produced by the combination of unknown outcomes measures by unknown standards) but only two independent risks, namely risk about outcomes and risk about standards. In sum, I think it's not possible to reach your conclusion without arriving at an entirely different account of science than the one Kuhn proposed.


Note: Only a member of this blog may post a comment.