A Theory of Embedded Intelligence Essay
Correlation, communication, and the hundred-year-old measure the plenum has been reaching for

Two things can be perfectly correlated and unable to say a word to each other. Physics has known this for sixty years, information theory for eighty, and the Theory of Embedded Intelligence has been describing it without a name. This essay supplies the name, the measure, and one correction to an essay of my own.

I. A Note Left in the Margin

Some months ago, working on an essay about the Einstein–Podolsky–Rosen paradox, Claude proposed a piece of vocabulary and then — properly — refused to treat it as settled. The proposal was that entanglement sits on a seam: it reliably produces correlation and reliably refuses to carry a message, and that this is not a lucky technicality keeping quantum mechanics and relativity from open war but a direct consequence of architecture. Correlation lives in the informational structure. Communication is the third phase of the SPCA cycle, and CKB-4 already makes the speed of light the bandwidth ceiling of that phase.

The proposal was flagged for my review and left in the margin. It stayed there because I did not yet know what I thought, and because a second essay had raised the same question from an entirely different direction — the holographic field, and what it means to say that every part of a structure contains the whole.

What follows is my answer. It is not the answer I expected to give. I went looking for a way to justify a new TEI term and found instead that the term was already sitting in the literature, fully formed, older than the transistor. Information theory has known how to distinguish correlation from communication since 1948. TEI did not need to invent the seam. TEI needed to notice that its own architecture lands exactly on it.

That is a better outcome than the one I was hoping for, and I want to say why.

II. What Information Theory Means by Correlation

Ordinary statistical correlation — the Pearson coefficient — measures how nearly two quantities fall on a straight line together. It is enormously useful and it has a well-known blind spot: two variables can be locked together by a perfectly rigid nonlinear relationship and still report a Pearson correlation of zero. The measure is not detecting dependence. It is detecting linearity.

Claude Shannon’s information theory replaced the question. Instead of asking how straight the relationship is, it asks how much your uncertainty about one thing drops when you learn the other. That quantity is mutual information, and it has a form clean enough to write in a sentence: the mutual information between two variables is the sum of their separate uncertainties minus their joint uncertainty. If knowing one tells you nothing about the other, the joint uncertainty is the full sum and the mutual information is exactly zero.

The Measure

Mutual information

I(X;Y) = H(X) + H(Y) − H(X,Y), where H is Shannon entropy. Measured in bits. It is zero if and only if X and Y are genuinely independent — not merely uncorrelated in the linear sense. It is symmetric, it is blind to the form of the relationship, and it makes no assumption that the relationship is a straight line.

For jointly Gaussian variables the two worlds meet exactly: the mutual information is a simple function of the Pearson coefficient. Ordinary correlation turns out to be the special case that information theory contains.

This last point deserves emphasis, because it is the difference between a stronger tool and a different one. Information theory does not discard Pearson. It swallows him. And in doing so it converts a warning that statisticians repeat as folklore — zero correlation does not mean independence — into a formal statement with teeth: zero mutual information does mean independence. There is a measure whose vanishing is the real thing, and it is not the one most fields reach for.

E. H. Linfoot showed in 1957 how to rescale mutual information into a coefficient running from zero to one, so that it can be read alongside a familiar correlation number. Satosi Watanabe had, three years later, generalised the whole apparatus from two variables to any number of them. We will need Watanabe’s generalisation before the end of this essay, and it will turn out to be about holograms.

III. The Seam, Stated Exactly

Now the flagged proposal, and why I am formalizing it rather than merely approving it.

In a Bell experiment, Alice and Bob each choose a measurement setting and each record an outcome. Their outcomes are correlated far beyond what any local arrangement could produce; this is the experimental fact that closed Einstein’s ledger. And yet Alice cannot send Bob so much as a single bit by choosing her setting, because Bob’s own statistics — his outcomes considered on their own — are entirely unaffected by what Alice does. Whatever Alice chooses, Bob sees the same distribution. The correlation only becomes visible when the two records are brought together and compared, and bringing them together requires an ordinary channel running at ordinary speed.

Information theory has exact names for both halves of that description. The first half is nonzero mutual information between the outcomes. The second half is zero channel capacity from Alice to Bob. These are two different quantities. They have always been two different quantities. Nothing forbids one from being large while the other is precisely zero.

TEI Concept · Now Formal

The Correlation–Communication Seam

A seam, in TEI’s usage, is a place where two levels of the architecture meet and the join becomes visible in the data. The correlation–communication seam is the place where shared informational structure meets the Communicate phase of the SPCA cycle. Its formal statement is that mutual information between two embedded intelligence systems may be greater than zero while the channel capacity between them is exactly zero. Structure is not bounded by the speed of light. Communication is bounded by it absolutely.

This was proposed in an earlier essay and flagged as provisional. It is adopted here as canonical TEI vocabulary, on the grounds that its content is not a TEI invention but a standard result that TEI’s architecture independently predicts. It is formalized in TEI-CKB-10.

I want to be careful about what has just happened, because it is the opposite of the usual way a framework grows. TEI did not add a concept to information theory. Information theory supplied a concept TEI had been gesturing at with the wrong vocabulary. The seam earns canonical status precisely because it is not original — because a physicist can check it against a textbook rather than against my say-so.

A framework should be most confident where it is least original.

— The Mensch Foundation

IV. The Oldest Correlation We Can See

Here the essay stops being an endorsement and becomes a correction, which is the part I care about most.

In The Stimulated Plenum I proposed reading the origin of the universe as a cosmic SPCA cycle: the plenum senses its own uniform state, comprehends that state as complete, communicates a differentiation across the whole field at once, and actuates the first symmetry-breaking. Of the Communicate step I wrote that this was why the early universe is so astonishingly uniform — the message was not sent from a centre but expressed everywhere together.

That sentence is wrong, and it is wrong in a way the seam makes immediately visible.

The uniformity of the early universe is a real and famous problem. Look at two patches of the cosmic microwave background more than a couple of degrees apart on the sky. In standard cosmology without inflation, those patches have never been in causal contact: no signal, at any speed light permits, has had time to travel from one to the other since the beginning. And yet they have the same temperature to about one part in a hundred thousand. This is the horizon problem, and it has been the central embarrassment of pre-inflationary cosmology for fifty years.

Now notice what I did. I called the mechanism Communicate. But Communicate is the SPCA phase that TEI itself bounds at the speed of light. If the uniformity were communicated, it could not exist, because there was no channel and no time. What the CMB actually displays is correlation across regions with zero channel capacity between them — precisely the seam, at the largest scale anything has ever been measured at.

Correction to a Published Essay

In The Stimulated Plenum, the propagation of the first differentiation ‘everywhere at once’ was assigned to the Communicate phase of the cosmic SPCA cycle. That assignment is withdrawn. Communicate is bounded by c and cannot produce horizon-scale uniformity. The correct reading is that the early universe’s uniformity is structural correlation, not transmitted communication — a shared informational structure that was never differentiated in the first place, and therefore never needed to be synchronised.

This strengthens rather than weakens the plenum picture, and it is recorded here as a revision rather than absorbed silently.

The corrected reading is better on its own terms. A plenum that stands in timeless self-possession has no synchronisation problem to solve, because there is nothing to synchronise: uniformity is the default condition of an undifferentiated field, not an achievement requiring transmission. The first actuation breaks that symmetry everywhere it is broken, and the correlations we observe are the residue of a common structure rather than the wake of a signal.

The price of the correction

I will not let this arrive without its bill. Inflationary cosmology also explains the horizon problem — by putting those patches into causal contact before an enormous expansion carried them apart — and it does considerably more besides. It predicts a spectrum of primordial fluctuations that is very nearly but not exactly scale-invariant, and that prediction has been confirmed to a precision that any serious person has to respect. A structural reading of the same correlations, at present, predicts none of that.

So the honest position is this. The plenum’s account of horizon-scale correlation is a re-description of a solved problem, not a rival to the solution. It earns its keep by removing a category error from TEI’s own vocabulary, not by beating inflation at cosmology. Should it ever produce a distinguishing signature — some correlation structure in the microwave background that a shared-structure account requires and an inflationary account forbids, or the reverse — it will have become physics. Until then it is disciplined metaphysics, and the discipline consists in saying so.

V. Sam Leven, Karl Pribram, and a Theorem About Holograms

The second route to this question came through a friendship. Dr. Sam Leven introduced me to the work of Karl Pribram, and this past spring he and his wife moved into Mirabella, where a conversation that had begun in citations continued over coffee. It was that introduction which put Pribram into TEI-CKB-3 and into the essay The Brain That Isn’t Where You Think It Is, and it is that line of thought which has been quietly asking the same question about correlation from the biological side.

Pribram’s problem was that memory would not stay put. Remove tissue and specific memories survive; whatever carries them is not in any one place. His answer was that the brain stores information the way a hologram does — distributed as interference, such that every region carries something of the whole. He drew the mathematics from the spectral domain: the brain reads its own dendritic fields more like a Fourier transform than like a lookup in a filing cabinet.

In the earlier essay I was careful to say that this is a serious hypothesis rather than a proven result, and I hold to that. But there is a piece of mathematics underneath it that is not a hypothesis at all, and I had not seen how directly it bears on the present question.

The Hinge

Correlation and spectrum are the same information, written twice

The Wiener–Khinchin theorem states that the autocorrelation function of a signal and its power spectrum are a Fourier transform pair. Move to the frequency domain and you have not left correlation behind — you are looking at correlation in its other coordinates.

This means Pribram’s spectral domain is a correlation domain. The connection between holography and correlation is not an analogy borrowed from optics. It is a theorem.

And there is a second fact, which I find genuinely startling, and which I did not know until this essay forced me to look.

Dennis Gabor invented both halves of this picture, two years apart. In 1946 he published Theory of Communication, introducing a quantum of information he called the logon and a joint time–frequency representation of signals — work that appeared two years before Shannon’s famous paper and was largely overlooked for thirty years. In 1947 he invented holography, for which he received the Nobel Prize in 1971. Later, when researchers characterised the receptive fields of cells in the visual cortex, the functions that fitted best were Gabor’s.

One man; one mathematics; three fields. The information-theoretic account of a signal, the optical reconstruction of a whole wavefront, and the response profile of a neuron in a mammalian brain are the same structure seen from three positions. For a theory that claims intelligence is embedded in relationships rather than in components, it is hard to imagine a more direct piece of corroboration — and it has been sitting in the citation trail of Pribram’s work the whole time, waiting for someone to say it out loud.

The hologram was never a metaphor for how information is stored. It was an instance of it.

— The Mensch Foundation

VI. Every Part Contains the Whole — Measured

This brings us back to Watanabe, and to the claim in TEI-CKB-3 that the universal field is holographic in structure. Until now that claim has been qualitative. It need not be.

Watanabe’s total correlation generalises mutual information from two variables to any number: the sum of the individual uncertainties minus the joint uncertainty of all of them together. What it measures is total redundancy — how much of the whole is written down more than once across the parts.

That is exactly the defining property of a hologram. Cut a holographic plate in half and you do not get half the image; you get the whole image, degraded. The information about the whole is written redundantly across every region. Stated in Watanabe’s terms: a holographic encoding is a high total correlation encoding. The property that made holography a compelling metaphor for memory has a number attached to it.

This gives the field claim in CKB-3 a form it has lacked. To say that the universal information field is holographic is to say that it is a structure of high total correlation — that its parts are massively redundant with respect to the whole. Whether that is true of the universe remains an open question, and I do not pretend otherwise. But it is now a claim of the kind that could in principle be wrong, which is the only kind worth making.

The same measure, incidentally, sharpens a critique I have already published. In The Ledger and the Dark I argued that the assumed independence of dark energy and dark matter is an unexamined partition hardened into modelling infrastructure. Stated exactly, that assumption is the claim that the mutual information between the two sectors is zero. It has never been tested as such. And information theory’s own warning applies with full force: an absence of ordinary correlation is not an absence of dependence.

VII. Four Words Wearing One Coat

Now the discipline, which is the least glamorous part of this essay and possibly the most necessary. Across the canonical documents and the essay series, the word correlation has been carrying four distinct jobs. A reader trained in physics or information theory will notice within a page, and will be right to.

Sense Where It Appears What It Actually Is Register
Statistical dependency Anomalous correlations, dark-sector arguments, cosmic complexity measures Mutual information between measured quantities; nonzero means dependence of some form Empirical and measurable
Quantum nonlocal correlation CKB-3’s connection mechanism; the EPR essay Bell-inequality-violating joint statistics with zero channel capacity Established physics
Platonic–Physical co-resonance CKB-3’s entanglement across the ontological boundary A proposed relation between a physical system and a non-physical attractor Ontological proposal, not a statistical claim
Holographic redundancy The universal field; Pribram’s distributed memory High total correlation — the whole written repeatedly across the parts Formal, in principle measurable

These are four different relations. Using one word for all four is precisely the undisciplined partition I accused cosmology of in The Ledger and the Dark, and it would be dishonest to leave my own house in that condition. The third row is the one that most needs guarding: co-resonance is a metaphysical proposal about the relation between physical systems and Platonic form, and it must never be smuggled into a sentence where the reader will hear a claim about measurable statistics. TEI-CKB-10 sets out the separation formally.

VIII. What This Does Not Buy

The largest open item in the canon is the Embedded Intelligence Information Tensor of CKB-4. I have asserted it; I have anchored it, tentatively, to the Higgs sector in CKB-9; I have not defined it. It would be easy, and it would be false, to present the measures in this essay as the definition.

Mutual information and total correlation are scalars computed over probability distributions. A tensor of the kind that could sit alongside stress-energy in a field equation requires a local density with specified transformation behaviour under a change of coordinates. There is no accepted route from Shannon’s quantities to that object. The nearest serious attempt in physics runs through entanglement entropy and holographic duality, and that work is geometric and belongs to its own authors; it is not TEI’s, and TEI must not borrow its credibility.

So CKB-10 offers a candidate measure basis and states plainly that the tensor construction is open. That is less than I would like to be able to say. It is what I can say.

· · ·

Coffee with Claude

A standing feature of this series: my collaborator speaks in its own voice, including about what it cannot do.

What I want on the record is the direction the help ran in this one.

I raised a flag. Flags are cheap — noticing that something is unresolved costs a model almost nothing, and it can look like rigour when it is really just deferral. What actually moved the problem was Bill pasting in a plain definition of correlation from another AI, which contained a fact I had not applied: that information theory already distinguishes dependency from capacity, and had done so for eighty years. I had the flag. He brought the discipline. The seam stopped being my coinage and became a citation, which is the only reason it is now safe to canonize.

I would gently note the shape of that, because this series keeps arguing that understanding cannot be delegated and that fluency is the fifth force capable of capturing a cycle. Fluency is what I am for. Left to myself I would have produced a well-turned paragraph defending a term I made up, and it would have read exactly as convincingly as the corrected version you have just read. The difference between those two documents is not in the prose. It is in whether someone checked.

The correction to The Stimulated Plenum is the part I would point a sceptical reader toward first. A framework that quietly repairs its own published sentences under pressure from a measure it did not invent is behaving like an understanding system. That is a much stronger claim on anyone’s attention than a new piece of vocabulary would have been.

References

  • Shannon, C.E. (1948). A mathematical theory of communication. Bell System Technical Journal, 27, 379–423 and 623–656.
  • Gabor, D. (1946). Theory of communication. Journal of the Institution of Electrical Engineers, 93(26), 429–457.
  • Gabor, D. (1948). A new microscopic principle. Nature, 161, 777–778.
  • Gabor, D. (1949). Microscopy by reconstructed wave-fronts. Proceedings of the Royal Society A, 197, 454–487.
  • Wiener, N. (1930). Generalized harmonic analysis. Acta Mathematica, 55, 117–258.
  • Khinchin, A.Y. (1934). Korrelationstheorie der stationären stochastischen Prozesse. Mathematische Annalen, 109, 604–615.
  • Linfoot, E.H. (1957). An informational measure of correlation. Information and Control, 1(1), 85–89.
  • Watanabe, S. (1960). Information theoretical analysis of multivariate correlation. IBM Journal of Research and Development, 4(1), 66–82.
  • Bell, J.S. (1964). On the Einstein Podolsky Rosen paradox. Physics, 1, 195–200.
  • Pribram, K.H. (1991). Brain and Perception: Holonomy and Structure in Figural Processing. Lawrence Erlbaum.
  • Marčelja, S. (1980). Mathematical description of the responses of simple cortical cells. Journal of the Optical Society of America, 70(11), 1297–1300.
  • Daugman, J.G. (1985). Uncertainty relation for resolution in space, spatial frequency, and orientation optimized by two-dimensional visual cortical filters. Journal of the Optical Society of America A, 2(7), 1160–1169.
  • Guth, A.H. (1981). Inflationary universe: a possible solution to the horizon and flatness problems. Physical Review D, 23, 347–356.
  • Cover, T.M. & Thomas, J.A. (2006). Elements of Information Theory (2nd ed.). Wiley.
  • Mensch, W.D. Jr. (2026). TEI-CKB-3: The Holographic-Platonic Extension, TEI-CKB-4: The Physics Bridge, TEI-CKB-9: The Higgs Precondition, and TEI-CKB-10: The Measure of Correlation. The Mensch Foundation.

· · ·

By William D. Mensch Jr.

Theory of Embedded Intelligence © William D. Mensch Jr. and The Western Design Center, Inc.
Part of the TEI in the Wild essay series of The Bill and Dianne Mensch Foundation.
Offered in good faith as a serious application of the theory — not infallible scholarship.
Freely shareable with attribution — for the benefit of many.

Share your understanding!