Epistemic status: exploratory

I read a lot of books. And during my current training to improve as an epistemologist, I ended up reading even more than I usually do, on topics and ideas that I expect could interest you. But since a real book review takes too much energy, I’m just writing a bit on each book, shotgun style.

Last week I focused on general philosophy of science books instead of on more concrete analyses of special sciences. I (re)read Against Method by Paul Feyerabend, a classic of critical philosophy of science; The Knowledge Machine by Michael Strevens, a modern future-classic about the methodologies of science; and Understanding Nature: Case Studies in Comparative Epistemology by Hub Zwart, a weird book comparing the approaches of science and literature to understanding life.

Against Method, Paul Feyerabend

Feyerabend is legendary for his epistemological anarchism and proclaiming that in science, “anything goes”. This is the book where he introduces and defends all that.

Many of the comments, criticism and mentions I see of Feyerabend call him an epistemic relativist who denies the special position of science in terms of knowledge. And to some extent he deserves it, because he really likes to push for anarchy and liberality of perspectives. But what he actually accomplishes in Against Method IMO is something far more palatable to LW readers: pointing out the failures of Traditional Rationality (and Popper — he really likes to diss on Popper). As the title implies, Feyerabend defends the thesis that any methodology in Science is too restrictive for doing what scientists claim to do. Which means that at some point, progress will be made by breaking the rule instead of following it blindly.

Feyerabend calls this “opportunism”, and argues that the great minds in science (he names Einstein and Bohr but mostly uses Galileo as an example) are the ones who don’t limit themselves in such a way.

At this point I need to deal with the frustrating part: Feyerabend is really, as far as I can see, an epistemological anarchist. He doesn’t think that science has a special place in terms of knowledge production and perspective, even though he thinks it has a place (and in a later edition explicitly pushes for more of it against the rise of political correctness and postmodernism). The main consequence is that he likes to jump to “Anything Goes” when there’s no reason to. His analysis of how science works is incredibly deep and subtle, and then, when the time comes from correcting the prescriptive doctrine, he just jumps to “anything goes and we should let people go into whichever direction they fancy”. Similarly, he argues from the perspective of traditional rationality not because he believes in its incredible power, but as an undercover agent:

An anarchist is like an undercover agent who plays the game of Reason in order to undercut the authority of Reason (Truth, Honesty, Justice, and so on).

Yet because he is playing the undercover agent, he’s arguing that Traditional Rationality is incoherent with scientific progress as commonly understood. So we don’t need to share his anarchism to take something out of his analysis. Because the jump to anarchy only happens in the last step, all the rest still stands without it. And what’s left is a great argument about the possibility and even the need to use counterinduction in science: to propose and keep refining theories that contradict known theories and known observations and facts.

This is the main thread of Against Reason: in order to show that methodologies always fail, Feyerabend home in on the need for consistency and coherence with facts and accepted theories required by Traditional Rationality, and dismantle it.

Feyerabend modus operandi is to take an episode in the history of science (mostly Galileo’s defense of Copernic, but many other examples from Physics appear) that every scientist will consider as progress and a step in the right direction, and then show that this episode would not have happened if the players followed the need for coherence with previous theories and facts that Traditional Rationality requires. At some point he even manages to argue that Traditional Rationalists should side with the Church in Galileo’s trials, because Galilieo and the Copernican perspectives only had a couple of nice conceptual tricks going for them instead of hard data and compatibility with previous known facts and theories!

What are his main points?

For understanding a given theory, alternatives (especially incoherent ones) are necessary, both for clarifying the underlying assumption and to propose stronger empirical tests.
- His main example is the incompatibility of Brownian Motion and the classical second law of thermodynamics, which he argues could not have been noticed without an alternative theory of heat (the kinetic one) because of the impossibility to measure the breaking of the second law by Brownian Motion. I don’t feel like I can really judge that point, but it sounds reasonable to me. Would love a comment/criticism if you feel this is nonsense.
New theories can’t survive direct comparison with old ones because they lack the latter’s auxiliary science and apparatus.
- This just makes a lot of sense, and is confirmed by all the modern history of science that I have read. When some theory has hundreds of years behind it and the other has 15, we should expect the latter to be wobblier than the former.
Because new theories can’t be compared with old ones directly, they should be preserved and explored long enough to gather the strength for such a comparison.
- Follows from the previous point. Here I feel the pull of Feyerabend’s anarchism, because he wants to allow any theory to survive that way. Whereas I feel there are stronger criteria to apply even at that stage, beyond beating the previous theory. Maybe being one of the alternatives that clarify the established theory? Not sure if that would apply to relativity though, which is an important example. (This is probably where simplicity plays a role too)
- One of the most controversial part of the book is when Feyerabend argues that Galileo ensured this preservation of Copernican physics by sheer propaganda and manipulation.
Theories incompatible with the facts should be allowed at an initial stage, because they can point to the right direction while having quantitative and qualitative failures.
- At this point, Feyerabend mostly lists a bunch of productive theories that had initial problems, quantitative and qualitative: Bohr’s model of the atom, special and general relativity, classic electrodynamics, Newton’s theory of colors, Kepler’s rule of telescopic optics…
Sometimes a theory is incompatible with the facts because the facts have been contaminated by a previous theory.
- This part is so interesting that I will write a post detailing it. Basically, it’s the theory-ladenness of observation that Kuhn constantly talks about, but demonstrated in a clear way. Feyerabend's main example is that at the time of Galileo, there was an obvious and probably unconscious equivalence of observed motion with absolute motion. Which meant that if the Earth moved like Copernic claimed, it should be noticeable (when you drop a stone from a tower for example). Yet in this case, as Galileo saw, it was the “facts” and the natural interpretation of the observations that was at fault: only relative motions can be observed.

On the negative side, I was not really interested in Feyerabend's argument for humanism (which is all about letting all perspectives roam free), and he often sounds like he’s overstating his case. That doesn’t change the insight of his analysis, especially with regard to natural abstraction.

Still, I find myself frustrated. Feyerabend clearly has a deep understanding of the history of physics and the underlying process that opportunists like Galileo, Bohr and Einstein used to advance their craft. Yet when the time arrives to extract some guiding principle out of these more concrete and organic processes, he stays true to his beliefs and jumps to anarchism. He thus has very little to say about my main hamming question: how to hit a small target in the high-dimensional space of hypotheses ?

The Knowledge Machine, Michael Strevens

This is a great modern book on philosophy of science. Just the initial short and clear explanation of Popper and Kuhn’s views, both their inherent weirdness but also their strengths, is worth the price of the book alone.

Strevens wants to explain two mysteries: why is science so powerful and why did it take so long to arrive. I feel like he does a decent job of answering both at a structural level, taking the mantle from Kuhn in many ways. His answer, the iron rule of explanation, captures many of the subtleties of science as a process, and accounts for progress despite constant irrationality. It just fails to help with hitting a small target in the high-dimensional space of possibilities; it simply assumes this ability without really stating it. Or said another way, Strevens articulates well the social process of Science as described by Yudkowsky, but fails to explain how scientists themselves (and Bayes) can be faster than Science.

Strevens’ iron rule amounts to only accepting empirical evidence to settle a disagreement. That’s it. And he makes a strong case that this basic prescription (and its ramifications) explain a great deal about the power of science. Basically, this pushes every scientist to strive for vindicating her own theory and view, in full swing of subjective preferences, but the arguments can only be accepted through empirical testing. Strevens describes an objective game that subjective and irrational scientists play, which result in forward progress by focusing more and more empirical testing on the disagreements (he calls that Baconian Convergence following Francis Bacon).

As for why Science only happened when it did, he claims that the iron rule is so counterintuitive (because it refuses any other type of argument, be it philosophical or theological) that it took the highly powerful and compartmentalized mind of Newton, in addition to the liberal atmosphere of 17th century Western Europe to give it a chance for long enough to see the result.

Most interesting to me was Strevens’ emphasis on motivation: one big use of the iron rule is to clarify enough the rules of the game that everyone feels like investing time in it will be valuable, even the tremendous effort and whole lifetimes needed by some modern experiments. In that way, Strevens captures a meta-paradigm of science, so that Kuhn’s restricted paradigms are not needed any more to explain the doggedness of scientists. This motivation aspect was definitely not on my radar, and I think Strevens model will be particularly useful here for my goals of forging a meta-paradigm of alignment.

I also enjoyed Strevens’ Tychonic principle (after Tycho Brahe), which says that disagreement between theories depends on more and more precise measurement. It’s not particularly new for anyone used to modern science, yet I have never seen it explicitly stated and argued for.

On the negative side, Strevens' blindspot about hitting small targets in a large possibility space corrupt some of his arguments. He notably ends up defending narrow-mindedness against humanism because the iron rule clearly leads to superior progress. Except this progress also comes from generation of good hypotheses, which quite often comes from weird intuitions around beauty and simplicity. I think he’s somewhat aware of this, as he has a full chapter on Beauty and Elegance which concludes by a weird “Yes, often beauty helps in finding better theory, but the iron rule is right for not following that lead.” Maybe because it leads to fewer failures in his mind?

This weakness aside, The Knowledge Machine is simply an awesome book for anyone interested in these questions of philosophy of science. It focuses on the real meat of the subject, is clearly written and brilliantly structured, and contains many great examples, not only from physics! (which is rare in general philosophy of science)

Understanding Nature: Case Studies in Comparative Epistemology, Hub Zwart

I didn’t finish this one. Not that it’s bad; I did learn a few things about, of all things, thes history of philosophy and a form of scientific wrestling in the interventionist work of Claude Bernard and Ivan Pavlov. Problem: I was instead looking for was insights and ideas on how fiction and narrative can complement more traditional scientific methods in understanding the world.

This is what the book set out to do: compare the epistemologies of science and literature with regard to life, notably animals and plants. And it sent all the right signals in the intro: the author clearly sees the issue with anthropomorphisation and projection of one’s own subjectivity on the fiction, implying that fiction can somehow transcend these issues.

Yet in the chapters I read, there wasn’t a single convincing contribution of the literary canon analyzed, except in weird metaphysical and phenomenological senses.

Okay, that’s not true; the notion of experimental novel from the french realist novelists was interesting. I knew about it (because that’s the sort of thing you have to learn in french high-schools), but Zwart did deepen my understanding of it. The idea is to see a novel as a laboratory, by creating characters with some dispositions, and confronting them to intervention and stimuli, seeing how they react.

The basic structure of an experiment can be articulated in the form of a rather straightforward question: “What happens if?” This is the basic sentence that guides experimental scientists in their laboratories as well as literary authors in their novels. Research subjects in an experimental setting, as well as principal characters in a novel, are exposed, by scientists and novelists alike, to a variety of circumstances and events in order to provide their readership with a viable answer to the question What happens if? Although novelists usually work with human subjects, they may decide to work with animals at times. In that case, their novels and stories become similar to animal experiments, to some extent at least.

That being said, I see such “experiments” as more about scenarios than fiction; these are narratives, but they don’t really capture the cluster that “stories” point at IMO. Something is missing there, which relates to how stories and fiction make us feel rather than how they treat their characters. And for articulating that point, this book doesn’t help me at all.