Chapter 14: Theorem or Pattern?

Thesis: This is where the book faces its central question. Is the convergence across fields a law, something that forces any bounded actor facing the unverifiable into these moves? Or is it only a strong empirical pattern: one we keep seeing but have not proved must hold, and whose rhyme a selection effect may explain?

This is the chapter where the book has to answer for itself. It puts on the table the question I have been carrying all along.

Is the convergence across fields a law? That would mean something forces any bounded actor facing the unverifiable into these moves, of necessity. Or is it only a very strong empirical pattern? That would mean we keep seeing it but have not managed to prove it could not be otherwise, and a selection effect alone might be enough to explain the rhyme.

I will make both cases as strong as I can, including the one against me. If this chapter leans at all, it should lean toward doubt.

The Case for a Law

The first argument is the lever decomposition from the previous chapter. If the eight moves really do occupy every position the risk decomposition leaves open to attack, then the convergence is not a coincidence. Structure forces it: anyone who handles the problem well will sooner or later rediscover them, because there is no other choice. If this argument holds, it carries great weight.

The second is independent rediscovery. In the sociology of science, Merton studied "multiple discovery": the same idea is often reached, almost at the same time and independently, by people who knew nothing of one another¹². Newton and Leibniz both arrived at the calculus. Darwin and Wallace both arrived at natural selection. Bell and Gray went to the patent office with the telephone on the same day. The history of science is crowded with such cases, too many to look like chance. The eight moves in this book were likewise reinvented again and again in cryptography, statistics, number theory, organization theory, and security engineering, fields that at the time were not talking to one another much. When something keeps being found independently, it smells more of necessity than of borrowing.

The third comes from a solid precedent inside science. In physics, the renormalization group and universality classes show that systems with wildly different structures converge on the same behavior near a critical point, and there it really is a theorem¹⁷¹⁸. One strikingly concrete example: the behavior of a liquid near its critical point and the behavior of a magnet near its Curie point are described by the same set of "critical exponents." Molecules and magnetic moments have nothing to do with each other, yet they fall into the same universality class. What governs critical behavior is coarse-grained features like symmetry and dimension, not microscopic detail. Wimsatt's robustness analysis holds that a conclusion reached again and again by many independent routes is more likely to be true²⁰. Whewell's 1840 "consilience of inductions"¹¹ and Wigner's essay "The Unreasonable Effectiveness of Mathematics"²² both speak to the credibility that such convergence across fields confers. Convergence is one of science's oldest signs that something is real.

The Case for a Pattern, or Something Weaker

Now for the case against me. I think it weighs at least as much.

The most damaging objection is that those fields are not nearly as independent as they look. They share a mathematical foundation: probability, optimization, and information theory run through the roots of every one of them. They share human cognition, since the same kind of brain built all these disciplines. And they borrow from and cite one another. They were never sealed off. Shannon's information theory flowed into almost everything, and the language of decision theory spread everywhere. Suppose the convergence exists only because everyone draws on the same mathematical toolbox, is shaped by the same kind of mind, and has been imitating everyone else all along. Then "independent rediscovery" counts for much less, and the apparent convergence may be only the echo of a common origin.

Second, the lever decomposition may well be a frame built after the fact. Perhaps it is just flexible enough to fit many different sets of moves. The table in the previous chapter already gave this away. Redundancy was filed under lowering joint failure, yet it also looked like a special case of screening, and that overlap shows the ruler itself is not rigid enough. Dennett asked a key question: when does a pattern count as real, rather than imposed¹³? His criterion is predictive and compressive power. A real pattern lets you predict something new. So far, my framework has mostly organized moves we already knew. It has not predicted a move never seen before and later confirmed. That is a test it has not yet passed.

Third, the selection effect. I went into these fields looking for convergence, so I may well have filtered out, without noticing, the fields and counterexamples that did not fit, and kept only what rhymed. The replication crisis is the loudest warning of all: an empirical pattern that looks thoroughly robust may be nothing more than the product of systematic bias²⁹³⁰. I have no reason to think I am immune to such bias.

Fourth, even if the convergence is real, it need not point to a deep law. Laudan's pessimistic meta-induction reminds us that one successful theory after another was later overturned⁶. Converging on a pattern is not the same as converging on the truth. Cartwright argues that the fundamental laws do not actually describe the world truthfully¹⁴, and Anderson's "more is different" points out that higher levels have laws of their own¹⁵. Perhaps this book's convergence is only a regularity at the scale of a "special science," not a fundamental law. Worrall's structural realism offers a middle path⁷: perhaps what really survives is the structure (the levers themselves), even if the gloss I wrap around it is wrong.

What It Would Take to Settle It

Really closing this question would take something I cannot supply. You would need a formal model of "a bounded actor plus an unverifiable system," and a theorem proving that, under that model, the optimal (or the only) strategy is exactly these few levers. As far as I know, no such model exists yet. The closest real theorem is no free lunch²⁴, and of all things it points the other way: no method is best everywhere. Until that model and that theorem appear, the question is open, and I do not intend to pretend it is closed.

So let me say plainly what this book offers. It is a conjecture, a strong and useful one with clearly drawn limits, together with a shared vocabulary that ties many fields together. It is not a theorem. This is the stance the book has recommended all along: a calibrated belief, not a yes-or-no verdict.

A Book That Cannot Verify Itself

Here, something that has run beneath the whole book since the preface comes to the surface. A book about how to act when you cannot verify cannot verify its own central claim.

It sounds like an awkward bit of self-reference, but it is the book's most honest moment. Faced with a central claim it cannot verify, the book has no choice but to do what it has been describing all along. It states a calibrated belief (I think the convergence is real, but I cannot prove it). It marks where the claim stops (this is a conjecture, not a theorem). It openly invites refutation: go and find a field that does not converge, or a counterexample where a single move breaks the decomposition. That would be its black swan. And then it keeps talking anyway, and keeps handing you this vocabulary, because the vocabulary is useful even if it is unproven.

In other words, the book has run its own moves on itself. It used proxy substitution, replacing "prove the convergence" with "exhibit and organize the convergence." It used calibration, putting a price on its own confidence. And it used a falsificationist audit trail, writing down in black and white a claim that can be overturned. We should be used to this by now: a self-referential system cannot fully justify itself from within²⁶²⁷²⁵. But being unable to prove itself from within does not mean it cannot act. If the book's thesis is right, the fact that it is written the way it describes is not a flaw. It is a small but fitting piece of supporting evidence.

That leaves one last question. If verification is usually out of reach, and even this book can offer only an unproven belief, what is the great mass of things we ordinarily call "knowledge"? The next chapter brings the argument to rest in epistemology.

Next chapter: 15. Knowledge Without Verification →← 13. The Eight Levers

References

Waypoints: 1. historical scientific judgment; 2. theoretically studied material; 3. how science progresses; 4. how to live in an unverifiable world. This section was checked source by source.

T. S. Kuhn (1962). The Structure of Scientific Revolutions. University of Chicago Press. Google Books [3][1] Kuhn argues that science does not advance by linear accumulation but alternates between normal science (solving puzzles within an existing paradigm) and scientific revolution (the replacement of paradigms), with "incommensurability" lying between old and new paradigms. The first edition appeared in 1962, also published as a separate volume of the International Encyclopedia of Unified Science. It is a founding text for understanding "how science progresses," and it prompts this chapter's very question: is cross-domain convergence shaped by a single paradigm, or forced out independently?
K. R. Popper (1959). The Logic of Scientific Discovery. Hutchinson. Google Books [3][1] Popper systematically advances falsificationism: a scientific theory cannot be empirically verified, only falsified, so falsifiability becomes the boundary between science and non-science and the criterion of scientific progress. This book is the expanded English edition of the German original Logik der Forschung (1934, copyright page marked 1935), published by Hutchinson of London in 1959. It is the common starting point for the later debates of Kuhn, Lakatos, and Feyerabend, and it grounds this book's posture of openly inviting refutation and writing down assertions that can be overturned.
I. Lakatos (1970). "Falsification and the Methodology of Scientific Research Programmes." In I. Lakatos and A. Musgrave (eds.), Criticism and the Growth of Knowledge. Cambridge University Press. doi:10.1017/cbo9781139171434.009 [3] Lakatos reconciles Popper and Kuhn by proposing that the unit of evaluation is not a single theory but a "research programme": a programme has a protected hard core and an adjustable protective belt, and if it keeps predicting and delivering new facts it is progressive, otherwise degenerating. The paper grew out of a 1965 colloquium at Bedford College, London, with the collection published by Cambridge University Press in 1970. Its "progressive/degenerating" criterion provides exactly an operational methodological frame for this chapter's "theorem or pattern" dispute.
P. Feyerabend (1975). Against Method: Outline of an Anarchistic Theory of Knowledge. New Left Books. Google Books [3][1] Feyerabend, with his "epistemological anarchism," opposes any universal, fixed scientific method, maintaining that in the actual history of science "anything goes," and noting that major breakthroughs often came precisely from breaking established methodological rules. The book was first published in 1975 by New Left Books of London (later Verso). It forms the strongest opposing stance to whether cross-domain convergence is forced out by some methodological theorem: if there is no unified method at all, convergence is harder still to explain as necessary.
L. Laudan (1977). Progress and Its Problems: Towards a Theory of Scientific Growth. University of California Press. Google Books [3] Laudan argues that the measure of scientific progress is not approach to truth but "problem-solving effectiveness": how many empirical problems a theory solves and how many conceptual difficulties it creates. This frees the judgment of scientific progress from the metaphysical burden of truth and grounds it in comparable functional indicators. It bears directly on "how science progresses," and supports this chapter's restrained stance: converging on a useful pattern need not amount to converging on truth.
L. Laudan (1981). "A Confutation of Convergent Realism." Philosophy of Science, 48(1). doi:10.1086/288975 [3][2] Laudan, with a historical list, advances the "pessimistic meta-induction": many historically successful theories that made accurate predictions had core terms later judged to refer to nothing at all (such as ether and phlogiston), so empirical success does not reliably guarantee a theory's truth. The paper appears in volume 48, pages 19 to 49. It directly challenges the theorem-like claim that "cross-domain convergence points to truth," and is one of this chapter's most crucial opposing sources.
J. Worrall (1989). "Structural Realism: The Best of Both Worlds?" Dialectica, 43(1-2). doi:10.1111/j.1746-8361.1989.tb00933.x [3][2] Worrall proposes structural realism, seeking a middle path between the "no-miracles argument" and the "pessimistic meta-induction": what survives across theory change is not the description of the ontology but the mathematical structure (as Fresnel's optical equations still held after the ether was discarded). The paper appears in volume 43, issues 1 to 2, pages 99 to 124. It offers this chapter a middle reading: even if the gloss I wrap around the levers is wrong, what truly survives may be that structure itself.
I. Hacking (1983). Representing and Intervening: Introductory Topics in the Philosophy of Natural Science. Cambridge University Press. doi:10.1017/cbo9780511814563 [3][2] Hacking shifts the center of gravity in the philosophy of science from "representing" (how theory describes the world) to "intervening" (how experiment manipulates the world), proposing an experimental realism: if we can stably use an entity to intervene and to produce other phenomena (if you can spray it, it is real), we have reason to believe it exists. The book was published by Cambridge University Press in 1983. It offers this chapter another explanation for where the convergent pattern comes from: convergence may stem from shared experimental practice rather than from theoretical necessity.
W. V. O. Quine (1951). "Two Dogmas of Empiricism." The Philosophical Review, 60(1). doi:10.2307/2181906 [2][3] Quine attacks the two dogmas of logical empiricism: the sharp divide between the analytic and the synthetic, and reductionism. He holds that knowledge is a "web of belief" tested against experience as a whole, and that no single statement can be verified or falsified in isolation. The paper appears in volume 60, issue 1, pages 20 to 43. His confirmation holism is the philosophical bedrock of this book's central condition: no single claim can be lifted out and verified completely on its own.
M. Polanyi (1958). Personal Knowledge: Towards a Post-Critical Philosophy. University of Chicago Press. Google Books [1][4] Polanyi proposes "tacit knowledge": we know far more than we can tell, and all explicit knowledge rests on a layer of personal judgment and skill that cannot be fully formalized. Scientific cognition therefore cannot do without the scientist's personal participation and commitment. The book was published by the University of Chicago Press in 1958. It bears directly on this book's landing point: when verification cannot be exhausted, how scientists form a calibrated belief by judgment and act upon it.
W. Whewell (1840). The Philosophy of the Inductive Sciences, Founded Upon Their History. John W. Parker. Google Books [3][2] Whewell here proposes the "consilience of inductions": when a theory induced from one class of facts turns out to explain another, originally unrelated class of facts, this unexpected convergence is a strong mark of the theory's truth. The work is in two volumes, published by John W. Parker of London in 1840. It is the earliest methodological statement of the idea that "cross-domain convergence is credible," providing the historical source for this chapter's side for "law."
R. K. Merton (1961). "Singletons and Multiples in Scientific Discovery: A Chapter in the Sociology of Science." Proceedings of the American Philosophical Society, 105(5). link [1][3] Merton systematically examines the phenomenon of "multiple discovery" in the history of science: the same discovery is often made, almost simultaneously and independently, by people who knew nothing of one another, from which he argues that such multiples are not the exception but the norm of scientific discovery, with discovery depending more on the state of accumulated knowledge than on individual genius. The paper appears in volume 105, issue 5, pages 470 to 486. It is the key sociological evidence that "convergence is a strong empirical pattern," and this chapter draws on it precisely to weigh the force of independent rediscovery.
D. C. Dennett (1991). "Real Patterns." The Journal of Philosophy, 88(1). doi:10.2307/2027085 [2][3] Dennett asks: when does a pattern count as "real," rather than imposed by the observer? His criterion is compression and prediction: if describing the data as a certain pattern yields genuine information compression and supports predictions about new cases, that pattern is real. The paper appears in volume 88, issue 1, pages 27 to 51. This is the core conceptual tool of this chapter's "theorem or strong empirical pattern," and on its basis the chapter concedes that the lever decomposition has not yet predicted a move never before seen and later confirmed, a test it has still to pass.
N. Cartwright (1983). How the Laws of Physics Lie. Oxford University Press. doi:10.1093/0198247044.001.0001 [2][3] Cartwright argues that the fundamental laws of physics are universal precisely because they do not describe the real world truthfully: the more fundamental a law, the more idealization and approximation it needs to fit phenomena, while what truly describes concrete systems are local, phenomenological laws. The book was first published by Clarendon Press / Oxford University Press in 1983. It challenges at the root whether, behind the convergence this chapter sees, a "theorem" truly stands, or merely a tidiness at the level of models.
P. W. Anderson (1972). "More Is Different." Science, 177(4047). doi:10.1126/science.177.4047.393 [2][3] Anderson opposes the reductionist "constructionist" inference: even if everything is composed of fundamental particles obeying fundamental laws, it does not follow that higher-level behavior can be derived from those laws. With each step up in scale, wholly new, self-contained regularities emerge. The paper appears in volume 177, issue 4047, pages 393 to 396 (August 4, 1972). It supports a weakened reading of this chapter: the book's convergence may be only a regularity at the scale of some "special science," not a fundamental law.
H. A. Simon (1962). "The Architecture of Complexity." Proceedings of the American Philosophical Society, 106(6). link [2][3] Simon argues that complex systems that evolve stably and endure tend to be hierarchical and "nearly decomposable," with interactions within a subsystem far stronger than those between subsystems, and uses the parable of the watchmakers to show that systems with stable intermediate parts are more easily assembled. The paper appears in volume 106, issue 6, pages 467 to 482. It gives cross-domain convergence yet another "common origin" explanation: different fields may stumble on similar structures because complex systems are subject to the same set of architectural constraints.
K. G. Wilson (1979). "Problems in Physics with Many Scales of Length." Scientific American, 241(2). doi:10.1038/scientificamerican0879-158 [2][3] Wilson explains the renormalization group to a general readership: when a system spans many length scales (as near a critical point), it can be handled by coarse-graining step by step, thereby explaining why systems with wildly different microscopic details fall into the same "universality class" and exhibit exactly the same critical behavior. The paper appears in volume 241, issue 2, pages 158 to 179 (August 1979). It is the hardest physical evidence for this chapter's side for "law," because there the convergence of different systems to the same behavior is a provable theorem.
R. W. Batterman (2002). The Devil in the Details: Asymptotic Reasoning in Explanation, Reduction, and Emergence. Oxford University Press. doi:10.1093/0195146476.001.0001 [2][3] Batterman analyzes "asymptotic reasoning" in physics: many explanations (especially of universality phenomena) depend on the singular behavior that emerges when some limit is taken (such as a scale going to zero or infinity), and such explanations cannot be simply reduced to the underlying theory, residing precisely where the details vanish. The book was published by Oxford University Press (cover year usually marked 2002, some reviews dating it 2001 from the pre-publication catalog). It offers philosophical analysis for this chapter's side for "law": cross-domain convergence may hold precisely because of this asymptotic universality, independent of microscopic detail.
R. Levins (1968). Evolution in Changing Environments: Some Theoretical Explorations. Princeton University Press. doi:10.1515/9780691209418 [2][3] Levins, through a series of theoretical models, explores how organisms evolve in fluctuating, uncertain environments, introducing tools such as the "fitness set" to analyze which strategy is optimal under variable conditions, throughout a modeling style that approximates complex reality with multiple simplified models. The book was published by Princeton University Press in 1968 (Monographs in Population Biology, no. 2). The multi-model, approximate modeling it represents is a biological forerunner of Wimsatt's robustness analysis, echoing this chapter's emphasis on "the convergence of many independent routes."
W. C. Wimsatt (1981). "Robustness, Reliability, and Overdetermination." In M. B. Brewer and B. E. Collins (eds.), Scientific Inquiry and the Social Sciences. Jossey-Bass. Google Books [3][2] Wimsatt systematically expounds "robustness analysis": if a conclusion can be reached again and again by many mutually independent routes of detection, derivation, or measurement, then it is more likely to be true rather than an artifact of one means, and this "overdetermination" is the key to telling real things from artifacts. The paper appears in that collection, pages 125 to 163. It is the methodological core of this chapter's "why cross-domain convergence is credible," and just the argument on which the side for "law" leans.
W. C. Wimsatt (2007). Re-Engineering Philosophy for Limited Beings: Piecewise Approximations to Reality. Harvard University Press. doi:10.2307/j.ctv1pncnrh [3][4] Wimsatt argues for remaking the philosophy of science into a tool fit for "limited beings": real knowers have bounded computation, are error-prone, and are trapped in their own perspective, and so rely on heuristics, approximation, and piecewise approximation to inch toward reality rather than pursuing idealized complete rationality. The book was published by Harvard University Press in 2007. It is highly attuned to this book's theme, answering head-on how a bounded actor advances science in an unverifiable world and lives accordingly.
E. P. Wigner (1960). "The Unreasonable Effectiveness of Mathematics in the Natural Sciences." Communications on Pure and Applied Mathematics, 13(1). doi:10.1002/cpa.3160130102 [2][3] Wigner marvels at a fact: mathematics developed for purely internal motives turns out, again and again, to describe natural laws with precision, a fit he calls "a wonderful gift which we neither understand nor deserve." The paper appears in volume 13, issue 1, pages 1 to 14, from the 1959 Courant Lecture. It is the prototype of the "theorem or pattern" question: is the cross-domain effectiveness of mathematics a necessity, or a vast rhyme we cannot yet explain?
H. Putnam (1975). Mathematics, Matter and Method: Philosophical Papers, Volume 1. Cambridge University Press. Google Books [2][3] This collection gathers Putnam's early papers on the philosophy of mathematics and scientific realism, among them "What is Mathematical Truth?", which gives the classic statement of the "no-miracles argument": realism is the only philosophy that does not make the success of science a miracle, for if the entities in a theory did not exist, the success of its predictions would be inexplicable. The paper appears at pages 60 to 78. It is a positive weapon for this chapter's side for "law," set squarely against Laudan's pessimistic meta-induction.
D. H. Wolpert and W. G. Macready (1997). "No Free Lunch Theorems for Optimization." IEEE Transactions on Evolutionary Computation, 1(1). doi:10.1109/4235.585893 [2][3] Wolpert and Macready prove the "no free lunch" theorem: averaged over all possible objective functions, any two optimization algorithms have exactly the same expected performance, so there is no universally dominant algorithm, and any algorithm's advantage is bought with specific assumptions about problem structure. The paper appears in volume 1, issue 1, pages 67 to 82. It is a real theorem this chapter takes as a contrast, and one that points, of all directions, against me, reminding us that conclusions of the "unique optimal strategy" kind require very strong premises.
D. R. Hofstadter (1979). Gödel, Escher, Bach: An Eternal Golden Braid. Basic Books. Google Books [2][4] Hofstadter, through Gödel's incompleteness, Escher's visual paradoxes, and Bach's canons, weaves a common motif: self-reference and "strange loops," and from it explores how mind and meaning emerge from meaningless formal hierarchies. The book was published by Basic Books in 1979 and won the Pulitzer Prize for nonfiction. It renders self-reference and recursive closure as an art, echoing this chapter's moment when a book cannot prove itself from within yet still writes itself "in the way it describes."
E. Nagel and J. R. Newman (1958). Gödel's Proof. New York University Press. Google Books [2][4] Nagel and Newman, with as little technical detail as possible, make clear to the general reader the proof strategy of Gödel's first and second incompleteness theorems: any sufficiently strong consistent formal system contains true propositions it cannot prove, and cannot prove its own consistency from within. The book was published by New York University Press in 1958. It provides a readable and accurate basis for this chapter's account of the "intrinsic limits of formal systems" and of how "a self-referential system cannot prove itself from within."
G. J. Chaitin (1982). "Gödel's Theorem and Information." International Journal of Theoretical Physics, 21(12). doi:10.1007/bf02084159 [2][3] Chaitin reinterprets incompleteness through algorithmic information theory: the information content implied by a formal system's axioms is finite, so it cannot prove any proposition whose complexity exceeds that content (such as that a sufficiently long random bit string is indeed random), and incompleteness is thereby reduced to an information ceiling. The paper appears in volume 21, issue 12, pages 941 to 954. It pins down the intrinsic limits of formal systems from the angle of information, providing another formal ground for the "unverifiable world."
M. Mitchell (2009). Complexity: A Guided Tour. Oxford University Press. doi:10.1093/oso/9780195124415.001.0001 [2][3] Mitchell surveys the science of complex systems for the general reader: from information, computation, and evolution to networks, introducing themes such as emergence and self-organization that recur across biological, computational, and social systems, and frankly admitting that the field still lacks a unified theory. The book was published by Oxford University Press in 2009. It provides a reliable introductory map to the state of research on cross-domain common patterns, and reminds us that these patterns remain, to this day, mostly empirical observation rather than theorem.
J. P. A. Ioannidis (2005). "Why Most Published Research Findings Are False." PLoS Medicine, 2(8). doi:10.1371/journal.pmed.0020124 [3][4] Ioannidis uses a statistical model to argue that under conditions of small samples, small effects, large researcher degrees of freedom, and pervasive interests and bias, the probability that a published "positive" finding is true often falls below one half. The paper appears in volume 2, issue 8, e124 (August 2005). It is a classic alarm of metascience, reminding this chapter that a seemingly robust empirical pattern may be only the product of systematic bias, against which the author has no reason to assume immunity.
Open Science Collaboration (2015). "Estimating the Reproducibility of Psychological Science." Science, 349(6251). doi:10.1126/science.aac4716 [3][4] The Open Science Collaboration, coordinating over a hundred researchers, replicated one hundred published psychological studies, with only about a third of the replications yielding an effect significant and in the same direction as the original, and replication effect sizes generally smaller than originally reported. The paper appears in volume 349, issue 6251, article number aac4716. It turns the "replication crisis" from worry into quantifiable fact, providing this chapter with direct and weighty empirical data on whether a strong empirical pattern is truly robust.