The Deutschian deadend

critical rationalism is full of tortured rationalizations

Preamble: David Deutsch is one of the most brilliant physicists and philosophers of our time. He is also wrong about key aspects of his philosophical worldview. The essay below attempts to explain why.

---

I. Introduction

For the reader: this is about some of the technical details of David Deutsch's philosophy. If all you took away from Deutsch is that optimism is good, problems are solvable, and humans are significant—those things are kept in tact, and you can carry on with your day. But, if what you took away from reading Deutsch was an entire philosophy of knowledge, a worldview that percolates deep into your understanding of humans, truth, and reality—then I have some bad news for you.

I should include some background on myself: I originally discovered Deutsch about four years ago, and I became obsessed. I read multiple of his and Karl Popper’s books, papers, lectures and interviews, and took many thousands of words of notes on all of it. I really felt like I was coming face to face with a profoundly significant set of ideas—a deeply clear, comprehensive, and coherent picture of reality. I now believe this was wrong, and that I had fallen prey to a number of fairly obvious conceptual and psychological errors.

A note on terminology: in this essay I’ll be grouping together the ideas espoused by David Deutsch, Brett Hall, Karl Popper, and Chiara Marletto into one loose bucket, which I’ll refer to interchangeably as “critical rationalism,” or the “Deutschian” or “Popperian” worldview. While I recognize they don’t all have the exact same position on every question, they share enough common philosophical ground to be grouped together for the sake of argument.

Put most succinctly, the critical rationalists’ mistake is that they view the world in terms of simplistic dichotomies. They formulate theories for how things work—say, how humans obtain knowledge about the world—and then fit every conceivable piece of evidence into that theory, making their ideas effectively immune to criticism. When taken seriously, their ideas lead them to a very distorted understanding of how humans work, and how we know things about the world.

The way I’ll make many of the arguments in this essay is oblique rather than direct. Some of my points will be about the meta of Deutsch’s statements rather than the substance of them; I’ll even make some sociological observations rather than purely technical/philosophical ones. I think this is necessary because, as I’ll try to demonstrate, critical rationalism is based on a set of mutually reinforcing assumptions and framings that are effectively impossible to refute all at once, purely on the substance of their claims. There is no single, definitive contradiction within the philosophy—if there was, Popper and Deutsch themselves would have found it long ago. But it’s possible to step just outside of the philosophy—to see the web of framings, rationalizations, and everyday behaviors that make up the worldview—to see the problems in it. The Deutschian worldview is a kind of philosophical “local maximum,” from which it’s impossible to move towards greater conceptual clarity without first taking apart a lot of conceptual scaffolding.¹

A big part of my motivation for writing this essay is psychological: what I would like to do is help a version of myself three years in the past who, by virtue of being so absorbed in the Deutschian worldview, felt a sense of existential loneliness, and got into a number of silly arguments in which he tried to convince others of this worldview (you might go as far as calling it a religion), and failed, mostly because in all these conversations he and his counterpart were talking past each other. I'm trying to help a past version of myself waste less of his time and emotional energy on silly arguments trying to defend Deutsch.

What I am trying to do is also sociological—I'm trying to explain the apparent contrast between a large number of people who are convinced that Deutsch's books (The Beginning of Infinity in particular) are some of the most powerful and important books ever written, and an even larger part of the population (including much of academic philosophy and a decent chunk of the scientific community) who tends to respond to Deutsch’s work with ambivalence, appreciating its value but not recognizing it as revolutionary. I am trying to explain why it is that some people come away from reading Deutsch totally transfixed and transformed, and others come away feeling somewhere on the spectrum between “this was silly” and “it was pretty good I guess.” It is not, as many Deutschians believe, simply a matter of the latter group not understanding what Deutsch is saying. There is something deeper going on.

Now, Deutsch and the critical rationalists deserve a ton of credit: it’s a fundamentally well-intentioned way of thinking. There are many basic tenets of theirs that I still agree with: knowledge is possible and it’s good; humans have an important role to play in the universe; the scope of what we can know about the world is unimaginably large; technological progress is a good thing and much more of it is attainable. The thing is, all of these principles remain true, even without all the tortured philosophizing that Popper and Deutsch do. Critical rationalism adds an unnecessary, pseudo-rigorous veneer onto these more obvious and basic principles.²

By the nature of Deutsch and Popper’s ideas being abstract, this essay will also necessarily be abstract. To combat this, let me ground the whole essay in a concrete empirical bet: Popper’s ideas about epistemology, and David Deutsch’s extensions of them, will forever remain in the footnotes of the history of philosophy. Popper’s falsificationism, which was the main idea that he’s widely known for today, will continue to remain the only thing that he’s widely known for. The frustrating fact that Wittgenstein is widely regarded as a more influential philosopher than Popper will continue to remain true. Critical rationalism will never be widely recognized as the “one correct epistemology,” as the actual explanation (or even the precursor to an explanation) of knowledge, progress, and creativity. Instead it will be viewed, like many philosophical schools before it, as a useful and ambitious project that ultimately failed. In other words, critical rationalism is a kind of philosophical deadend: the Deutschian deadend.

II. Fitting everything into your theory

Critical rationalists formulate theories for how they think certain things should work, and then they view everything as an example of that theory.

We can start with one of the most basic of Popper’s ideas—that of conjecture and refutation—which Chiara Marletto describes in her book The Science of Can and Can’t:

Given that knowledge has such an essential role in the survival of complex entities, it is essential to understand the process by which new knowledge is created from scratch in our mind. Fortunately, this process was elucidated by the philosopher Karl Popper in the mid-twentieth century. He argued that knowledge creation always starts with a problem, which we can think of as a clash between different ideas someone has about reality. For example, when writing a story, the clash in the author’s mind might be between the desire to use elegant, lyrical language and the necessity of keeping the attention of the reader alive with a gripping plot. The author has to find a way of meeting both these criteria, which may clash in certain situations: a long passage describing an idyllic landscape might give a perfect chance to meet the former criterion but might result in the reader dropping the book and switching on the TV, because it slows down the pace of storytelling. To address problems such as this, one has to create new knowledge.

Marletto claims that Popper has “elucidated” the process by which new knowledge is created. But has he really elucidated anything? This passage exemplifies one of the central critical rationalist tendencies: framing your theories in terms of highly abstract concepts, which then allows you to frame any conceivable situation as an example fitting your theory. “A clash between different ideas someone has about reality” can be applied to any conceivable human experience. You could describe anything as a conflict between two ideas. Me wanting to write a book is a conflict between the fact that I want to write a book and the fact that I haven’t written a book yet. This is not to say that it’s untrue: there are of course conflicts in our mind all the time. But this alone doesn’t bring us any closer to an understanding of how we create knowledge. It’s like saying: “humans learn about the world by asking questions and then answering them.” Great, now what do we do about this?

What makes critical rationalism appealing is the way it seems to “unify” disparate ideas. It literally unifies all of human activity into one big category of “knowledge-creation via conjecture and refutation.” As you’re going about your day and walking around, you’re creating conjectures and criticizing them; when you’re working through an emotional problem, you’re creating and criticizing conjectures; if you’re talking to someone, you’re creating and criticizing conjectures; if you’re dancing or making art, you’re creating and criticizing conjectures. Evolution is also doing the same thing—creating and criticizing (implicit) conjectures. Everything is about knowledge: DNA contains knowledge, technology contains knowledge, all of culture is a bunch of knowledge. What distinguishes good art from bad art is the fact that it has “more knowledge.”

If you’re into neat conceptual unifications and have a fetish for “grand unified theories that explain everything” this will be very satisfying—but again, it’s not clear what we actually get from this grand unification. The question for critical rationalists is this: what specific advances have we made across fields as diverse as fiction-writing, painting, biology, neuroscience, and music composition, by virtue of viewing all of them as instances of “conjectures and refutations”? I argue that because the model is at such a high level of abstraction, it hasn’t actually been useful across all the domains it claims to apply to.³

To be fair, Popper and Deutsch’s epistemology goes deeper than this one idea. To see how this “rationalizing reflex” is present throughout Popperian thinking, it helps to look at some of the other core principles that critical rationalists espouse.

Observation is not always theory-laden

There are two ideas are the core of Popper’s epistemology which are, as far as they go, true, useful, and eloquently put—except that critical rationalists, once again, over-use and over-apply them.

The first is that “observation is theory-laden.” Here Popper is saying that there is no such thing as a “totally pure observation statement.” A statement that you might naively think of as purely observational—“the temperature of the water in this cup is 40ºC”—is actually loaded up with a bunch of implicit theories and concepts, like the concept of water, temperature, dimensions, the notion of “celsius,” the idea of numbers, and the idea of a “cup.” So when you make such a statement, it’s not a pure objective observation because it carries this baggage of implicit theories with it. There are even theories embedded in the very structure of the language you’re using to make the observation—e.g. the subjective-verb-object structure of clauses in English.

Why is the theory-ladenness of observation so important? This brings us to the second Popperian principle, which is that “the truth is not manifest.” Popper stakes this claim in contrast to some other philosophers who claim that we have “direct access” to the truth of the world through our senses. Popper very rightly points out that the truth is not obvious, because we can be deceived in many ways. There are countless examples of this, from mundane optical illusions, to errors in our measuring instruments, to superstitions that taint our experience of the world. All of these are “sources of fallibility”—they act as barriers to our “direct access” to the truth.

Now, to see the limits of both of these claims—that observation is theory-laden and the truth is not manifest—you need to appreciate the underlying frame in which they’re made. The underlying assumption here is that we, as humans, are irrevocably “cut off” from the actual objective world, only making tentative conjectures about it, unable to ever “verify” anything we know. Critical rationalists literally claim that we can never assert the truth of anything—not even the most basic observations like “it’s Wednesday” or “I’m alive.” For critical rationalists, not only are we prohibited from saying these statements are true—we can’t even say that they’re likely to be true. The entirety of our knowledge is, as they put it, a bunch of “unjustified untruths.”

The critical rationalist’s picture of the world is one of radical skepticism, born of a metaphysical tradition dating back hundreds of years to Descartes and others, which Deutsch and Popper never seriously question. In this picture, we are always “groping around in the darkness of a cave,” mired in our “infinite ignorance.” This picture misses something absolutely crucial, which is that we, as humans, are fundamentally part of the objective world—we are embedded in it, and even continuous with it, rather than irrevocably severed from it. Brian Cantwell Smith writes about transcending this metaphysical frame when he writes:

What would it be to recognize, finally, that the various forms of metaphysical separation—between representation and represented; between the parts of divvied-up reality; between and among concepts, types, or properties—what would it be to recognize that these forms of separation, like the separations we maintain in our political and emotional lives, are all partial: negotiated, gradual, welling up and subsiding, dynamically maintained, in a kind of on-going dance? What would it be to see the world as partially pulled apart, that is, making room for pluralism, error, autonomy, individuality, and heterogeneity, and as partially put together, making room for normativity, communion, humility, and transcendence?

It is not so difficult an image. Think of a potter pulling apart a particular sticky kind of clay, pushing globs of it away, stretching and squishing and clumping it together, forming shapes and drawing out spaces between and around it—except that we potters are just more clay.

Once we recognize that we are part of the world, constantly in contact with it and enmeshed in it—or as Cantwell-Smith puts it, “partially pulled apart” and “partially put together”—it becomes easier to accept the possibility of “direct observation” and “manifest truths.” There are many things you can directly observe, and which are “manifestly true” to you: what you’re wearing at the moment, which room of your house you’re in, whether the sun has set yet, whether you are running out of breath, whether your parents are alive, whether you feel a piercing pain in your back, whether you feel warmth in your palms—and so on and so forth. These are not perfectly certain absolute truths about reality, and there’s always more to know about them—but it is silly to claim that we have absolutely no claim on their truth either. I also think there are even such “obvious truths” in the realm of science—like the claim that the earth is not flat, that your body is made of cells, and that everyday objects follow predictable laws of motion.

As Cantwell-Smith puts it, our theories do partially mediate our access to the objective world—but they don’t entirely determine our perception of it:

There is much that is right in this argument [that we don’t have direct access to objective reality], especially the claim that we have no conscious access to the world independent of historical, cultural, and personal interpretation. But from the fact that we have no access independent of such interpretation, it does not follow that we have no access to it at all, or that the access we have is entirely determined by those interpretive schemes.

We are not disembodied explanatory souls trying to pierce into the nature of reality through a infinitely many miles of darkness and ignorance. We are in the world, part of the world, in direct contact with it. Our “direct access” to the world is a crucial ingredient to our success in obtaining knowledge—just as much as our capacity to create conjectures.

The triviality of “universal explainers”

One of Deutsch’s famous quips is that we are “universal explainers”: we, as humans, can explain everything that can be explained by any entity. Anything in the world that can be understood, can be understood by us. He uses this argument to claim, for example, that superintelligent AI couldn’t possibly be fundamentally different from us, and likewise with aliens. The difference between advanced AI and us would not be one of kind, but rather one of degree: whatever intelligence they have, we could supplement our own intelligence with more memory and faster processing speeds (e.g. with computers, or trivially, notebooks and pencils) in order to get to the same level of intelligence.

Dwarkesh Patel writes about why this is a silly argument:

This is a bit like saying that it’s inaccurate to suggest that a car has a higher range or speed than a bicycle. With future technology, we could replace the bicycle rider’s ATP stores with an atomic battery, and its muscles with miniature jet engines. In fact, bicycles and cars are both “universal vehicles” which share the same fundamental capacity for transportation. The only differences which can exist between them are the size of their energy supply and the power of their engines. Since you can always extend these two attributes, one vehicle cannot be more powerful than another.

In Deutsch’s view, our brains are basically Turing machines but with a special “creativity algorithm” installed—that’s what distinguishes us from other creatures. A Turing machine can always be augmented with more memory and faster processing, so we are functionally equivalent to aliens and superintelligent AI’s. This is technically true, and functionally unhelpful. Those who are concerned about AI safety are worried about AI systems becoming much more intelligent and more competent than us very quickly. If these systems do surpass us that quickly, it doesn’t really matter that we could technically augment our brains with a bunch of memory—we become functionally much less capable than them.

Deutsch’s claim that we are universal explainers involves a highly speculative connection he draws between the Church-Turing-Deutsch principle (which asserts that a sufficiently powerful computer can simulate every physical process) and our own brain’s capacity to understand the world. It assumes that when we form explanations, we are “running a program that simulates the outside world” in the exact same way that a computer simulates weather patterns. This seems plausible, but is not a given – many people disagree with this framing, and in the absence of a clear understanding of how our brain forms explanations, it’s hard to settle the question.

Beyond this tenuous connection, Deutsch’s point about universal explainers basically boils down to an instrumental argument. The argument for universal explainers is: if there are things we truly can’t understand, then by definition we can’t talk about them (otherwise we’re basically “talking about the supernatural”), so why bother talking about them. It’s an argument of the form, “if I can’t see it, it doesn’t exist”—which is a decent enough argument in practice. Except that it’s an instrumental argument, and in most other situations, critical rationalists staunchly oppose instrumentalism.

False predictions on AI

In his 2012 essay Creative Blocks, Deutsch argued that the dominant AI paradigm was doomed, because it was ignoring the key capability of humans: our ability to create and criticize conjectures. He was convinced that the approach of building AI by “learning from experience” would never work, because we as humans don’t “learn from experience.” This is part of Deutsch and Popper’s critique of empiricism—the view that knowledge is derived from our sense experiences. In Deutsch’s world, theories always come first. We formulate a theory (perhaps consciously or unconsciously), and that theory is the framework through which we make observations. The observations might refute our theory, in which case we discard it and come up with a new theory.

The predominant paradigm in AI for the past few decades has been the opposite: start with a bunch of data, and train your algorithm to figure out patterns in the data, and hope that it will generalize to uncover “deeper knowledge” contained in the data. Many people were convinced this wouldn’t work at first, that merely extrapolating from statistical patterns in data would never amount to true intelligence. And then, year after year, this exact approach in AI helped to solve problems that we had previously been unable to get computers to solve, and we now know it’s enough to pass the Turing test. The very approach of “learning from data” which critical rationalists would deride as “empiricist” can now outperform doctors at diagnosing illness and support graduate-level math research, among many other capabilities. But because this AI is built under an “empiricist” paradigm, critical rationalists still insist that they are not and will never be intelligent in the way we are.

Deutsch believes that everything that is special about humans comes down to the “creativity algorithm” that’s implemented somewhere in our brains. This is the thing that enables us to make conjectures about the world, and Deutsch has speculated that it’s also the same thing that gives us free will and consciousness. And on top of that, he asserts that this is the same algorithm that any other “generally intelligent entity” in the universe must have. To be fair, the questions of intelligence, consciousness, and free will are all open questions, and Deutsch’s idea of the “creativity algorithm” is intriguing and worth exploring—but at this point it seems too simplistic to pin down every big question about our humanity to a single, discrete algorithm in our brains that we just haven’t found yet. Many other creatures seem to be conscious even if they don’t make “conjectures” about the world, there are plenty of examples of intelligence in the absence of “explanatory knowledge”, and there are now machines that convincingly speak and think, but who never had a creativity algorithm installed in them. I would not bet on all of these capabilities coming from the exact same place, in a single “jump to universality” that we made.

An intolerance of ambiguity

There is one other aspect of the Deutschian frame that Popper and Deutsch never seriously question: it’s that the correct way to think about “how we obtain knowledge of the world” is through the lens of formal logic.

The title of Popper’s first book is A Logic of Scientific Discovery—he was specifically after a logic of science, and that quest has been continued by Deutsch. The critical rationalists think about science in the frame of “forming logical statements,” and figuring out the truth or falsity of those statements.

The problem with formal logic is that it is intolerant to ambiguity, and it can’t accommodate any notion of “partial truth.” If any part of a system of logical statements is untrue, its falsehood “bleeds out” into the entirety of the system. This is why critical rationalists make statements like “Newton’s laws are false”—because for as much as we’ve gotten out of Newton’s laws, the fact that we’ve found conditions in which they do not apply makes the theory as a whole strictly false. And this is also why it’s impossible, under fallibilism, to ever establish that any theory of ours is true—because we are concerned with logical truth, and logical truth requires us to check every conceivable instance and implication of a statement, which is completely impossible.

(Of course, the more commonsense view of science is that we are not after strict, perfect truths, and so it makes no sense to call Newton’s laws “false”—they are true in many circumstances, and false in others. Like all theories, they have domains in which they apply, and domains in which they don’t. The fact that they are imperfect does not mean that we’ve discarded them entirely.)

In a conversation on statements, propositions, and truth, Deutsch makes his metaphysical assumptions explicit when he says that “reality is pristine”—in his view there is a perfectly precise truth of the matter to every aspect of reality. Not that we’ll ever reach it, but it’s there.

But we have very little reason to think that “reality is pristine”—what we’ve found again and again is that the world is not amenable to exact, perfect separation into well-defined categories. Hofstadter illustrates how the concepts we use to slice up the world all ultimately have poorly defined boundaries, and David Chapman does the same for everyday objects. The aforementioned Brian Cantwell-Smith paper likewise refutes this stubborn “metaphysical discreteness.” Franklin and Graesser make the same point when they comment offhand that “the only concepts that yield sharp edge categories are mathematical concepts, and they succeed only because they are content free.”

For the critical rationalists, all this ambiguity and fuzziness that we find in our concepts is simply a matter of our own fallibility—there is a perfectly precise truth of the matter about everything in the world, we just have trouble seeing it because of our limited perspective. According to Deutsch, there is a collection of “infinitely precise and perfectly unambiguous abstract propositions” which describe every single facet of the universe in complete detail. These propositions exist in the abstract realm, independent of space and time. Now, it’s impossible to say whether these propositions really do exist, but what’s clearer is that we really don’t need them, if we stop framing our quest for knowledge as a search for logically true propositions (or approximations to such propositions).

I said earlier that the critical rationalist project will ultimately be deemed a failure, and this is the exact sense in which I mean it: we will eventually give up on trying to formulate science and knowledge-creation as approximations to formal logic. We already have plenty of reasons to believe this project is doomed—whether it’s in the intractability of the frame problem and relevance realization, or the pesky fact that semantics cannot be reduced to syntax. And it’s not a problem: we carry on creating new knowledge regardless. What I suspect will happen is that we will, collectively, “set aside” this project, in the same way that we’ve set aside previous philosophical projects, as Richard Rorty describes in Philosophy and The Mirror of Nature (emphasis mine):

Wittgenstein, Heidegger, and Dewey are in agreement that the notion of knowledge as accurate representation, made possible by special mental processes, and intelligible through a general theory of representation, needs to be abandoned. For all three, the notions of "foundations of knowledge" and of philosophy as revolving around the Cartesian attempt to answer the epistemological skeptic are set aside. Further, they set aside the notion of "the mind" common to Descartes, Locke, and Kant—as a special subject of study, located in inner space, containing elements or processes which make knowledge possible. This is not to say that they have alternative "theories of knowledge" or "philosophies of mind." They set aside epistemology and metaphysics as possible disciplines. I say "set aside" rather than "argue against" because their attitude toward the traditional problematic is like the attitude of seventeenth­ century philosophers toward the scholastic problematic. They do not devote themselves to discovering false proposi­tions or bad arguments in the works of their predecessors (though they occasionally do that too). Rather, they glimpse the possibility of a form of intellectual life in which the vocabulary of philosophical reflection inherited from the seventeenth century would seem as pointless as the thirteenth-century philosophical vocabulary had seemed to the Enlightenment. [...] Wittgenstein, Heidegger, and Dewey have brought us into a period of “revolutionary” philosophy (in the sense of Kuhn's “revolutionary” science) by introducing new maps of the terrain (viz., of the whole panorama of human activities) which simply do not include those features which previously seemed to dominate.

We can put aside the Deutschian project. Rather than trying to eradicate ambiguity and keep seeking “perfect truths,” we can embrace the inherent nebulosity in the world, and accept that exactness and ambiguity are two inseparable elements of the world and our relationship to it.

III. Systematic mistakes in Deutschian thinking

With these last few sections, what I’d like to do is take a more meta/sociological stance towards critical rationalism, where I try to tackle why it’s possible for someone to be as confused as I claim that critical rationalism is, while also being staunchly defensive of it.

Popperians ignore the problem of meaning to their peril

This isn’t so much an obvious logical problem with critical rationalism, but rather a way of thinking that leads a lot of critical rationalists into confusion.

Popper famously described the problem of meaning as a “pseudoproblem” – as long as we think we understand our counterpart’s view, we don’t need to obsess over the meaning of words. He wrote, for example:

Linguistic precision is a phantom, and problems connected with the meaning or definition of words are unimportant. Words are significant only as instruments for the formulation of theories, and verbal problems are tiresome: they should be avoided at all cost. (37-38)

Popper is right that it’s possible to focus too much on the definitions of words, but in his broad dismissal he also ignores genuine problems about the relationship between our ideas, our language, and the world. The meaning of words is neither inherent to the shape of the words themselves nor to our brains – it lies in our interaction with the world, as David Chapman illustrates with the parable of the pebbles:

In the 1980s, this conundrum, “the problem of intentionality,” became the central issue in the philosophy of mind. Cognitivism collapsed when it became clear that no answer is possible—not because we don’t know enough details about how brains work, but even in principle.

Representation is not a property of the bucket, pebbles, or sheep. It’s a property of the whole history of interaction of the bucket, pebbles, shepherd, sheep, and gate. Likewise, beliefs aren’t in your head; they too are dynamics of interaction. Representation can’t be found in a snapshot of the state of the world, nor in a timeline of brain activity. It’s necessarily a process extended in both time and space.

Why does this matter for critical rationalism? This bias against problems of meaning leads some critical rationalists to never question the meaning of their utterances, even their most abstract philosophical claims, because they’ve internalized the idea that investigating meaning is a waste of time. A common pattern in critical rationalist thinking is to define words in very particular ways, and then refuse to question those definitions. Deutsch finishes each chapter of his books The Fabric of Reality and The Beginning of Infinity with a list of such definitions. Sometimes the definitions are interesting reframings of everyday concepts (like defining a “person” as “an entity that can create explanatory knowledge”), but other times they are caricatures of opposing positions that are specifically framed in a way that bolsters his argument.⁴ It’s fine to define words in new and interesting ways, but the denigration of “problems of meaning” leads some critical rationalists into a blindspot, where they imagine that their disagreements with other philosophers are always about substance and never about mere meanings, and fail to see the way they and their opponents are talking past each other.

Popper dismisses introspection

Another common tendency among critical rationalists is a dismissal of introspection as a useful tool in doing philosophy. This happens because a lot of critical rationalists have internalized a strict distinction between the logical content of our ideas (which, in their view, is the only thing that really matters for epistemology), and our psychological states, like the emotions we have about those ideas. Popper does this explicitly in the first chapter of The Logic of Scientific Discovery, in a section literally titled “elimination of psychologism.”

I now believe that cleanly separating the content of our ideas and our feelings about our ideas is extremely difficult if not outright impossible, and attempting to do so only makes it harder to think clearly. Raymond Smullyan’s essay Is God a Taoist? is a good example of how introspecting a bit can help partly resolve abstract philosophical problems: he shows that much of the confusion about free will stems from an implicit analogy we make between “the laws of physics” and “laws” in the sense of civil code:

Your acts are certainly in accordance with the laws of nature, but to say they are determined by the laws of nature creates a totally misleading psychological image which is that your will could somehow be in conflict with the laws of nature and that the latter is somehow more powerful than you, and could "determine" your acts whether you liked it or not. But it is simply impossible for your will to ever conflict with natural law. You and natural law are really one and the same. […]

Don't you see that the so-called "laws of nature" are nothing more than a description of how in fact you and other beings do act? They are merely a description of how you act, not a prescription of of how you should act, not a power or force which compels or determines your acts.

Here’s a specific example: I think many of the biggest fans of critical rationalism would be well-served by introspecting on what they find so appealing about David Deutsch’s books. If you’re not careful, a book like The Beginning of Infinity, despite all its claims about how everything is conjectural, can ultimately become a kind of “existential grounding” for you, psychologically. You can test this by asking a simple question: if you found out that Deutsch was deeply wrong – like, stupidly and obviously wrong, about many important points – would you have a negative emotional reaction to this discovery? If that's not the case for you – then great, you are not emotionally attached to Deutsch's ideas. But I imagine there are at least a few people out there who are fans of Deutsch specifically because of the sense of meaning it gives them, as it did for me.

This is especially true if you’re someone who has a soft spot for philosophy, for big ideas. And in particular, a soft spot for the power of thought. You have always kind of wished that, by virtue of just sitting in a room with your books and pencil and paper, and just sitting down and thinking hard about things, you could potentially land at a truly transformative, earth-shaking insight. And not only would it be a powerful insight, but it would be an intuitively understandable insight – an explanation. It would upset you if this is not possible. Deutsch’s work is emotionally satisfying because it puts theorizing at the very center of human activity, as the most significant and consequential thing that we humans do.

To be clear, this entire argument on its own is not a refutation of Deutsch’s ideas. Rather, it’s a nudge: that if you’re someone for whom all of the above psychological points apply, then you might be well served by inspecting whether some of your defenses of Deutsch’s ideas are genuine arguments you believe, or if they stem from a desire to rationalize a worldview you’re emotionally attached to.

Critical rationalists blind themselves to the ladder of abstraction

In a podcast discussion between Jake Orthwein and Chris Lovgren, they point out one of the funny quibbles between fans of Popper and fans of David Chapman: both parties criticize the other party as being “too abstract.” When I first encountered this claim, I actually had difficulty understanding how critical rationalism is abstract. And I think at least part of the answer is that the critical rationalist worldview blinds you from seeing just how abstract it is.

In the critical rationalist worldview, everything in your mind is an idea. Feelings are ideas, perceptions are ideas, thoughts are ideas. Every piece of knowledge we have is made up of ideas, or theories. And as I mentioned in the section on “observation is theory-laden,” all of our observations are also inextricably enmeshed with ideas. Deutsch also argues forcefully for “the reality of abstractions” (one of the chapters in his book), asserting that any idea that plays a role in our “best explanation” of something is a real thing. In effect, he says that highly abstract concepts like complex numbers, causality, and counterfactuals are just as real as the table in front of you. Your understanding of the table, after all, is mediated by a bunch of abstractions! We exist in a thick morass of abstractions and there is simply no way out of it.

Once again, we’re in a situation where what Deutsch claims is true in a technical sense, but it distorts your understanding of the world in a profound way. In the more commonsense view of the world, there is a very obvious distinction between “abstract” things and “concrete” things. What makes something concrete is that it’s here, now, tangible. What makes something abstract is that it requires layers of concepts to talk about: to talk about complex numbers, you first need to talk about the abstract notion of “real numbers”, which are an abstraction built on top of “whole numbers,” which are still abstractions over the concrete notion of “two apples in front of you.” There is an important sense in which the “concrete” things are “more real” than the “abstract” things—as in, given our particular brains and bodies, they are easier to grasp and their existence is more easily verifiable.

When you enmesh yourself deeply in the Deutschian worldview, this hierarchy is wiped away.⁵ The notion of “complex numbers” and that of “two apples in front of you” somehow end up on the same footing, because they are both theory-laden and conjectural. This is what makes it so that critical rationalists have trouble seeing why their claims are so abstract, and as a consequence so frustrating to argue with. (They’ll make statements like “Popper proved that theories are not generalizations of observations,” which is only true when you take on their particular, strained definitions of “theory” and “generalization,” but is otherwise a hopelessly vague claim.⁶)

This brings us to another, related hierarchy that critical rationalists blind themselves to:

Critical rationalists dismiss the “hierarchy of reliability”

Most people subscribe to the commonsense view that there’s a certain “hierarchy of reliability” between math, science, and philosophy, where mathematical truths are taken to be “certain,” scientific truths are “likely true, or strongly supported by evidence”, and philosophical truths are “largely a matter of taste.” In an example of this way of thinking, Erik Hoel writes that “Most disagreements in a highly abstract field break down into quibbling over language. Wittgenstein showed us that.”

Deutsch claims that this intuition is incorrect, because all of our knowledge is conjectural, and so we are equally unsure about everything. For example, he says that our knowledge of solipsism being false (the claim that there is nothing outside your individual mind) is equally as compelling as our knowledge that the square root of 2 is an irrational number. The claim that “knowledge is created via conjecture and refutation” is equally as compelling as the claim that “the earth is not flat.”

Deutsch writes:

Some philosophical arguments, including the argument against solipsism, are far more compelling than any scientific argument. Indeed, every scientific argument assumes the falsity not only of solipsism, but also of other philosophical theories including any number of variants of solipsism that might contradict specific parts of the scientific argument.

There are two different mistakes happening here.

First, what Deutsch is doing is assuming a strict logical dependency between any one piece of our knowledge and every other piece of it. He says that our knowledge of science (say, of astrophysics) implicitly relies on other philosophical arguments about solipsism, epistemology, and metaphysics. But anyone who has thought about the difference between philosophy and science recognizes that in practice they can be studied and argued about independently. We can make progress on our understanding of celestial mechanics without making any crucial assumption about metaphysics. We can make progress studying neurons without solving the hard problem of consciousness or the question of free will.

The second, bigger problem with Deutsch’s claim is that, like many of the other points in this essay, it’s true in a technical sense but completely misses the point. Deutsch is right that ultimately, we are not certain of mathematical truths, just as we are not really certain of anything.⁷ But here he’s using a literally impossible standard of “certainty,” where it means “known without any possible sliver of doubt and there is no chance in any conceivable world where this turns out to be false in the future.”

A more commonsense definition of certainty might be: we’ve thought long and hard about it, and we have extreme difficulty seeing how it could possibly be otherwise. In this way, we are certain of mathematical truths, less certain of many scientific truths, and completely uncertain about philosophical truths. This is obvious in everything from the way we talk about progress in math versus science and philosophy (we talk about “proofs” in math, compared to “arguments” in philosophy), to the rate at which we resolve disagreements (philosophical debates tend to take hundreds of years to resolve, and even then, are only ever really resolved once the question becomes concrete enough to become part of science).

To put it in terms that might make more sense to a critical rationalist: while our knowledge in all these fields is conjectural, we have wildly different methods of criticism for the ideas in them, and some of those methods are more robust than others. It’s very easy to come to agreement about the validity of mathematical proof (so much so that we have literally programs that do this mechanically), whereas philosophical arguments can be litigated endlessly because they are abstract and very imprecise. Science occupies a middle ground, where its claims are concrete enough that they can be tested and falsified with experiments, and it’s easy to agree on the outcomes of experiments.

When one twitter user once asked why so many critical rationalists are perceived as dogmatic, Brett Hall responded that it’s because people subscribe to this false hierarchy of reliability, and “no one accuses a mathematician of being dogmatic when they prove a theorem.” My claim is that it takes a particular kind of mental straining to defend yourself against claims of dogmatism by saying that “no one talks about mathematicians as dogmatic”—it betrays a belief that your philosophy has as much argumentative force as a literal mathematical proof.

IV. Coda

The problem with critical rationalism is not that it’s wrong in one fundamental way – but that it’s right in a bunch of unimportant ways. How many scientists actually use the idea that “beliefs don’t exist” to do their work? Who relies day-to-day on the maxim that “we don’t know if anything is true or likely to be true, only that it’s not yet falsified,” other than conspiracy theorists and self-help gurus? In response to this, Deutschians would say: the fact that these ideas are not widely used says nothing about how true they are. And that is, once again, strictly true – but it misses the point. You’ve backed yourself into a conceptual corner where all you’re doing is telling convenient stories about what the world is like, and no one’s really going to disagree with you, because your stories are “not even wrong.” That is why I call it the “Deutschian deadend.”

Critical rationalism does have some nuggets of truth in it, and thus it has value as one of many ways of trying to understand our relationship to the world—but it’s certainly not “the best explanation” for how we obtain knowledge. If you ask cognitive scientists, AI researchers, linguists, neuroscientists, and psychologists – do any of them credit critical rationalism and Karl Popper with giving them foundational insights into how our minds work and how we learn about the world? They don’t. Of course, many critical rationalists will then assert that this is why All Of Those Fields Are Wrong, and if everybody just sat down and understood science and knowledge in the exact way that we do we would no longer have this problem.⁸

To close, there are two sociological observations I’d like to make that, while not directly refuting this worldview, help illustrate the flaws in it. First, critical rationalists are staunchly anti-dogma: Popper’s whole schtick was that we are fallible beings, constantly prone to error, and so we should never hold any of our ideas as immune to criticism. And yet, critical rationalists are often described as more dogmatic than their counterparts. In a second contradiction, critical rationalists are emphatically against focusing too much on the meanings and definitions of words—and yet in many discussions, they seem particularly fixated on word choice and specific definitions.⁹ The point is: there are multiple contradictions between some critical rationalists’ stated beliefs (anti-dogma, against fixating on definitions) and their actions. And in my view this is yet another meta-problem that points to flaws within the philosophy itself.¹⁰

The reason this essay was important for me to write is that I was once the kind of person I’m describing here. After I read a lot of Deutsch and Popper, I started viewing most other philosophers of science as fundamentally confused, and even much of commonplace thinking about epistemology as mistaken, and made lots of bold pronouncements on twitter to that effect. I got into debates with friends about how their commonsense intuitions are wrong, and this new idea is actually the correct epistemology and is one of the most important ideas ever articulated. But then, over the course of another few years, more reading and thinking, and many conversations with friends who had gone through a similar evolution, I began to see the way that Popperian thinking rationalizes any critique as a form of misunderstanding, and ultimately makes its adherents parrot-like in their utterances. I eventually realized that all those principles I cherished—human fallibility, the possibility of understanding the world better despite it, and our potential to transform the world—are still true, even without all these tortured rationalizations.

---

Thanks to Ben and Jake for feedback on earlier drafts.

1. I want to emphasize that I’m specifically critiquing the philosophy of critical rationalism, rather than Deutsch or Marletto’s work in physics, which I don’t have a strong opinion about.

2. One other point in their defense: critical rationalism helps counter the unquestioned pessimism and antihumanism that pervades much of our culture. For that reason alone, their work should be admired. But again, I don’t think we need their philosophical arguments to defend optimism and humanism—there are much more basic, commonsense arguments for those views. And in practice, I don’t see critical rationalism making a lot of progress in convincing staunch pessimists/nihilists to become optimistic—it mostly seems to give people who were already partial to these ideas stronger conviction about them.

3. One counterargument: the “conjecture and refutation” model of creativity might inspire someone, because it makes them more optimistic, as they realize that making mistakes and criticizing bad ideas is part of the process of coming up with good ideas. This is fair, but it effectively reduces the whole theory to a self-help tool (“you can learn from your mistakes!”). If the whole point of Popper and Deutsch’s work is self-help, I support it, but I imagine a lot of critical rationalists believe there’s something deeper here than that.

4. For example, Deutsch defines instrumentalism as “the misconception that science cannot describe reality, only predict outcomes of observations.” When defined this narrowly, instrumentalism is obviously false. But wikipedia has a more nuanced definition: “instrumentalism is a methodological view that ideas are useful instruments, and that the worth of an idea is based on how effective it is in explaining and predicting natural phenomena.” Many of the arguments Deutsch makes actually subscribe to this kind of instrumentalism, like his argument against reductionism, and even his argument for universal explainers. Deutsch does the same thing when writing about Wittgenstein and other philosophers in The Beginning of Infinity: he presents the views of his opponents as simplistic caricatures, rather than actually attempting to engage with the substance of their views.

5. I’m making a psychological point here, and it might not be true of others! Reminder that this whole section is more about how critical rationalism can lead to errors in thinking, but if there are staunch critical rationalist who agree with me about abstract vs concrete concepts and don’t think it poses any issue for Popper/Deutsch, I would love to hear from them.

6. Here’s my attempt at explaining what critrats mean when they say that “theories are not generalizations of observations”—they are trying to refute induction. They’re opposed to the idea that theories just “pop out” of repeated observations. My view is that this is sometimes true, and sometimes not. Some of our theories really are just generalizations from data – “curve-fitting,” so to speak (e.g. Kepler’s laws, Ohm’s law, Boyle’s law, Mendel’s laws of inheritance). Other theories require some creative insight that transcend our previous way of thinking and they’re not just generalizing from past data (e.g. Einstein’s theories).

7. See this piece for an eloquent illustration of how our beliefs about math are based on tentative evidence just like everything else.

8. To be clear, neuroscience has its problems, and so does psychology, and I’m sure linguistics does too. And Deutsch’s chapter on “explanationless science” in The Beginning of Infinity does a great job at illustrating some of these problems. But again, I don’t think these scientists need to swallow the entire conjecture-and-refutation worldview, they just need to look at their implicit assumptions. They need grade-school level critical thinking (as well as cultural and institutional changes), not Popperian philosophy.

9. For example, Deutsch criticizes the Royal Society for saying that “the nature of science is to establish truth,” and counters by insisting that “scientific truth can be discovered, but never established.” See also footnote #4.

10. To be explicit about my beliefs, I think what some Deutschians do is hold their general philosophy as immune from criticism, which is easy to do because it’s always possible to come up with more rationalizations for why you’re right (given how abstract the claims are), and I think part of the way they’re able to do this is to define words in a particular way that makes their theories work, and then if anyone questions those definitions, they respond by saying “stop arguing about definitions, it’s a waste of time.” (For completeness, I’d also like to credit Deutschians who are generally not like this and actually are open-minded in online discussions, e.g. Carlos and Lulie.)