Chapter 17 - Information, Entropy, and the Shape of Memory

In 1948, a Bell Labs mathematician named Claude Shannon published a paper that quietly reorganized how an entire century of subsequent technology would be built, by doing something deceptively simple: he showed that "information" could be defined precisely, mathematically, and completely independently of meaning, of physical medium, and of the specific content being communicated. A string of coin flips and a symphony, reduced to their information-theoretic essentials, are just sequences of symbols with certain statistical properties, and Shannon's mathematics could quantify exactly how much "information," in a rigorous technical sense, either one contained, how much could be compressed out of them, and how reliably they could be transmitted across a noisy channel. This chapter asks whether Shannon's framework - and its descendants in modern information theory - offers a more promising way to think about morphic resonance than the physical-field language we've been using so far, one that sidesteps some of the specific problems Chapter 16 identified, while being honest about where it introduces new problems of its own. By the end of this chapter, we will have arrived at the single most important reframing in this whole book - the specific, narrowed version of the hypothesis that the rest of the book is actually built to test.

What Information Actually Means, Precisely

Let's build the core concept in plain language before formalizing it. Shannon defined information in terms of surprise, or more precisely, in terms of the reduction of uncertainty. If I tell you the sun rose this morning, I have given you almost no information, because you were already almost completely certain this would happen - the message resolves almost no uncertainty. If I tell you the result of a coin flip you haven't seen, I've given you exactly one "bit" of information (in the technical sense that underlies the word "bit" in computing), because there were exactly two equally likely possibilities beforehand, and my message has narrowed that down to one. The formal measure Shannon developed, called information entropy, quantifies the average amount of uncertainty (or "surprise") inherent in a probability distribution - mathematically, for a set of possible outcomes with probabilities $p_1, p_2, \ldots, p_n$, entropy is defined as $H = -\sum_i p_i \log_2 p_i$, measured in bits. A fair coin flip has maximal entropy of exactly 1 bit, because both outcomes are equally likely and equally surprising; a heavily loaded coin that comes up heads 99% of the time has much lower entropy, because most outcomes are highly predictable and carry little informational surprise when they occur as expected.

A closely related and, for our purposes, even more useful concept is mutual information: a precise measure of how much knowing the value of one variable reduces your uncertainty about a second variable. If two variables are completely independent - knowing one tells you nothing whatsoever about the other - their mutual information is exactly zero. If two variables are perfectly correlated, their mutual information is maximal. Crucially, and this is the point that makes mutual information such a useful concept for evaluating claims like morphic resonance, mutual information can be calculated and is mathematically meaningful regardless of whether the correlation between the two variables arises through any known physical causal channel, through a shared common cause, through pure statistical coincidence in a finite dataset, or (hypothetically) through some genuinely mysterious non-local mechanism. Mutual information is a purely statistical, pattern-based measure; it describes the presence and strength of a correlation without, on its own, telling you anything about the mechanism producing that correlation.

Why This Reframing Is Genuinely Useful

This distinction matters enormously for thinking about morphic resonance research in a modern, computationally sophisticated way, and it points toward the single most productive reframing this book can offer: rather than asking "is there a physical field transmitting morphic resonance, and if so, what carries it," which runs immediately into the serious physics problems of Chapter 16, we can instead ask a logically prior, more modest, and more directly testable question:

Is there detectable mutual information between the developmental, behavioral, or organizational history of spatially and causally separated systems, beyond what known confounds - shared genetics, shared environment, shared culture, coincidence, publication bias - can explain?

Hold onto that sentence. It is the precise, narrowed, mechanism-agnostic target that every research protocol in Part IV is aimed at, and it is worth being explicit about exactly what work it is doing. Notice everything it does not claim: it does not assert a non-local mechanism, does not require Sheldrake's morphic fields to be real in any literal sense, and does not commit you to any position on the physics problem of Chapter 16. It asks only whether a specific statistical pattern exists in the world, beyond what conventional explanations already predict. This reframing is valuable for three reasons. First, it is a purely statistical, mechanism-agnostic question, investigable with rigorous statistical and computational tools - developed in full technical detail beginning in Chapter 20 - without first having to resolve the deeply challenging physics question of what could possibly transmit such information if found. Second, it properly separates two questions popular treatments of morphic resonance often collapse together: the empirical question of whether such correlations exist beyond chance and known confounds, and the theoretical question of what would explain them if they do. It is entirely possible to establish strong evidence for the first while remaining agnostic, or even skeptical, about candidate explanations for the second - exactly the position physicists occupied for decades regarding many confirmed but poorly understood phenomena throughout the history of science. Third, framing the research question in terms of detectable mutual information across large datasets is exactly the kind of problem modern machine learning - beginning in Chapters 20 and 21 - is extraordinarily well suited to, arguably far better suited than any statistical toolkit available in 1981, or even in the early 2000s.

Landauer's Principle: Information Has a Physical Cost

Before getting too excited about information theory as an escape route from the hard physics of Chapter 16, it's important to introduce a crucial corrective, one of the most elegant and underappreciated results in twentieth-century physics: Rolf Landauer's 1961 principle, which established that information is not, in fact, a free-floating abstraction fully separable from physics - it has an irreducible physical cost. Specifically, Landauer showed that erasing one bit of information in a physical system must, as a matter of thermodynamic necessity, dissipate a minimum amount of energy as heat, at minimum $k_B T \ln 2$, where $k_B$ is Boltzmann's constant and $T$ is the system's temperature, connecting the abstract world of information theory directly and quantitatively to the material world of thermodynamics. This principle has since been experimentally confirmed, with landmark demonstrations in the 2010s directly measuring the predicted minimum heat dissipation associated with information erasure in carefully engineered microscopic systems.

Landauer's principle matters here because it closes off a tempting but ultimately mistaken escape hatch: the idea that reframing morphic resonance in terms of "pure information" transfer, rather than physical field transmission, somehow exempts the hypothesis from physical constraints altogether, because information is supposedly a more abstract, less physically constrained category than energy or matter. Landauer's result shows this is not correct - information processing, storage, and transmission are always, in every confirmed physical system, tied to physical substrates and subject to thermodynamic costs and constraints. A morphic field's information, if it exists, would still need to be physically instantiated somewhere, in some physical medium, obeying some physical dynamics, even if we remain agnostic about exactly what that medium and dynamics might be - meaning the information-theoretic reframing doesn't dissolve the hard physics question so much as productively re-pose it: not "is there a field" but "what physical substrate, if any, could support the kind of long-range, low-decay mutual information structure morphic resonance requires," a question that remains exactly as unresolved, and exactly as much in tension with known physics per Chapter 16, as before - but now stated with more mathematical precision, which is itself real intellectual progress.

Entropy, Order, and the Direction of Time

There's a second, related information-theoretic concept worth introducing carefully, because it bears directly on one of Sheldrake's own recurring arguments: the relationship between information entropy (the statistical measure of uncertainty we defined above) and thermodynamic entropy (the classical physics measure of disorder, central to the second law of thermodynamics, which states that the total entropy of an isolated system can never decrease over time, giving physics its fundamental arrow of time).

Sheldrake has argued, in some of his more theoretically ambitious writing, that the reliable emergence of increasingly complex, organized biological form over the course of evolution and development represents a kind of local decrease in disorder - organisms are, in an intuitive sense, highly ordered, low-entropy structures relative to the disorganized raw materials they're built from - that seems to run against the universe's overall thermodynamic tendency toward increasing disorder. Morphic resonance, on this reading, might be understood as an organizing principle that helps explain this local entropy decrease.

This argument rests on a fairly well-known and, I think, resolvable misunderstanding, one worth clearing up directly because it's a genuinely common point of confusion, not unique to Sheldrake. The second law of thermodynamics applies to isolated, or more precisely closed, systems: total entropy cannot decrease for a system with no external energy or matter exchange. Living organisms are emphatically not isolated systems. They are open systems, constantly importing energy - as food, or sunlight for photosynthetic organisms - and exporting entropy as waste heat and disordered waste products. This allows local order, the organism's own low-entropy, highly organized structure, to increase substantially, while the total entropy of the organism plus its surrounding environment still increases overall, fully consistent with the second law. No exception or special new principle is required.

This is a mainstream, well-understood, thoroughly unmysterious resolution, developed in detail by physicists and biologists working on non-equilibrium thermodynamics since at least Erwin Schrödinger's influential 1944 book What Is Life? It means the apparent "puzzle" of biological order increasing against a supposedly universal tendency toward disorder does not actually require morphic resonance, or any other novel organizing principle, to resolve. It is fully and elegantly explained by existing, confirmed non-equilibrium thermodynamics. Any version of the morphic resonance argument that leans on this specific "order versus entropy" framing is resting on a misunderstanding of what the second law actually forbids - a place where a more careful reading of established physics removes, rather than strengthens, the case for a non-conventional organizing principle.

Kolmogorov Complexity and Compressibility

One final, genuinely useful information-theoretic concept deserves introduction here, because it will become directly relevant to the machine learning methods of Part IV: Kolmogorov complexity, which measures the complexity of a specific string of data (rather than a probability distribution, as Shannon entropy does) as the length of the shortest possible computer program that could generate that exact string. A highly repetitive, patterned string (like a thousand repetitions of "01") has low Kolmogorov complexity, because a very short program ("repeat '01' one thousand times") can generate it; a string of genuinely random digits has high Kolmogorov complexity, because no shorter description exists than the string itself - you simply have to specify every digit.

This concept is relevant to morphic resonance research because it offers a rigorous, mechanism-agnostic way of asking whether a given pattern (a specific crystal lattice structure, a specific learned behavioral sequence, a specific developmental trajectory) is more "compressible," relative to some reference population of similar patterns, when it has occurred more times before - a question that maps directly onto Sheldrake's core claim that repeated patterns become "easier" (in some operationalizable sense related to how readily a system settles into them) over time. Modern machine learning systems, particularly the large neural network models we'll examine beginning in Chapter 20, are, in a very real technical sense, sophisticated compression engines. A well-trained language model's ability to predict the next element in a sequence is mathematically related to how effectively it has learned to compress the statistical structure of its training data. Comparing how "surprised" - in the precise Shannon-entropy sense - such a model is by a new instance of a pattern, relative to how surprised it would have been by an equally novel but structurally unrelated pattern, gives us a genuinely new, computationally powered instrument for asking Sheldrake's core question in operational, quantitative terms. Chapter 25 develops this specific methodological proposal in full.

Looking Forward

Information theory has given us the book's operative research target - detectable, anomalous mutual information between separated systems, considered independently of mechanism - while Landauer's principle has reminded us that this reframing is not a way of escaping physics altogether, only a way of re-posing the physical question more precisely. We've also cleared up a specific, common misreading of thermodynamic entropy that sometimes gets marshaled, mistakenly, in morphic resonance's favor.

Chapter 18 continues this theoretical deep dive by turning to systems theory and the science of complexity - asking how self-organization, emergence, and attractor dynamics in complex systems (the same toolkit that, as we saw in Chapter 7, successfully explained starling murmurations through purely local rules) might offer yet another, perhaps more promising, non-mystical framework for understanding exactly the kinds of robust, convergent pattern formation that originally motivated Sheldrake's hypothesis - without requiring any appeal to non-local field transmission at all.