S1-302 – Communication as model transfer
The miracle and the mess
In 1948, a mathematician at Bell Telephone Laboratories named Claude Shannon published a paper that changed, in a single stroke, what it was possible to think about communication. Shannon was not interested in the meaning of messages: he was explicit about this from the first pages of his paper, which noted that the semantic aspects of communication are irrelevant to the engineering problem he was solving. What Shannon was interested in was something more fundamental: given a source of information and a channel through which it must be transmitted, how much information can the channel reliably carry, and what determines that limit?
The paper introduced a framework (information theory) that provided precise mathematical answers to these questions and that has since become the foundation of every digital communications technology from the telephone to the internet to the smartphone. But Shannon’s framework, developed for engineering purposes and deliberately stripped of semantic content, captures something about human communication that is easy to overlook precisely because it is so far from what we usually think communication is about. Communication, Shannon’s framework implies, is the transfer of a signal from a source to a receiver through a channel. The signal is subject to noise: distortion that the channel introduces between transmission and reception. And the receiver cannot simply read off the original signal from what arrives, because what arrives is the original signal plus whatever the channel added to it.
Human communication is not a telephone line. But the structure Shannon identified (source, channel, noise, receiver) describes something real about the difficulty of getting a thought from one mind into another. This article is about that difficulty: about what actually happens when one person tries to communicate something to another, why it so consistently and predictably fails, and what the conditions for success actually require.
The compression problem
The fundamental challenge of human communication is not the channel: not the physical medium through which words travel from speaker to listener. It is the compression that must occur before the signal enters the channel, and the decompression that must occur after it leaves it.
A thought, in the mind of the person who has it, is embedded in a rich context: a network of associations, memories, experiences, emotional resonances, background assumptions, and unstated implications that give it its specific meaning for that person. To communicate the thought, the person must compress this context into a sequence of words (a linear, one-dimensional signal) and then transmit that sequence through a channel (sound waves, marks on a page, pixels on a screen) to a receiver who must reconstruct the original thought from the sequence of words alone, without access to the context in which it was embedded.
The reconstruction is not a decoding. Decoding, in Shannon’s sense, is a mechanical operation: given the code and the received signal, one can recover the original message exactly, provided the noise has not corrupted it beyond the code’s error-correction capacity. Human communication is not like this. The words do not encode the thought in a way that allows mechanical reconstruction. They provide cues (pointers to structures in the receiver’s own network of associations, memories, and background assumptions) that the receiver must use to construct a thought that is, with luck, approximately similar to the thought the speaker had. The receiver does not recover the original. They build something new, guided by the cues, from the materials of their own mind.
The miracle of communication is that this works as often as it does. The mess of communication is that it fails in ways that are systematic, predictable, and largely invisible to the parties involved, because each party has access only to their own reconstruction, and has no direct way to verify that their reconstruction matches the other’s original.
Shannon’s framework: what it captures and what it misses
Shannon’s information theory gives precise mathematical form to the engineering problem of communication: how to transmit a signal reliably through a noisy channel. Its central concept (information) is defined not in terms of meaning but in terms of probability: the information content of a message is a function of how surprising it is, how much it reduces uncertainty. A message that tells us something we already knew conveys no information in Shannon’s technical sense. A message that tells us something completely unexpected conveys a great deal.
This counterintuitive definition of information has direct implications for human communication. The words that carry the most information in Shannon’s sense (the most surprising, unexpected, contextually anomalous words) are often the hardest to process and the most likely to be misunderstood, because the receiver’s expectation-based processing is disrupted by them. The words that are easiest to process (the expected, contextually predictable words) carry the least information in Shannon’s sense. Human communication navigates a permanent tension between informativeness and processability: the most informative messages are the hardest to receive accurately, and the easiest to receive accurately are the least informative.¹
Redundancy is the other half of what Shannon measured, and it dissolves an apparent paradox. This article has called communication compression, the squeezing of a richly contextual thought into a thin line of words, and yet natural languages are, at the level of the signal, lavishly redundant. The two live at different levels. Language compresses hard where meaning is concerned, throwing away almost all of a thought’s context, and pads generously where the symbols are concerned, sending far more letters than the message strictly needs. Shannon estimated in 1951, in a study of the entropy of printed English, that the language is roughly 50 percent redundant, that about half of what we write could in principle be reconstructed from the rest, and later analyses put the figure higher still. The effect is easier to feel than to be told: a sentence stripped of its vowels, like “y cn stll rd ths wtht mch trbl,” still delivers its meaning almost intact. This padding is not a flaw but a form of error correction, the slack that lets a message survive a noisy room, a bad connection, an unfamiliar accent, or a smeared line of handwriting. And it quietly confirms this article’s central claim, because if half the letters can go and the meaning still be recovered, then the meaning was never in the letters. It was rebuilt by the reader from the redundancy of the language and the context of the sentence, which is only to say, once more and now unmistakably, that words are cues and not a code.
What Shannon’s framework does not capture is precisely what makes human communication different from the transmission of a telephone signal: the role of shared context. Two people who share a rich common context (shared experiences, shared background knowledge, shared vocabulary, shared assumptions about what is relevant and what can be left unsaid) can communicate efficiently with very few words, because each word activates a large amount of shared context and allows an enormous amount to be left unsaid. Two people who share no common context require an enormous amount of explicit content to communicate anything precisely, and even then the reconstruction will be approximate because the receiver’s context (the materials from which the reconstruction must be built) is different from the sender’s.
The efficiency of human communication is therefore not primarily a function of the words used. It is a function of the shared context within which the words operate. The same sentence (“the bank was steep”) has entirely different meanings depending on whether the shared context is a conversation about rivers or a conversation about finance. The same phrase (“can you pass the salt?”) is a request in one context and an insult in another, depending on the relationship between the speakers and the history of their interaction. The words are cues. The context is the meaning.
Grice’s maxims: the implicit contract of conversation
In 1975, the philosopher Herbert Paul Grice published an account of the implicit principles that govern cooperative conversation: the principles that speakers and listeners rely on to make communication possible without having to make everything explicit.² Grice observed that conversation is not merely a sequence of sentences exchanged between two parties. It is a cooperative activity governed by what he called the cooperative principle: the implicit assumption that participants in a conversation are trying to make their contributions as informative, truthful, relevant, and clear as the purpose of the exchange requires.
Grice identified four maxims through which the cooperative principle operates. The maxim of quantity: make your contribution as informative as required, but not more so. The maxim of quality: do not say what you believe to be false, and do not say what you lack evidence for. The maxim of relation: be relevant. The maxim of manner: be clear, brief, and orderly, and avoid obscurity and ambiguity.
These maxims are not rules that speakers consciously follow. They are background assumptions that listeners bring to the interpretation of what speakers say: assumptions that shape how every sentence is interpreted. When a speaker says “I have read some of Kant’s work,” the listener, applying the maxim of quantity, infers that the speaker has not read all of Kant’s work, because if they had, the maxim of quantity would have required them to say so. When a speaker responds to “How do you like my haircut?” with “Well, you got it cut,” the listener, applying the maxim of relation, infers that the speaker’s response is relevant to the question, and since the literal content seems irrelevant, infers an implicature (an implied meaning) that the speaker did not like the haircut but was unwilling to say so directly.
What makes Grice’s account important for this series is not the maxims themselves but what happens when they break down. The maxims assume a cooperative relationship between speaker and listener: an assumption that both parties are trying to communicate honestly and helpfully. When this assumption is violated (when a speaker is trying to mislead, when a listener is trying to catch the speaker out, when the social dynamics of the exchange make honesty costly) the maxim-based inferential system produces systematically wrong interpretations. The listener who infers a Gricean implicature from a speaker who is deliberately flouting the maxim of quality will arrive at a false belief that they have generated themselves, through perfectly valid Gricean inference, from a misleading cue. The communication has succeeded in the technical sense (a signal was transmitted and a reconstruction was produced) and failed in the substantive sense: the reconstruction is not even approximately similar to the original.
The telephone game: how information degrades in transmission
There is a children’s game that has been played in many cultures under many names (in English it is called the Telephone game, or Chinese Whispers) in which a message is whispered from person to person around a circle, and the final message, when announced aloud, typically bears little resemblance to the original. The game is played for its comedy value: the absurd transformations that occur through the chain of transmissions. It is also a demonstration of something important about how information degrades in sequential transmission.
Each transmission in the telephone game involves compression and decompression: the receiver reconstructs the message from the cues they received, and then re-compresses their reconstruction into a new signal to pass along. The errors at each stage are small (a word misheard, a phrase slightly reinterpreted) but they accumulate, and they accumulate in a specific direction. The receiver’s reconstruction is not random: it is shaped by the receiver’s own associations, expectations, and background assumptions. The distortions are systematic. The message tends to become more coherent with the receiver’s prior expectations, more consistent with familiar story structures, and shorter (less informative in Shannon’s sense) as the surprising and unexpected elements are dropped or assimilated to the expected.
This is not only a children’s game. It is the mechanism by which information propagates through any social network, through any organization, through any chain of communication that involves multiple transmissions. Historical events become narratives. Scientific findings become headlines. Complex policy arguments become slogans. Legal judgments become precedents. At each stage, the compression strips away context, the decompression fills in context from the receiver’s own stores, and the result drifts from the original in the characteristic direction of the receiver’s prior expectations and schemas.
The practical implications for organizations and institutions are direct. Information that matters (about what is actually happening at the operational level, about the effects of policies, about the accuracy of models) must travel from its source through multiple layers of compression and decompression before it reaches the people who need to act on it. At each layer, the systematic biases of the compression process operate. The unexpected and inconvenient tends to be stripped away. The expected and reassuring tends to be preserved and amplified. The result is that the information that reaches decision-makers is not a random sample of available information. It is a systematically biased sample, skewed in predictable directions by the cumulative effect of the compression process.
The illusion of mutual understanding
The most consequential failure of communication is not the obvious failure (the one where both parties know that they have not understood each other) but the invisible failure: the one where both parties believe they have understood each other perfectly, and are wrong.
This failure is invisible because each party has access only to their own reconstruction. The speaker transmits a signal and has no direct access to the receiver’s reconstruction. The receiver constructs a reconstruction and has no direct access to the speaker’s original. Both parties may be confident that communication has occurred (they may have nodded, signaled agreement, and moved on) while their respective versions of what was communicated differ in ways that will only become visible when they act on their respective understandings and discover the gap.
The gap is not always discovered. In many conversations, the reconstructions of the two parties are similar enough that the differences never produce a visible consequence. In the conversations that matter most (agreements about what is to be done, commitments about how things will proceed, shared understandings of what a situation means) the invisible failure is most likely to be consequential and least likely to be discovered until the cost of discovering it is high.
George Bernard Shaw’s observation (attributed to him, though without a specific verified source) that the single biggest problem in communication is the illusion that it has taken place, is not an epigram. It is a description of a specific failure mode: the confident mutual misunderstanding in which both parties believe they have communicated and have not. The failure mode is particularly likely in interactions between people with different professional backgrounds, different cultural frameworks, or different levels of familiarity with the subject being discussed, because in these interactions the shared context is thinner, the words carry less shared meaning, and the reconstructions are therefore more likely to diverge from each other and from the original. But it is also likely, and often more consequential, in interactions between people who believe they share a context (between colleagues, between partners, between friends) because the belief in shared context reduces the checking that would otherwise reveal the gap.³
The Conscious Look, applied to communication
The Conscious Look, applied to the problem this article has been describing, is not primarily a practice of speaking more carefully (though that helps). It is a practice of assuming, as a default, that communication has not occurred until there is evidence that it has: not the social signal of agreement and acknowledgment, which can be produced without genuine understanding, but the substantive signal of the receiver demonstrating, through their own words and their own actions, that their reconstruction approximates the original.
The practical tools for reducing the invisible failure are few, simple, and consistently underused. The most reliable is the technique that educators call teach-back or the teach-back method: ask the receiver to explain, in their own words, what they understood. Not “do you understand?” (which invites a social response) but “can you tell me what you’re going to do?” or “can you explain what this means in your own terms?” The teach-back reveals the reconstruction. The gap between the reconstruction and the original, if there is one, becomes visible before the cost of the gap is realized.
The second is the discipline of distinguishing between what was said and what was meant: of noticing when one’s own interpretation of what another person said is not the only possible interpretation, and of checking whether the intended interpretation was in fact the one the speaker had. This discipline requires the specific form of epistemic humility that the series has been recommending throughout: the willingness to hold one’s reconstruction as a hypothesis rather than as a direct access to what the other person said, and to remain genuinely open to the possibility that the hypothesis is wrong.
The practice that turns that discipline into a concrete method is the receiver’s counterpart to the teach-back: when one suspects a gap, to say back to the speaker, in one’s own words, what one took them to mean, and to treat the correction that follows as the aim of the exercise rather than an interruption of it. This is the substance of what Carl Rogers called active, or reflective, listening, which is not a performance of attentiveness but a deliberate surfacing of one’s own reconstruction so that its errors can be caught by the one person able to catch them, the one whose meaning it was. Where the teach-back has the speaker check the listener, this has the listener check themselves, out loud, and between them they close the loop from both ends. Said back and corrected, a reconstruction stops being a private guess and becomes a tested one.⁴
The third, and in many ways the most important, is the cultivation of an awareness of the degree of shared context. Communication that relies on dense shared context (that leaves large amounts unsaid because the context will fill it in) is efficient and elegant between people who genuinely share that context and unreliable and dangerous between people who believe they share it but do not. The practice is to make explicit, in proportion to the uncertainty about shared context, what would otherwise be left as implicature: to say what one means rather than assuming that what one says will be interpreted as what one means. This is not the same as saying everything (which would be both tedious and counterproductive). It is the calibration of what is left unsaid to the actual degree of shared context, rather than the assumed degree.
Further reading
Claude Shannon and Warren Weaver’s The Mathematical Theory of Communication (1949) is the foundational text: Shannon’s original 1948 paper together with Weaver’s more accessible introduction explaining its significance for communication more broadly. The mathematical parts require a background in probability, but Weaver’s introduction is readable and precise.
Paul Grice’s “Logic and Conversation,” published in Peter Cole and Jerry Morgan’s Studies in Syntax and Semantics (1975), is the original statement of the cooperative principle and the four maxims: a short and unusually readable piece of analytic philosophy that has generated an enormous subsequent literature in pragmatics and philosophy of language.
Steven Pinker’s The Language Instinct (1994) provides the biological and cognitive background for understanding why communication works as well as it does: why the compression and decompression process is as reliable as it is, and what the specific capacities are that make human communication possible. It is the most accessible treatment of the science of language for a general reader.
Deborah Tannen’s You Just Don’t Understand: Women and Men in Conversation (1990) documents in specific detail the ways in which different conversational styles (styles shaped by different cultural and subcultural backgrounds) produce the illusion of shared understanding while concealing systematic differences in what is being communicated. It is the most accessible available account of the invisible failure mode this article describes, in the specific domain of cross-gender conversation.
For the organizational implications (how information degrades through organizational hierarchies) Karl Weick’s Sensemaking in Organizations (1995) provides the most thorough account of how organizations construct shared meaning from ambiguous information, and how the construction process systematically produces distortions that are invisible to the people producing them.
Notes
¹ Shannon’s formal definition of information as a function of surprise (technically, the negative logarithm of the probability of the message) has a counterintuitive implication that is directly relevant to human communication: the most surprising messages carry the most information but are also the most likely to be misprocessed or misremembered, because the receiver’s expectation-based processing is disrupted by them. Predictive processing, described in article S1-201 of this series, provides the cognitive mechanism: the brain generates predictions about incoming signals and processes the discrepancy between prediction and signal as the information-carrying component. A highly surprising message generates a large discrepancy, which is processed as high information but is also more likely to be incorrectly processed, because the absence of a strong prediction means the reconstruction has less prior structure to guide it.
² Grice, H. P. (1975). Logic and conversation. In P. Cole and J. Morgan (Eds.), Syntax and Semantics, Vol. 3: Speech Acts (pp. 41-58). Academic Press. The concept of conversational implicature (the meaning that is conveyed by the manner of saying something rather than by its literal content) is one of Grice’s most important contributions to philosophy of language. The implicature in the haircut example is generated by the receiver’s inference that the speaker’s contribution must be relevant (maxim of relation) and sufficiently informative (maxim of quantity), so if the literal content seems irrelevant and uninformative, the receiver infers that the speaker is conveying something else. This inferential mechanism is responsible for a very large proportion of the meaning conveyed in ordinary conversation and is a major source of misunderstanding when speakers and listeners have different assumptions about what counts as relevant and informative.
³ The specific failure mode of confident mutual misunderstanding (in which both parties believe communication has occurred and have in fact produced significantly different reconstructions) has been studied in the context of medical communication, where the consequences of the failure are most clearly measurable. Patients consistently interpret what physicians say differently from what the physician intended, and both parties consistently believe they have understood each other. Schillinger, D., Piette, J., Grumbach, K., Wang, F., Wilson, C., Daher, C., Leong-Grotz, K., Castro, C., and Bindman, A. B. (2003). Closing the loop: Physician communication with diabetic patients who have low health literacy. Archives of Internal Medicine, 163(1), 83-90. The study found that physicians who used the teach-back method (asking patients to explain in their own words what they had been told) identified significant discrepancies between intended and received communication in a majority of interactions, despite both parties having believed communication was successful.
⁴ The same say-it-back loop, pushed to its limit in the setting of disagreement, is what is now called steelmanning: restating another person’s argument in its strongest form, strong enough that its author would accept the restatement as their own, before answering it. That is a move in the ethics of argument rather than in the mechanics of communication, and it is treated in the project’s work on disagreement and structured dissent, the articles on talking across the gap and on adversarial collaboration (S1-208, I-25).