Turtles All the Way Down: The Exquisite Lie That Science Tells Itself
The most successful enterprise in human history cannot tell you whether it is telling the truth. This should trouble you more than it does.
There is something almost indecent about the confidence with which the modern world genuflects before science. Not before science as method—that rigorous, often tedious habit of testing one’s convictions against the stubbornness of reality—but before science as oracle. As the deliverer of capital-T Truth. As the final court of appeal in a civilisation that has, with admirable efficiency, demolished every other court it once recognised.
The question I wish to put to you today is not whether science works. Of course it works. The device upon which you are reading these words is proof enough that somebody, somewhere, got their equations tolerably right. The question is far more dangerous, far more elegant, and far more consequential than that. It is this: Does the fact that science works mean that science is true?
The distinction may seem trivial. It is anything but. It is, in fact, the fracture line running beneath the entire cathedral of modern knowledge, and the fact that most people—including most scientists—have never noticed it is itself a phenomenon worthy of investigation.
The Two Great Camps
The philosophy of science has, for centuries, been divided between two fundamental orientations, and as with all great disputes, neither side has had the decency to be entirely wrong.
On one side stand the scientific realists. These are the optimists, the believers, the noble souls who hold that science does not merely organise our experience of the world but describes the world as it actually is. Electrons are not bookkeeping devices. Genes are not convenient fictions. Tectonic plates are not metaphors that happen to predict earthquakes. These things exist, the realist insists, and our best theories tell us, approximately, what they are like. Science, on this view, is humanity’s grandest achievement precisely because it pierces through appearance to reality. Each successive theory is a better portrait, a closer approximation, a more faithful rendering of what is genuinely out there.
There is something magnificent about this conviction—the sheer ambition of it, the refusal to settle for anything less than truth itself. One cannot help but admire people who insist that the mind is adequate to the universe. The realist treats the human intellect with the seriousness it deserves, refusing to grovel before the unknowable or to retreat into the comfortable mediocrity of agnosticism.
The realist position rests upon three pillars. First, a semantic thesis: the claims of science are genuine assertions, capable of being true or false—not elaborate fictions dressed up in the language of mathematics. Second, a metaphysical thesis: the entities science describes exist objectively, independently of whether anyone is looking. Third, an epistemic thesis: we have good reason to believe that our best theories are getting it approximately right. Together, these pillars support the most optimistic vision of human knowledge ever articulated—the vision of a species that can, through discipline and method, comprehend the universe that produced it.
On the other side stand the instrumentalists. These are the pragmatists, the skeptics, the thinkers who observe that a map need not resemble the territory in order to get you where you wish to go. Scientific theories, on this view, are supremely effective tools—instruments for prediction, for technological mastery, for the systematic organisation of what we observe. But the question of whether they correspond to some mind-independent reality sitting patiently behind the curtain of phenomena? That question is either unanswerable, meaningless, or—most damningly—irrelevant to anything that science actually does.
The instrumentalist does not deny that science achieves extraordinary things. The instrumentalist merely denies that achievement is the same as truth. And here one must concede that the instrumentalist has, at the very least, good taste in arguments.
The Miracle and the Graveyard
The realist’s most powerful weapon is elegance itself. It is called the no-miracles argument, and it goes like this: if scientific theories were not at least approximately true, then the staggering predictive success of science would be an inexplicable coincidence—a cosmic fluke of such proportions that only a fool or a mystic would accept it. The fact that quantum electrodynamics predicts experimental results to twelve decimal places is not the sort of thing that happens by accident. Something must be going right. And what is going right, says the realist, is that the theory is tracking reality.
It is a beautiful argument. It has the clean lines and satisfying inevitability of a well-designed building. And like many beautiful things, it conceals a structural deficiency.
The instrumentalist’s counter-weapon is called pessimistic meta-induction, and while it lacks the elegance of its rival, it possesses something more useful: history. Consider the graveyard of abandoned theories that were, in their day, every bit as successful as the ones we now champion. The phlogiston theory of combustion explained everything the eighteenth-century chemist needed it to explain—until it didn’t. The luminiferous aether was not a fringe hypothesis; it was the considered opinion of the finest minds in physics, a theoretical entity so obviously necessary that its non-existence seemed inconceivable. The caloric theory of heat, the miasma theory of disease, the Ptolemaic model of the solar system—each of these was, by the standards of its era, predictively successful, explanatorily powerful, and wrong.
The pattern is not subtle. If every previous generation’s best theories turned out to be fundamentally mistaken about the furniture of the universe, what rational basis do we have for supposing that our generation’s theories are the exception? The realist’s response—that theory change represents progressive refinement rather than wholesale rejection—has a certain plausibility, but it also has the faintly desperate quality of a man moving the goalposts while insisting they have been in the same place all along. If the content of “approximate truth” shifts with every theoretical revolution, one begins to suspect that the concept is less a rigorous standard than a rhetorical device—elastic enough to accommodate any outcome and therefore incapable of being refuted by any.
The realist has attempted to salvage the position through what might be called a strategy of selective inheritance: carve up the old theory, identify the bits that were genuinely doing the predictive work, and argue that those bits were preserved in the successor theory while only the dispensable ornaments were discarded. It is clever. It may even be correct in certain cases. But it has the distinct flavour of a retrospective rationalisation—of someone examining the ruins after the earthquake and declaring that the parts still standing were the only parts that mattered.
The Fragility of “Literal Truth”
Here we arrive at what I consider the deepest and most neglected problem in this entire debate. Both the realist and the instrumentalist conduct their argument as though we possess a clear, stable, and philosophically robust understanding of what it would mean for a scientific theory to be “literally true.” We do not.
This is not a minor technicality. It is a chasm.
The phrase “literal truth” has been the subject of philosophical investigation for millennia, and the results of that investigation should give every confident realist pause. Even in legal contexts—where the practical consequences of lying are immediate and severe—the meaning of “literal truth” proves remarkably elusive. The distinction between a statement that is literally true and one that is practically true—that is, one that conveys a genuine insight without being exactly correct—is far harder to draw than common sense suggests.
The difficulty compounds when we consider the nature of language itself. The truth conditions of natural language sentences are not fixed independently of context. There is an irreducible pragmatic dimension to the determination of what any sentence “literally” says. And the specialised languages of science—riddled as they are with approximations, idealisations, metaphors, and mathematical abstractions—are not exempt from this problem. They are, if anything, more deeply entangled in it.
Consider: when a physicist says that an electron “orbits” the nucleus, this is not literally true in any ordinary sense of the word “orbit.” When a biologist speaks of a gene “coding for” a trait, the metaphor of a code is doing real conceptual work that a literal description would struggle to replicate. When an economist models rational agents maximising utility, everyone involved understands that no actual human being has ever behaved in the manner described. Scientific discourse is saturated with approximation and metaphor not as a deficiency but as a structural necessity. The map is not the territory, and the language of science is not the language of reality—if such a language exists at all.
There is experimental evidence, moreover, that human beings systematically prefer pragmatically useful statements over literally accurate ones. We are, it seems, cognitive pragmatists by nature—creatures whose mental architecture is oriented not toward the apprehension of literal truth but toward the construction of models that work well enough for the purposes at hand. If this is so, then the entire scientific enterprise may be better understood not as a truth-tracking mechanism but as the most sophisticated prediction-generating apparatus our species has yet devised. A glorious instrument. But an instrument nonetheless.
The Problem of Many Sciences
There is a further difficulty that the debate’s usual framing obscures. We speak of “science” as though it were a single enterprise with a unified method and a common epistemic standard. It is nothing of the kind.
The methodological practices, explanatory standards, and evidential requirements of fundamental physics bear almost no resemblance to those of evolutionary biology, which in turn share little with clinical psychology or macroeconomics. The physicist can isolate variables, repeat experiments, and achieve predictions of breathtaking precision. The economist cannot. The molecular biologist operates in a domain of staggering complexity where causal pathways are tangled beyond any hope of complete enumeration. The sociologist confronts systems in which the objects of study are themselves interpreting, reacting to, and modifying the theories constructed about them.
To ask whether “science” delivers literal truth is therefore to ask a question that is ill-formed until we specify which science, which domain, and which level of description. It is entirely possible—indeed, I think it is probable—that something resembling realism provides a defensible account of certain domains of mature physics, while instrumentalism offers the more honest characterisation of the biological and social sciences. The idea that a single epistemological verdict can be returned across the entirety of human inquiry is not philosophical sophistication. It is a refusal to look at the evidence.
The distinction between correlation and causation—that elementary caveat drilled into every statistics student—is itself far more treacherous than the textbook version suggests. The criteria we use to assess causal claims are not algorithmic; they are heuristic, fallible, and shot through with judgment calls that reasonable minds may resolve differently. In domains where controlled experimentation is impossible—which is to say, in most of the domains that matter most to human life—the gap between “this model works” and “this model is true” may be unbridgeable in principle, not merely in practice.
We are, after all, not gods peering down at the machinery of the universe from some Archimedean vantage point. We are participants in the system we are attempting to describe. Our instruments are part of the causal order they seek to illuminate. Our observations alter the phenomena we observe. Our theories shape the very questions we think to ask. The idea that such creatures—embedded, partial, historically situated—could nonetheless produce descriptions that correspond literally to the way things are in themselves is not a modest claim. It is an extraordinary one. And extraordinary claims, as someone or other once observed, require extraordinary evidence. The evidence we have is extraordinary—but it is evidence of predictive power, not of metaphysical correspondence, and it is a fatal error to conflate the two.
The Quantum Embarrassment
And then there is quantum mechanics.
Here is the most empirically successful theory in the history of science. Its predictions have been confirmed to a degree of precision that borders on the absurd. The technological civilisation of the twenty-first century—from semiconductors to lasers to magnetic resonance imaging—is built upon its foundations. If any theory has earned the right to be called “true,” surely it is this one.
And yet. The physics community cannot agree on what quantum mechanics means. The Copenhagen interpretation holds that physical properties do not exist in a determinate state until they are observed—a position that makes the act of measurement constitutive of reality rather than revelatory of it. The many-worlds interpretation holds that every quantum event causes the universe to split into branching copies—a metaphysical extravagance that makes the multiverse a necessary feature of reality. Bohmian mechanics restores determinism at the cost of non-locality. Decoherence-based approaches attempt to dissolve the measurement problem without solving it.
These interpretations are mutually incompatible. They describe fundamentally different realities. And yet they are empirically equivalent—they generate identical predictions for every conceivable experiment. There is no observation, no measurement, no test that can distinguish between them.
The working physicist’s response to this interpretive embarrassment has been immortalised in the phrase “shut up and calculate.” Use the equations. Make your predictions. Build your technologies. And for the love of all that is practical, stop asking what it all means.
This is instrumentalism, pure and unashamed, operating at the very foundation of our most fundamental physical theory. And if the scientists who understand quantum mechanics best have adopted the instrumentalist stance not out of philosophical conviction but out of sheer necessity—because the theory itself refuses to yield a unique ontological interpretation—then the realist’s position is not merely weakened. It is, at the deepest level of physical reality, without a leg to stand on.
Recent work has shown that the logical structure of quantum mechanics exhibits features of self-reference analogous to the liar paradox and the halting problem—those famous demonstrations that certain systems cannot, even in principle, fully describe themselves from within. If the universe at its most fundamental level is characterised by this kind of irreducible self-reference, then the aspiration to construct a literally true description of reality from within that reality is not merely unfulfilled. It may be incoherent.
Turtles All the Way Down
There is an old anecdote—attributed to various sources and almost certainly apocryphal—about a philosopher who, after lecturing on the structure of the solar system, is confronted by an elderly woman who insists that the world rests upon the back of a giant turtle. “And what,” asks the philosopher, “does the turtle stand on?” “Another turtle,” replies the woman. “And that turtle?” “It’s turtles all the way down.”
The story is usually told as a joke about infinite regress and the absurdity of certain metaphysical positions. But I wish to suggest that the joke is on us. For the epistemic situation of science is, in a precise and non-trivial sense, turtles all the way down.
Each scientific theory rests upon prior theoretical assumptions, which rest upon further assumptions, which rest upon still further ones—and at no point in this chain do we arrive at bedrock, at a foundation that is self-evidently and literally true. We arrive instead at another theory, another model, another instrument. The instruments become more refined with each generation—more predictively powerful, more technologically productive, more mathematically elegant. But refinement is not the same as truth, any more than a better map is the same as the territory it represents.
This is not a counsel of despair. It is a counsel of intellectual honesty—which is, when one thinks about it, a far rarer and more valuable commodity than optimism. To acknowledge that science provides us with instruments of extraordinary power and precision, without pretending that we have thereby secured correspondence with ultimate reality, is not to diminish science. It is to see it clearly. And there is a kind of dignity in seeing clearly, even when—especially when—what one sees is the limit of one’s own vision.
The scientific method has given us more mastery over nature than any other human invention. It has extended our lives, illuminated our origins, and expanded the boundaries of the possible in ways that would have seemed miraculous to every previous generation. These achievements are real. They are magnificent. They are ours.
But they do not settle the question of truth. It may well be that a particular scientific theory is fundamentally aligned with the way things are. The problem—and it is a problem that no quantity of empirical success can by itself resolve—is that we cannot know. The instruments work. The turtles hold. But no one has ever seen the bottom.
And perhaps—here is the thought that the honest mind must eventually entertain—there is no bottom to see.
This should not paralyse us. It should liberate us. The person who understands the nature of their tools uses them more skillfully, not less. The person who knows that every map is a simplification navigates with greater care, greater alertness, greater intelligence than the person who has confused the map with the landscape. The scientist who grasps the instrumental character of theory is not a lesser scientist but a more rigorous one—freed from the dogmatism that comes with mistaking one’s models for revelations.
The great temptation of every age is to believe that it has arrived—that its theories are not merely better than those that came before, but final. Every age that has succumbed to this temptation has eventually been embarrassed by its successor. There is no reason to suppose that our age will prove the exception. What we can say, with confidence and without apology, is that the instruments are extraordinary. The predictions are magnificent. The technologies are, by any historical standard, miraculous. That these achievements rest upon a foundation whose ultimate nature we cannot ascertain is not a scandal. It is simply the human condition, honestly described.
The turtles hold. They have always held. And that, for creatures such as ourselves—finite, embedded, incorrigibly curious—may have to be enough.
If you found this piece worth your time, consider subscribing. I write about the ideas that matter—which is to say, the ideas that most people would prefer not to think about.