Book review: The Knowledge Machine by Michael Strevens

Last spring I read and discussed Michael Strevens’ The Knowledge Machine: How Irrationality Created Modern Science with my lab group. Here’s my review.

tl;dr: a scientist who wants to dip a toe into philosophy of science could do much worse than to start with this book.

This is an unusual book in terms of both subject matter and approach.

In terms of subject matter, Strevens sets out to answer two big questions. First is a philosophical question: what distinguishes science from other human activities, such as pseudoscience or history or art or philosophy or religion or etc.? Second is a historical question: why did the activity we now recognize as science not emerge until the 17th century, in Europe? The subject matter is unusual because it’s unfashionable, even contrarian. Much recent philosophy of science, and history of science, argues that nothing distinguishes science from non-science. That there’s no clear bright line to be drawn between science and non-science, either in (philosophical) principle or in (historical) practice. And if there’s no clear bright line between science and non-science, then there can’t possibly be any Scientific Revolution–a more-or-less sudden invention of science, the date of which cries out for historical explanation.

In terms of approach, the book is unusual because it’s a popular book. If it’s not the first popular treatment of philosophy of science, it must surely be among the few. Further, it’s not a popular review of philosophy of science. Strevens does review some previous work in philosophy of science (primarily Popper and Kuhn), but only in order to set the stage for his own original answers to the two questions listed above. As best I can tell from Strevens’ cv, this book isn’t even a popular summary of his own papers. Rather, he seems to have written a popular book when most others in his shoes would’ve written a specialized scholarly monograph.

What I liked about the book:

  • A very effective baby-werewolf-silver bullet structure. I suspect that a lot of scientists and science students will instinctively think it’s pretty obvious what distinguishes science from non-science. Strevens does a great job of using historical examples to undermine those instincts. He’s quite convincing that there is something distinctive about science, and that we have yet to figure out exactly what it is. Strevens also is quite good at granting every point of the most vocal critics of science (that individual scientists are inevitably biased and subjective, etc.), while still arguing that science works.
  • A clear, interesting answer to his first question. According to Strevens, what distinguishes science from other ways of knowing, and makes it more effective than other ways of knowing, is that science alone follows what he terms the “iron rule”: all disputes are to be settled by empirical data. So if there are different competing scientific claims, well, figure out what experiment or other data would distinguish between those claims, and do whatever it takes to collect those data. Strevens doesn’t think there’s anything universal to be said about exactly how data can be used to settle disputes. In this, he disagrees with Popper and others. It’s not that, say, data always settles disputes via Popperian hypothetico-deductive arguments. And Strevens acknowledges that scientists always bring their own personal subjective biases to the table (while also emphasizing that they should try not to, and that science has professional practices that tamp down those biases). But he thinks that, in the long run, science will make cumulative progress towards the truth as long as scientists do their best to follow the iron rule. Strevens is quite good at emphasizing how counterintuitive the iron rule is. In particular, the iron rules says that scientists can’t appeal to other criteria when deciding among competing scientific claims. That might seem pretty obvious to us today, if the other criteria you have in mind are things like “the Bible” or “the authority of the Pope” or “seances” or whatever. But scientists who are following the iron rule also can’t favor one claim over another on grounds of, say, “parsimony”, or “mathematical rigor”, or “theoretical elegance”, or etc. It’s not at all obvious that appealing only to empirical data really is the best way to settle scientific disputes.
  • Following on from the previous bullet: I found myself thinking about an analogy between the iron rule, and arguments for the efficacy of evolution by natural selection. Arch-selectionists acknowledge the operation of evolutionary forces like genetic drift and migration, that can interfere with the operation of selection. But nevertheless (so the argument goes), selection is a distinctive evolutionary force: it’s the only one that has a tendency to push populations “uphill” on the “adaptive landscape”. So that, even if selection is weak or intermittent, and even if other evolutionary forces are at work, it’s still the case that populations under selection will tend to adapt (in the long run, on average), and that populations not under selection will not. I think Strevens’ “iron rule” is a bit like natural selection. It imparts a cumulative, directional tendency towards truth onto scientific research. Even if the iron rule is only imperfectly obeyed by practicing scientists, even if the relevant data often are difficult to collect, etc. Progress towards the truth might be slow in particular cases, it might even temporarily go into reverse sometimes. But in the long run, on average, science makes progress. In contrast, fields like philosophy or literary criticism or theology don’t obey the iron rule, and so there’s no force nudging those fields towards the truth.
  • Strevens illustrates his points with great historical examples, conveyed in a very readable style. In all seriousness, it’s a page turner. Now, they’re mostly canonical examples in philosophy of science, so they might be familiar to some of you. And it’s fair to wonder if it might be a bit misleading to focus on those examples, because the same features that made them canonical might also make them atypical (more on this below). But the audience for this book is people new to history and philosophy of science. For them, these examples are going to be totally eye-opening and captivating.
  • Strevens has provocative thoughts about how to organize scientific practice, and what values and behaviors to instill in individual scientists. At least, I think they’ll be provocative to many of you, because they’re very old school. For instance, Strevens is very much in favor of the traditional system of peer-reviewed scientific papers and all their conventions (dry language, strictures against speculation, etc.). He thinks it’s best if most scientists don’t pay attention to philosophy or other “ways of knowing”. He does not think that most scientific work involves, or should involve, much in the way of creativity. He thinks it’s good that science rewards scientists who are willing to spend inordinate time and effort on tedious data collection. Etc.
  • I found myself wondering if there are analogues of the iron rule in other fields. For instance, think of the old saying in software engineering, “code wins arguments“.

My criticisms and questions:

  • Strevens sometimes resorts to hypothetical examples rather than actual historical examples to illustrate his ideas. This is a letdown because the historical examples are so well-told, and because some of the hypothetical examples are a bit weird. I couldn’t see any reason why he couldn’t have stuck to historical examples throughout.
  • I think Strevens overgeneralizes from physics, and particularly from his hero Newton. In particular, I disagree with his chapter suggesting that “beauty” might constitute a partial exception to the iron rule. I know lots of physicists these days say that the “beauty” of a theoretical claim is a reliable guide to its truth. But I think Sabine Hossenfelder demolished that idea within physics. And it’s notable that scientists who aren’t physicists hardly ever stump for “beauty” as a reliable, data-independent marker of truth.
  • Strevens’ emphasis on implementing the iron rule via testing the quantitative predictions of phenomenological models also represents an overgeneralization from physics, I think. Yes, you can test Newton’s laws of motion, and quantum mechanics, without understanding why those laws work. But I feel like attempts to develop and test purely phenomenological models have a pretty mixed track record in science as a whole. Strevens is right that testing the predictions of purely phenomenological models is a way to implement the iron rule. But he kind of leaves the impression that it’s the only way, which it isn’t. I left the book fairly convinced that the iron rule is right, but less convinced by Strevens’ elaborations on how to apply the iron rule in practice.
  • I wish I knew more about Isaac Newton, to be in a better position to evaluate Strevens’ comments on Newton. Strevens’ presents Newton as more or less the inventor of modern science, because he invented the iron rule. In Strevens’ telling, Newton was a rare genius capable of compartmentalizing his scientific research from his other interests (his alchemical work, his work in Christian history and theology, etc.). So that Newton consciously followed the iron rule in his scientific work, while consciously ignoring the iron rule in his non-scientific work. I’m just instinctively suspicious that Newton himself saw a clear bright line between his scientific work in the Principia and his work in alchemy or whatever.
  • Strevens’ answer to his second big question–why science emerged relatively suddenly in 17th century Europe–isn’t very convincing, though I don’t really know enough history to judge. It’s basically just a list of favorable preconditions to adoption of the iron rule that were operative in Europe in the 17th century, plus the claim that Isaac Newton was a genius. There’s no discussion of why science didn’t emerge earlier in other places where some or all of those conditions held. And there’s no discussion of whether science could’ve emerged under some other set of conditions. As an exercise in counterfactual history, it’s too brief and superficial to be very convincing. (Click that last link for a more convincing exercise in counterfactual history of science.) But in fairness, most of the book is about Strevens’ first question–what distinguishes science from non-science. You can definitely read the book as an answer to the first question and not worry about the answer to the second question.

Bottom line: Brian and I have previously recommended some “philosophy of science 101” readings for ecologists, but having read Strevens’ book I think I’d recommend it instead. Strevens’ book is more fun to read and accessible than intro-level philosophy of science textbooks, such as Peter Godfrey-Smith’s Theory and Reality. Though it kind of depends what you want. Do you want the sort of piecemeal, haphazard introduction to philosophy of science that you’d get from reading philosophy of science papers about some bit of ecology? Or do you want a systemic, “big picture” introduction to philosophy of science that doesn’t mention ecology at all?

Here are links to other reviews from Science, the NYTimes, and the Guardian. And here’s a video of Strevens discussing his book.

15 thoughts on “Book review: The Knowledge Machine by Michael Strevens

    • Yes, thanks for the reminder about that piece, it’s interesting. Steven Frank has written a bit about this too, IIRC.

      I didn’t have anything quite that specific in mind when I made my own analogy. Strevens’ iron rule isn’t the same thing as Bayesian updating.

  1. At the end of the review in SCIENCE:
    ” While Strevens’s iron rule may indeed be the foundation of modern science’s success, the methods scientists use to come up with new ideas remain elusive.”
    Any comment?

    • Agree. Strevens doesn’t say anything about how to come up with scientific claims to test.

      Many years ago, I think philosophers of science took the view that there was nothing to be said here. That hypothesis development was just an inscrutable creative process. Nowadays, I think the standard view is that hypothesis development does have a logic to it, that it’s not a matter of just waiting around for random inspiration to strike. But I don’t know anything about the literature on this.

  2. I wish you’d written this a few months ago! Françoise Cardou and I read and discussed this then, so thoughts would have been fresh. Overall, agreement on several points: (i) the iron-rule part is stronger than the part explaining why science arose at a particular moment; (ii) the iron-rule part is very strong – basically anything goes as long as you rely on empirical evidence to support claims; this really seems to capture how science is actually done and what distinguishes it; (iii) the book is as readable and compelling as anything else you might find on the topic. Really enjoyable book.

    In terms of the previous commenter (anon), Strevens does say that anything and everything (religion, beauty, whatever) can be involved prior to actually deciding what hypothesis is or is not supported. Whatever beliefs or sources of inspiration or biases one has, it’s all fine when developing ideas – you just need set those aside when assessing claims about support from data. This was one area where we thought he overstated the hard line separating peer-reviewed publications from the rest. As best I can tell, there’s plenty of leakage from people’s biases into their peer-reviewed data papers.

    I don’t recall any implication that testing predictions of phenomenological models was the one best way to do things – I remember more the “anything goes as long as you use empirical data” gist.

    • “I wish you’d written this a few months ago! ”

      I too wish I’d written this a few months ago. 🙂 😦

      “the hard line separating peer-reviewed publications from the rest. ”

      Glad you brought that up, I forgot to put it in the post. Still mulling over what I think about that hard line Strevens draws between “official” science (peer-reviewed papers) and everything else (seminars, blog posts, tweets, pop sci books, one-on-one conversations, etc.) On the one hand, it certainly is true that there a bunch of conventions and enforceable rules associated with peer-reviewed papers to ensure (as far as possible) that they obey the iron rule. There aren’t any such conventions and enforceable rules for seminars, blog posts, tweets, conversations, etc. So yeah, you definitely can make the case that there’s an important distinction between peer-reviewed papers and other scientific outputs (even though, as you note, the conventions and rules associated with peer-reviewed papers are imperfectly obeyed and enforced). On the other hand, plenty of scientists do put a lot of effort into, and seem to be influenced by, seminars, blog posts, tweets, conversations, etc. It seems unsatisfying to just set all that stuff aside by saying it doesn’t count as proper “science”; only peer-reviewed papers really count. Surely there’s more that could be said about the relationship between peer-reviewed papers and other scientific outputs? I think Strevens kind of skirts the edge of committing the “no true Scotsman” fallacy when he carves off peer-reviewed papers from other scientific outputs (

  3. Sounds like a really interesting book. And sounds like pairing reading this with reading The Golem which largely takes the opposite point that there is no precise scientific method would make for a really great undergraduate philosophy of science course.

    It sounds like the first project makes a lot of sense (and I would agree with). But how does his 2nd project (why science started in Europe with Newton) not come across as naively simplistic and Eurocentric? Just within Europe, Bacon and Gallileo are both more strongly tied to empiricism in my mind than Newton (and both a tad earlier). But there was some pretty flourishing science in China and the Turkish empire several hundred years before Newton – I don’t have a deep knowledge of those (and they certainly had some astrology mixed in just like Newton), but I think there are elements of the Iron Rule in those traditions.

    • Re: undergraduates as an audience for Strevens’ book, my lab reading group included a strong final year undergrad student, who found the book accessible and very engaging. But I heard from an ecologist colleague at another uni who read the book with his lab group; he said his undergrads found it rough sledding. So YMMV.

      Strevens does talk about Bacon. Galileo doesn’t really come up.

      Re: China, the Turkish empire, and other places and times other than 17th century Europe: yeah, Strevens doesn’t really do any comparative historiography. I don’t really know enough myself about non-European scientific traditions to fill in the gaps myself.

      Strevens emphasizes that following the iron rule means *not* appealing to any other source of information or authority. You can’t appeal to scripture, or tradition, or some authority like Aristotle, or whatever. So I think Strevens might say that the iron rule traces to 17th century Europe because it’s a product of (and also helped produce) the Enlightenment. A product of (and cause of) widespread questioning of religious authorities and ancient authorities.

      Even if you grant that, it’s arguably still a problem for Strevens’ answers to both his questions that people in other places and times learned a lot of science that still holds up today. After all, how did people do a lot of still-valid science at those other places and times if they weren’t following the iron rule? I can imagine various ways of answering that question. One is that following the iron rule isn’t an all-or-nothing thing; it’s a gradient. You can follow the rule more or less well. So possibly, Strevens could grant that in China, the Turkish empire, and other places, there were people who did some still-valid science by following the iron rule imperfectly. He could then further argue that it’s in 17th century Europe that people first learned to apply the iron rule much more systematically and effectively (albeit still–always–imperfectly).

      • According to this UNESCO piece, the Arab science revolution 500 years before Newton were the real roots of the scientific method with empiricism and hypothesis testing ( – that piece also attributes the European origin to Galileo rather than Newton. I’m certainly not a scholar of the topic, but I think all scholars these days know enough to be suspicious of European exceptionalism as rarely being true. I’m surprised Strevens tried it.

        Personally, I wonder if banning appeals to authority and allowing any method of convincing not involving appeals to authority (parsimony, elegance, data) is not a closer description of the real method used by scientists. But I’m fine with cold hard empiricism too. I stopped calling myself a theoretician and started calling my self an empiricist early in my career.

      • “No appeals to authority” might well be a closer description of actual scientific practice than the iron rule. (Hard to say for sure, obviously; this isn’t the kind of thing you can really quantify…)

        Semi-related thought: I’ve been thinking about why the iron rule works. Is it just that the data is exogenous to the scientists? It’s an independent, exogenous input into the process of evaluating scientific claims. Ok, not *totally* exogenous. Scientists certainly can do (legitimate) things that affect what data they’ll see (choice of sample size, choice of study design, etc.). And they can and do bring their own subjective opinions to the table when deciding, e.g., how to interpret the data, or what data to dismiss as “outliers”, or whatever. But it gets increasingly hard to protect your subjective opinions in the face of more and more exogenous inputs. A bit like how, if you think you’re the best tennis player in the world, but you keep losing matches, well, at some point it becomes impossible to maintain your illusion that you’re the world’s best player. Because the outcome of a match is an exogenous input to your process of evaluating yourself as a tennis player. Yes, you certainly can affect the outcome of a match through your own play. And yes, there certainly are all sorts of factors that affect the outcomes of your matches, besides how good you are compared to your opponents (did the umpire make a bad call? was the crowd against you? etc.) But you don’t have *infinite* ability to explain away a sufficiently large amount of exogenous inputs. At best, you might manage to maintain your own illusions–but nobody else is going to share them.

        That’s why I’m suspicious of “beauty” as a possible partial exception to the iron rule. “Beauty” is in the eye of the beholder, it’s not an exogenous input. to the process of evaluating scientific claims. As illustrated by the fact that physicists’ and astronomers’ own sense of what’s a “beautiful” model has changed over time. Time was when perfectly circular orbits were considered “beautiful”. Same goes for appeals to “elegance” or “parsimony”, or how “interesting” a hypothesis is.

        To be clear, I do think scientists *should* value elegance, interesting-ness, etc. But they should value them because they help us decide what questions to ask, what lines of research to pursue, etc. We shouldn’t value them because they’re markers of truth.

  4. Naomi Oreskes provides a decent and fairly readable (to the interested, in any case) historical view of philosophy of science in her book Why Trust Science, and comes to some different interpretations and conclusions than what you describe in Strevens. I’m looking forward to giving it a read at some point, so thanks for the recommendation!

  5. Pingback: Friday links: science is “interesting” but not “awesome”, evolution of peer review, and more | Dynamic Ecology

  6. Sciencetific iron rule is pretty obvious at its practice. Follow the data, follow the outcome. Adjust following the 2. This is obvious and universally ( FOLLOWED). Nothing unusual about it. I think even animals follow this in their own way. Thr modern scientific method is a more conscious application of the principle. Why science snowballed starting from early modern Europe? Part of the story is Europe centrists and fact we ( as in modern world at this moment) are a direct and immediate heirs to it. And partly it is a critical mass of revival of lost Greek Roman thought ( with inputs from Indian thought via Arab world ) , developments of mathematics, testing equipments, a wider set of educated people, printing press etc.

Leave a Comment

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.