This paper is a translation from the Japanese original. Quotations from Japanese sources were translated for this version, and character counts (17, 189, 212, and so on) refer to the Japanese text. Where the reading is in doubt, the Japanese original governs. “Please RT” renders 拡散希望, the usual Japanese request to spread a post, which reads literally “diffusion wanted.”
“A wealth of information creates a poverty of attention.” Herbert Simon wrote this in 1971.[3] When information overflows, what becomes scarce is not information but the attention of those who receive it. A quarter of a century later, Goldhaber argued that the scarcest thing on the Net is attention, and that the economy would be reorganized around the struggle for it.[4] What happened next has been described in detail by Wu[5] and Zuboff.[6] The business of gathering people’s attention and selling it came to decide the very shape of the flow of information in society.
In the 2020s, what had been a warning became a finished mechanism. In 2023 X began sharing revenue with those who post, and in September of that year the total paid out was reported to have reached about 20 million dollars.[7] The number of times a thing is seen, that is, its impressions, turns into cash. In the Noto Peninsula earthquake of New Year’s Day 2024, many fake rescue requests were posted, apparently for the purpose of earning revenue, and the phrase “imp farming” (「インプ稼ぎ」: the farming of impressions, or “imps”) appeared in headline after headline in the news.[8][9] Neither truth nor content is asked about. Only the count is.
What sells for the highest price in this market is not explanation. It is a single sentence, clipped out of the middle of an explanation, whose meaning cannot be understood. Incomprehension draws reactions, reactions draw counts, and counts draw cash. Meanwhile, for the trouble of looking up the sentence that was not understood, nobody pays a single yen. Ridicule is cheaper than a search and, what is more, it sells well.
The blog article published on December 28, 2018, “[Wait, This Was About Mackerel?] On the Comedy Sketch Called WEEKLY OCHIAI, or on Hell” (「【サバの話だったの？】WEEKLY OCHIAIというコント、あるいは地獄について。」),[1] is a textbook specimen of this economy. The article asks in its title: was it about mackerel? One line is enough for this paper’s answer. Yes, it was about mackerel. In what follows, the paper shows from the recording and the book what was actually said on the program and how it was received (Section 3); confirms from the scholarly literature how stripes form (Section 4); extracts the method of the article (Section 5); and criticizes the mechanism that keeps paying rewards for that method, and imp farming, which is its present-day form (Sections 6 and 7). It then sets back on its feet the question that that one sentence was pointing at. Why does mimicry aimed at the optic nerve emerge from the discrete computation of genes? (Section 8.) The core of this paper lies there.
The object of study is the NewsPicks program WEEKLY OCHIAI, Season 2, “Update Japanese Society” (「日本社会をアップデートせよ」). It was streamed on June 13, 2018, and runs 60 minutes 12 seconds. The guest was Tsunehiro Uno (editor-in-chief of PLANETS). The official page describes the program as “an open discussion in which viewers of the program and the studio audience alike can take part in the debate.”[2] Remarks made on the program are quoted from the recording of the stream, with the playback position given as minutes:seconds. The wording of the article is quoted from its published text.[1]
The sources on Ochiai’s side are all in place, outside the program. They are the solo exhibition at GYRE in Omotesando, “Yoichi Ochiai: Beauty of Natural Resolution ∽ End to End Transformation of Material Things ∽ Digital Nature” (「落合陽一、山紫水明∽事事無碍∽計算機自然」; April 20 to June 28, 2018);[10] the book Digital Nature: Wabi and Sabi by Pantheized Computers Forming an Ecosystem (『デジタルネイチャー：生態系を為す汎神化した計算機による侘と寂』; on sale June 15, 2018);[11] Ochiai’s posts on X (formerly Twitter); and his writing on the platform note.[12] Among his books, this paper also cites The Century of Magic (『魔法の世紀』; 2015),[13] which is the premise of Digital Nature, and Matagi Drive (『マタギドライヴ』; published September 2026),[14] which is its sequel. Line up the dates, and the story becomes so simple that it is almost a letdown.
A little over six minutes into the main topic, the moderator asked Ochiai: “Ochiai-san, could you explain what Digital Nature is, in five minutes?” (21:20; wording as transcribed in the article[1]). In the recording the moderator goes on to say, “Could you diagram it for us?” What is to be diagrammed is the pair of sentences that open Chapter 1 of the book Digital Nature, due on sale two days later. They run to 189 Japanese characters.
Digital Nature is to reconstruct through computational technology the wisdom called quantization that living things brought forth, and thereby to update the existing nature and to implement it. And at the same time, it is a body of thought whose aim is, having deconstructed ‘modern human existence,’ to compose an environment in which computers and non-computers are inseparable and to construct a numerical nature, and thereby to overcome ‘modernity’ and to transcend, in a circular way, the binary oppositions of language and phenomenon, analog and digital, subject and object, landscape and scenery.— Digital Nature, opening of Chapter 1[11]
Drawing a diagram on the whiteboard, Ochiai worked through these two sentences from the top, one phrase at a time. He begins at 21:40 and ends at 27:05. That is 5 minutes 25 seconds. Along the way he remarks of a phrase he has just read aloud, “That’s the sentence, up to there” (22:13); having finished the first sentence, he marks it off, saying that this much is what is told “in the first half here, lines 1, 2, 3, 4” (24:36); and from “And at the same time” he goes into the second sentence. This is no off-the-cuff lecture. It is a word-by-word annotation of the definition he was asked to explain in five minutes. He overruns the five minutes he was asked for by just 25 seconds. What follows reads that annotation in order.
If you ask what the first computer born on Earth was, it’s living things that have DNA. There’s a digital computer that runs in base 4… and, well, this is the base code.— The program, 21:40 (wording as transcribed in the article[1])
Living things are the first information-processing quantization machines to have appeared on this Earth. … DNA is recorded using digital data, and information is transmitted to the next generation with errors reduced by means of quantized data and an “error-correction function.”— Digital Nature, preface[11]
The genetic information of DNA is written as a sequence of four kinds of bases.[18] Since each base is one choice out of four, one base corresponds to exactly two bits. The “digital computer that runs in base 4” is this code. The bases, in sets of three, specify one amino acid.[19] The information of life is stored and read out not as continuous quantities but as strings of discrete symbols. In 1994 Adleman actually solved a combinatorial problem by means of reactions among DNA molecules.[20] This is more than a metaphor. DNA is a medium that can be used for computation. On the same page as the definition, the book writes: “To begin with, the ‘digital,’ in the sense of quantization, was a form of information computation proper to living things. The records of DNA and RNA are quantized by four kinds of bases, and further encoded by codons.”[11]
And then, after that, well… how should I put it… you get living things with nervous systems, which turned nonlinear phenomena into, uh… into that thing.— The program, 21:54 (wording as transcribed in the article[1])
Here the article asks, “What goes in the place of ‘that thing’?” and, instead of looking for the answer, moves on to a joke.[1] The answer lies just ahead. Ochiai goes on to read aloud the first sentence of the definition. The article, too, transcribes that moment.
That’s what’s written at the very start: Digital Nature is to reconstruct through computer technology the wisdom called quantization that living things brought forth, and thereby—that’s the sentence, up to there.— The program, 22:08 (transcribed from the recording)
“That thing” is quantization. Ochiai takes up the talk of DNA and the nervous system with “That’s what’s written at the very start,” and reads aloud the first sentence of the definition. The word that belongs in the spot where, searching for a word, he put “that thing” is in that sentence. The preface defines the word: “Quantization here means quantization in the sense of information theory.”[11]
Quantization is the rounding of a continuous signal to a finite number of levels. Ever since Shannon set up the mathematical theory of communication,[21] treating information as discrete symbols has been the foundation of communication and computation. Nerves do the same thing. In 1943 McCulloch and Pitts formulated the nerve cell as a logical element that fires if the sum of its inputs exceeds a threshold and does not fire if it does not.[22] In August 1952 Hodgkin and Huxley wrote the current flowing across the membrane of the squid giant axon as nonlinear differential equations, and reproduced by computation how excitation that exceeds a threshold becomes an action potential and travels along the axon.[23] It was published in the same month as Turing’s paper on morphogenesis.[24] Nerves take a nonlinear, continuous phenomenon, reduce it to the discrete (fire, or do not fire), and carry it onward. Of the sense organs, too, the book writes: “The retina and the cochlea are also quantization devices: devices that digitize the light rays and air vibrations of space and connect them to the nervous system.”[11] The program’s way of putting it, a nervous system that quantizes nonlinear phenomena, states the standard understanding of neurophysiology in a single breath. The “that thing” that the article puzzled over was written in the first line of Chapter 1 of the very book the article named.
And, um, what “to reconstruct through computer technology” means is this… statistical error correction, and statistical differentiation. Uh, you reconstruct, through computer technology, the kind of approach that automates partial differentiation.— The program, 22:16 (wording as transcribed in the article[1])
Round a signal, and detail is lost. In exchange, symbols do not degrade when they are copied, and copying errors can be corrected. In the replication of DNA, the enzyme selects the correct base with a selectivity of ten-thousand-fold to a million-fold, and a proofreading function excises 90 to 99.9% of the misinsertions.[25] On the computer side, Hamming in 1950 devised codes that detect and correct errors.[26] In 1956 von Neumann discussed a probabilistic logic for synthesizing reliable mechanisms from unreliable components.[27] Neither living things nor computers eliminate errors one by one; they use redundancy to lower the rate at which errors occur. That is what “statistical error correction” means.
The second half, “statistical differentiation” and “the kind of approach that automates partial differentiation,” can be translated as it stands into the technical terms of machine learning. On the same page as the definition, the book traces the path from the perceptron,[28] which took its inspiration from the nervous system, to deep learning, and writes: “These show that, using quantized data, optimization problems can be computed by a statistical approach.”[11] Backpropagation, presented by Rumelhart and his colleagues in 1986, carries the error at the output backward by the chain rule, as partial derivatives with respect to the weight of each connection, and corrects the weights little by little.[29] The machinery that has the computer produce those partial derivatives automatically, with no person deriving them by hand, is automatic differentiation,[30] and it became the foundation of deep learning in the 2010s. In the words of the preface, the neural processing that living things devised is “being repeated, in the manner of reinventing the wheel, by the silicon-type computers that humankind has produced.”[11]
Here begins the passage that became the article’s title. One continuous stretch of speech is transcribed from the recording, just as it is.
You know how they say the Earth was born however many hundred million years ago? Made over 4.6 billion years, the iteration, the result that came out of the iteration, for example the pattern on a mackerel’s back took quite a long time to make, you see, but that iteration, in the case of a computer, this iteration, I mean, in one second, with a CPU you can run it something like 3 giga times, right? You can run it 3 giga times, and for a living thing this is, what, once a year or so, right?— The program, 22:37–23:03 (transcribed from the recording)
It is an ordinary sentence of comparison, with its subject and its predicate in place. Nature, over a long iteration, made, for example, the pattern on the mackerel’s back. A computer can run that iteration three billion times a second. A living thing runs it about once a year. The age of the Earth is about 4.55 billion years;[31] the tick of a 3-gigahertz CPU comes three billion times a second. A year is about 31.56 million seconds, so in the span of one generation of a living thing the computer runs about 9.5×1016 times. That is a difference of 17 orders of magnitude. Evolution is an iterative computation that repeats mutation and selection generation by generation, and Ochiai is comparing the orders of magnitude. The explanation comes to this: “to reconstruct through computational technology,” in the definition, means to run again, at a speed orders of magnitude greater, the iteration that living things have been running over the history of the Earth.
The article quoted these 25 seconds cut into three. The first piece it stops at “Made over 4.6 billion years, the iteration?” and follows with: “Don’t give me ‘iteration?’ Don’t check with us. You’re the only one who understands what you just said.”[1] In the recording, Ochiai says “iteration” and at once carries on from that same word into the next phrase. The second piece is the sentence from which the title came. Here the recording and the article’s quotation are set side by side.
What has vanished in the article’s quotation is the 17 Japanese characters 「を作るのに相当時間がかかってるわけ」 (“took quite a long time to make”). The predicate of this sentence is inside them. Take out the predicate, and any sentence at all becomes a case of “I don’t know what the talk was about.” Immediately afterward the article writes, “Lo and behold, this incomprehensible talk turned out to be about the pattern on a mackerel’s back,” and continues, “This catharsis is precisely what makes the big highlight of a comedy sketch.”[1] The third quotation (from “That iteration… in the case of a computer” onward) is placed on the far side of 391 Japanese characters of the article’s own prose, one heading included, and is prefaced with: “Once past the mackerel catharsis, the talk at once becomes incomprehensible again.” A single continuous sentence of comparison was in this way reassembled into a punchline, and into the “incomprehensible” before and after the punchline.
It’s an update of our view of nature, the kind where you go ahead and put people, and living things, and then computers here too, inside a humanly defined supernature. This thing is what I’m calling Digital Nature.— The program, 23:30 (wording as transcribed in the article[1])
This is the reading of the close of the first sentence, “to update the existing nature.” One stops setting human beings and nature face to face, and thinks of a single nature that takes in people, living things, and computers alike. Ochiai then goes into the second sentence. He explains “an environment in which computers and non-computers are inseparable” with the example of an environment in which one can no longer tell whether a goldfish merely seems to be floating there or is a physical goldfish (25:32); rephrases “a numerical nature” as “nature that is measured in numbers” (25:53); and goes on through “the continuous quantity called analog and this discrete quantity called digital” (26:23) to subject and object, landscape and scenery, following the words of the definition in order, and finishes at 27:05.
The premise of this view of nature is The Century of Magic,[13] which discussed the changes in expression and environment that computers bring. Later, while citing Turing’s theory of morphogenesis as an example of nature’s “computation,” Ochiai drew a distinction, “The ‘computation’ Turing used here is not computation in the sense of the Turing machine—an algorithmic procedure—but the description of a dynamical system by continuous differential equations,” and wrote, “If this distinction is blurred, the reach of the proposition ‘nature computes’ is unduly enlarged.”[12] The claim called Digital Nature holds a measuring stick to its own reach as well.
The article depicts this program as a place of “pretending to understand when nobody understands,” and writes, “Not a single person can say, ‘Huh? What are you talking about?’”[1] What remains on the recording is a different scene.
First, the guest, Tsunehiro Uno, is the very person who made this book. The publisher named in the colophon is Tsunehiro Uno,[11] and on the program he says so himself: “I’m this book’s editor, so” and “this is about the 18th time I’ve heard it, but” (18:10). He is the person with the least need of all to pretend to understand. The article introduces Uno with “An editor, apparently.” and “But in the end I don’t really get what it is he does.” and laughs at how he looks during the diagram: “No, you don’t get to be bored!!”[1] His own words, that this was the 18th time, are not quoted.
Second, when the diagram is finished, a voice in the studio asks, “Did everyone get that?” and the reply comes back, “Got it, got it” (27:06). One of those appearing on the program says, “So there’s this much of a story in a text this short” (27:17). That is, the diagram was taken this way: 5 minutes 25 seconds’ worth of content is folded into a definition of two sentences and 189 Japanese characters.
Third, when the moderator asked how this connects to the individual from here on, Uno rephrased it on the spot. He began by saying that language is a tool for dividing things up; said, “but that’s an expedient, and really it’s all connected, right?”; said that a computer, without forcing a division, “finds it easier than a human does to grasp things in their still-connected state, is the point”; added the caveat, “this is a super-loose translation, mind you”; and concluded that a world that keeps turning without the expedients called the human being and modern society is something “we’ve now, at last, for the first time, become able to envisage, is the point” (28:09–29:03). Ochiai takes this up with “In other words, it’s a society that works just by sending someone a picture of an apple, without saying ‘apple,’” and moves the talk forward (29:05). The speaker defines; the editor rephrases in his own words; the speaker takes that up and adds the next example. Between editor and speaker, understanding is established on the spot. The article quotes only the one remark that comes just before this rephrasing (“So the world will come to keep turning even without the expedient called the human”), writes, “No, you just stay bored!!!” and ends its account of the program there.[1] The rephrasing, about a minute long, begins some ten-odd seconds later.
Fourth, the very phrase “pretend to understand” is in the program itself. When the moderator starts to move on to the next question, a voice calls out: you gave up for a second just now, didn’t you. The moderator answers, “Myself included, I guess we’ve got to pretend to understand a bit,” asks for it “a bit more concretely,” and moves the talk on to Japanese society (29:48). The article itself introduces the moderator as someone who can say, “I don’t follow from there on either, so shall we talk about something else for now!”[1] It was not a place where one could not say that one did not understand. What was not understood was said aloud among those on the program, rephrased, and steered toward the concrete. The “pretending to understand” that the article discovered is a phrase the moderator spoke first, on the program.
A question; a word-by-word diagram of the definition; a check; the editor’s rephrasing; the speaker’s response; the moderator’s “a bit more concretely”; the next topic. As the running of a debate program, nothing in it is broken.
In these 5 minutes 25 seconds there are hesitations such as “uh…,” “um,” and “what was it… oh, that.” In research on conversation, this is not a defect but a mechanism. Schegloff and his colleagues showed that conversation has an organization of “repair” that deals with the troubles that recur in speaking, hearing, and understanding, and that speakers correcting themselves is distinctly more common than their being corrected by others.[32] Clark and Fox Tree showed, from large corpora of spontaneous conversation, that English uh and um are used as signals that tell the listener the speaker is about to search for a word, or is deciding what to say next.[33] To put down “that thing” and then go on to read a sentence with “quantization” in it is a perfectly ordinary instance of searching for a word.
According to Clark and Wilkes-Gibbs, the participants in a conversation repair, expand on, and replace a way of referring to something until it takes a form they mutually accept. The paper calls that process “an iterative process.”[34] Conversation, too, is iteration. The back-and-forth seen in 3.7, of definition, question, rephrasing, and response, is just that.
The position of one who listens from the far side of the screen has been studied too. In the experiments of Schober and Clark, even when both heard every word of the same conversation, the addressee understood more accurately than the overhearer listening alongside. This is because speakers choose their words to fit the common ground they share with their addressee, and the addressee can work with the speaker to check understanding, whereas the overhearer cannot.[35] The addressees on the program were the book’s editor and the moderator. The far side of the screen is the position of the overhearer. That there are things one does not understand there is ordinary. What is in question is what one does after that.
The effect of the form called transcription should be looked at too. Bucholtz argued that transcription involves interpretive decisions, about what is transcribed, and representational decisions, about how it is written, and that those decisions are made in accordance with the transcriber’s expectations and beliefs about the speakers, the intended readers, and the purpose of the transcription.[36] The hesitations of spoken language are, on the spot, hardly noticed by the listener. Set down in type, they all line up with the same weight, and the intermediate steps of someone thinking while speaking are fixed, just as they are, as defects. The order in which the diagram goes up on the whiteboard drops out, and so do the pointing gestures. The article’s transcription carefully kept the hesitations as “…” and dropped one predicate. What is kept and what is dropped is decided in accordance with the purpose of the transcription. At its outset the article says that what is entertaining about this program is that it is “a comedy sketch that arose naturally,” and states that it will explain this.[1] The transcription is a transcription made for that explanation.
This mackerel is not an analogy thought up on the spot. On the day of the broadcast, Ochiai’s solo exhibition at GYRE in Omotesando was entering the last stretch of its run,[10] and there was a mackerel in it. It was “Sketch of Wave Forms, Reflections, Sea and Sky” (《波の形，反射，海と空の点描》), in which the skin of a mackerel is printed at ultra-high resolution on ink-dyed washi paper and silver leaf. During the run Ochiai wrote, “It is a work in which the skin of a mackerel is printed at ultra-high resolution on ink-dyed washi and silver leaf, so, literally speaking, it is ‘mackerel,’” and each day he was handing out postcards of “Sketch of Sea and Sky (a.k.a. Mackerel)” to the first five visitors.[15] On April 28 he wrote this.
If you do not notice that the mackerel is a mackerel, you will not notice the oddness of the resolution either, you will probably take no interest in the material, and you will probably not understand what it means as a landscape painting, or what digital and iteration mean either, and that is fine. Just “I don’t know what it was, but it was beautiful” is fine.— Yoichi Ochiai, post of April 28, 2018[16]
Mackerel, and iteration. A month and a half before the broadcast, the two words are already standing side by side. On June 5 he announces that the release date of the book has been “set before the exhibition ends,” for the reason that “it is definitely easier to understand if you read it while the exhibition is on!”[37] And Digital Nature, which came out on June 15, closes its final chapter with a commentary on this mackerel work. The next page is already the afterword.
Mackerel. The protective coloration of blue-backed fish is a visual mimicry that looks like the boundary surface between wave and wave; it is the natural world’s optical imitation of the sea and the sun. … In the pattern on the back of the mackerel I saw the landscape made by the sun and the waves of the sea: imprinted, handed down in DNA through the succession of generations, and drawn.— Digital Nature, close of the final chapter[11]
DNA, the succession of generations, and then the pattern on the back of the mackerel. The same materials as in the program’s diagram are lined up, facing the same way. The mackerel was not the punchline. It was the conclusion. Moreover, the article names this book: “Yoichi Ochiai’s new book, ‘Digital Nature.’”[1] The article came out half a year after the book went on sale.
The program’s sentence is sufficient as it stands: the pattern “took quite a long time to make.” Then why can the pattern be called the result of iteration? To this question, one level deeper, Ochiai has been giving an answer since the day after the article was published, addressed to readers outside the program.
The mackerel in the middle is a story about hyperparameter tuning of a diffusion–reaction system, tied in with the visual information processing of natural enemies and told in today’s idiom; that’s what it means.— Yoichi Ochiai, post of December 29, 2018[17]
On the next day, the 30th, he continued, “The point is this: aren’t the hyperparameter tuning ‘that leads to a Turing pattern’ and the hyperparameter tuning ‘of the Turing pattern itself’ a case of adversarial generation against the visual neural circuits of the predator?” and he counts for himself the steps he skipped in the talk. “Well, I am talking with two steps skipped.” What they contain, he says, is “the visual information processing of the predator, the generation and hyperparameter tuning of Turing patterns and the like, the genetic algorithm for that, and adversarial generation as a process.”[38]
Therefore “the mackerel talk” is talk about Turing patterns. To put it a little more exactly, it is the story that the parameters of a pattern born of reaction–diffusion have been tuned by a long iteration played against the eyes of predators. How the stripes form is taken up in Section 4, and the question as a whole in Section 8.
In the same episode there is one more metaphor that the article made much of. It is “Right. ‘God, it turned out, was Doraemon,’ you know,” and what follows it, “Humans don’t have to make an effort, technology fills in for them, and whatever you say, it comes out of the fourth dimension” (20:23). Uno responded that this was the greatest god of the postwar middle-class illusion.[1] It is a metaphor that entrusts to a single character the faith in technology of postwar Japan: wish, and technology is handed to you from the other side. This metaphor has groundwork under it. In the preface of The Century of Magic, in 2015, Ochiai wrote, “Today, although technology is developing in a direction that widens the range of what we can do, its workings are becoming harder and harder to understand,” and gave the name “magic” to technology that is used without its workings ever coming to mind.[13] A tool that comes out of the fourth dimension whatever you say is the limiting case of this. If the century of magic is the one in which things can be used without knowing how they work, then being able to laugh without knowing how they work is its reverse side. Eight years after the program, Ochiai writes, “The author positions Physical AI as the practical consequence at which the lineage from Zuse has arrived.”[12] What the player operates in “It’s About Mackerel.” is a Physical AI: a blue arm reaching in from beyond the surface of the water, and a round hand. Technology no longer only produces tools out of the fourth dimension. It reaches a hand over to this side and takes hold of what is swimming there.
In his 1952 paper “The Chemical Basis of Morphogenesis,” Alan Turing showed that “a system of chemical substances, called morphogens,” reacting and diffusing within a tissue, can spontaneously produce a spatial pattern out of a uniform state.[24] This is where it runs against intuition. Diffusion ordinarily levels out differences of concentration and erases pattern. But when two substances make one another, or suppress one another, a difference in their speeds of diffusion makes the uniform state the unstable one, and stripes and spots arise. In 1972 Gierer and Meinhardt gathered this mechanism into a theory of how living things take form, as a pair, “activator and inhibitor,” that increases itself nearby and suppresses at a distance.[39]
Put another way, it goes like this. A substance makes still more of itself in its own surroundings and, at the same time, makes another substance that spreads faster and farther than it does, and so holds down its own increase a little way off. Then around a dense place there is always a thin place, and outside that another dense place. Nearby it calls in its fellows; far off it pushes them away. When this local rule works all at once over a surface, stripes and spots with a fixed spacing arise, though nobody holds a blueprint of the whole. To make form, no blueprint is needed. A balance between the speeds of reaction and diffusion is enough. With diffusion alone, the pattern disappears. Only when it is paired with reaction do stripes stand up.
Nowhere in the genome are the stripes written one by one. What the genes decide is only numbers: how fast the substances react, and how fast they spread. In the zebrafish, a change in just one gene, that of a protein of the channels that connect cell to cell (connexin 41.8), turns stripes into spots.[40] What was rewritten is not the picture. It is a coefficient. The picture is drawn by the following equations.
∂u/∂t = Du∇²u + f(u, v)∂v/∂t = Dv∇²v + g(u, v)
Here u and v are the concentrations of the two substances, Du and Dv are their respective speeds of diffusion, and f and g are the reactions. What Turing showed was that a uniform state which is stable without diffusion can become unstable the moment diffusion is put in. The conditions can be written as four inequalities (fu and the like are the partial derivatives of the reactions at the uniform state).[41]
fu + gv < 0fugv − fvgu > 0Dvfu + Dugv > 0(Dvfu + Dugv)² > 4DuDv(fugv − fvgu)
The wavenumber kc of the wave at which instability begins is given by kc² = (Dvfu + Dugv) / (2DuDv), and the spacing of the stripes comes to about 2π/kc.[41] The width of the stripes is not something anyone measured and drew. The equations settle it by themselves, from the ratios of the coefficients. Partial derivatives come into this story twice: once as a tool of learning (3.4), and once more as the equations of the pattern itself.
For a long time the theory remained an attractive hypothesis for explaining the patterns of living things. In 1995 Shigeru Kondo and his co-author showed that, as the emperor angelfish (Pomacanthus) grows, the way its stripes split and reconnect proceeds just as numerical computation of reaction–diffusion predicts.[42] The paper writes that until then there had been “no conclusive experimental evidence” of a Turing-type pattern at work in a living thing. After that, it was confirmed in the zebrafish that interactions among pigment cells actually produce the stripes,[43] and it was also shown that the patterns of hybrids, made by crossing species with different patterns, “blend” just as reaction–diffusion predicts.[44] Kondo and Miura have gathered these results into a framework for understanding pattern formation in living things.[45] The cellular machinery actually at work in the zebrafish differs considerably from the first simple model of chemical reactions, but in principle it is of the Turing type: that is how the review of 2015 sorts the matter out.[46]
The patterns of reaction–diffusion have a feature that is easy to recognize. The distance between stripe and stripe is decided by the ratio of the speeds of reaction and diffusion, and is not decided by the size of the body. When the body grows larger the spacing does not change; the number of stripes increases instead. What was shown in the emperor angelfish was exactly this phenomenon. As the fish grows, the stripes split, new stripes come in between, and they rearrange themselves while nearly keeping their spacing.[42] If the pattern were drawn from the start like a blueprint, the spacing of the stripes ought to stretch when the body stretches. It does not, because the pattern is still being computed on the skin, even now. The pattern is not a result that has been drawn. It is a process that is still going on.
The two-variable system that Gray and Scott derived from the analysis of a chemical reactor[47][48] produces, as Pearson showed in 1993, “a surprising variety of irregular spatiotemporal patterns” from slight differences of parameter.[49] In “It’s About Mackerel.” the golden maze that appears on the back of a mackerel that has settled in front of a mirror or the whiteboard is drawn by solving this system inside the browser. On a grid of 192×96, with the diffusion coefficients of the two substances at 1.0 and 0.5, the feed rate F at 0.0367, and the removal rate k at 0.0649, three to six updates are stacked up once every four frames ordinarily, and every frame once a mackerel begins to learn. This F and this k are hyperparameters that the makers chose and fixed. In the sea, the one who takes on that role is outside the mackerel (Section 8). The net of light swaying on the seafloor, too, is computed every frame, by refracting light at the rippling surface of the water and calculating the density of the light that gathers on the bottom. Neither is a picture that someone drew; both are forms that arise out of rule and iteration, and in that sense they are a small Digital Nature. To borrow the words of the laboratory, Digital Nature shows diverse options between human and machine, between the material and the virtual.[50]
In this game, when a mackerel finishes learning the reaction–diffusion equation in front of a mirror or the whiteboard, it is game over. It is a bad joke. But it copies the logic of the attention market just as it is. Once the mackerel has understood the meaning of its own stripes, the laugh of “Wait, this was about mackerel?” ends right there. Understanding is game over for the circulation of clips. That is why the market keeps on cutting just short of understanding.
Of the program, the article writes: “Of course, this is a program that deals with serious news and debate, and not one person among the production staff or the performers is trying to get a laugh.”[1] And then it laughs. From the text of the article, six methods for laughing can be extracted.
First, it drops the predicate. “The pattern on a mackerel’s back took quite a long time to make, you see, but” becomes, in the quotation, “the pattern on a mackerel’s back, you see, but…” (3.5). Stopping the sentence one step short of where its content arrives is the condition of the laugh. It is a technique of clipping with the same shape as “Beyond this point is paid content.” In the language of scholarship it is called quoting out of context, or contextomy. According to McGlone, a quotation taken out of context not only creates a false impression of the speaker’s intentions; it can also contaminate the reading of the quotation even after it is put back into its original context.[51][52]
Second, it chops. The one continuous sentence of 25 seconds was cut into three, and between the pieces were inserted 104 and 391 Japanese characters of the article’s own prose (3.5). How long a speaker’s sentence is, the one who quotes decides; and between sentence and sentence comes the voice of the one who quotes.
Third, it makes incomprehension the punchline. “What is this person saying?” “What goes in the place of ‘that thing’?” “You’re the only one who understands what you just said.” The questions are raised again and again.[1] But in the article there is no trace of anyone having gone to look for the answers. The answer to “that thing” was in the next sentence the article itself transcribed, and the output of the iteration was in the predicate the article dropped. What the article draws from this scene is the understanding, “I don’t know what the talk was about, but at any rate I understood that the pattern on a mackerel’s back has something to do with it.” What is not understood is fixed in place as not understood, and shared out with the reader. This is the true identity of what the article calls “catharsis.”
Fourth, it erases the listeners’ understanding. The book’s editor announcing that this was “the 18th time”; the “Got it, got it” after the diagram; the rephrasing that took a minute; the moderator’s own “pretend to understand.” None of these appears in the article (3.7). Take the people who understood out of the frame, and all that is left is talk that is not understood and people pretending to understand.
Fifth, it chooses what to count. The article counts a different sentence as “212 characters” and “38 seconds,” and writes, “You could cook a hot-spring egg in the time Yoichi Ochiai takes to say one sentence.” In its summary it concerns itself with the number of times a word is said: “He sure likes the word ‘iteration.’ How many times is he going to say it?” Yet of the scene in which, by its account, members of the audience looked at their phones, it writes, “Listening for tens of minutes to talk where you can’t tell what’s being said is rough, isn’t it~” The diagram is 5 minutes 25 seconds; even from the start of the main topic to the end of the diagram, it is about 12 minutes. The 38 seconds are timed to the second, and the 12 minutes are rounded to “tens of minutes.” Count back through the article’s own text, and “comedy sketch” (「コント」) comes up 20 times, “don’t understand” (「分から」) 20 times as well, and “incomprehensible” (「意味不明」) 9 times. The question of how the pattern forms comes up 0 times. The program does not speak the name “Turing.” What it speaks of is the pattern of the mackerel, and the iteration that made it. Neither “Turing” (「チューリング」) nor “reaction–diffusion” (「反応拡散」), where one arrives by asking what lies beyond that, appears in the article even once. The article that counted a sentence of 212 characters as too long has not walked a single step toward Turing’s paper, which is 36 pages long.[24] While we are at it: the preface of this book was made available free on Daily PLANETS on the day of release.[53] At the end of that preface it says, “Not far now to the hot-spring inn.”[11] It would have been quicker to read as far as the hot-spring inn than to count hot-spring eggs.
Sixth, it swaps the object. The article prints a photograph of the audience and appraises their expression: “If I were to express it in a single kanji, it would without a doubt be ‘無’ (nothing).” The critical eye moves from the content of the talk to the faces of the people listening. A photograph of faces teaches nothing about the pattern of the mackerel.
The article also analyzes Ochiai’s way of talking as having “the feature of pouring out a flood of talk in which you can’t tell what he is saying and then suddenly bringing out an analogy to something familiar.” As an observation it is correct, for to state the abstract and then bring out a concrete example is the very form of explanation. The error lies in having treated that concrete example not as the key to the answer but as the punchline of the laugh. The mackerel was not a familiar analogy that suddenly turned up. It was a work in a solo exhibition then running, it was the close of the main text of a book that would come out two days later, and it was the conclusion of the very book the article named.
On the basis of the above, this paper records its judgment. This article is not a critique. It is ridicule by clipping, and the cashing-in of attention. A critique tries to understand its object and then evaluates it. This article declared its abandonment of understanding to be the source of its laughter, called a sentence with its predicate taken out “incomprehensible,” and on top of that evaluated its object as a “comedy sketch.” Drop, chop, make a punchline, erase, count, swap. Not one of the six moves increases understanding; each increases attention. And attention, as the next two sections show, is the currency of this market. Elsewhere the article writes of what it did as “making fun of people while they are talking in earnest.”[1] Exactly so. On the side of the understanding it abandoned, all the answers were in place.
Eight years after its publication, the article is still going around. According to a full-archive search of X, there are 1,018 posts containing the article’s URL between December 28, 2018, and October 5, 2026 (excluding reposts).[54]
In 82 of these 95 months there is at least one post. Peaks rise again and again: 63 posts in April 2020, 31 in April 2022, 29 in September of the same year, 25 in March 2023, 40 in August 2024. The words that come attached are, for example, “A masterpiece I want to reread at regular intervals!” (November 2020) and “Seriously the kind of thing you get the urge to read at regular intervals.” (October 2026). Of the 1,018 posts, 563 contain the word “comedy sketch” (「コント」). Those containing “Turing” (「チューリング」) or “reaction–diffusion” (「反応拡散」) number 0.[54] In eight years it has gone around a thousand times, and among them there is not one post that touches on Turing or on reaction–diffusion. What goes around is the title and the place to laugh, not the content.
The Century of Magic wrote of us, as we look at Twitter, that we “do not share one timeline. Unlike film and television, which project the same image to be shared by everyone, each of us is gazing at a separate display.”[13] The clip goes around those separate screens handing out the same single picture. What that book calls “a society maintained by the sharing of images” comes back in its cheapest form. What is shared is not understanding. It is the place to laugh.
Here is the structure of the corruption. Ridicule, once written, goes around any number of times, gathers attention every time it goes around, and attention now converts directly into cash. Looking something up happens only once, is paid for by nobody, and, on top of that, breaks the laugh. Which of the two the market rewards is plain. And so only the sentence whose predicate was taken out comes around again, still without it, at regular intervals. Truth and content thin out with every round, and only the count piles up. That the fake rescue requests of the Noto Peninsula earthquake were reported as “turning tragedy into cash”[9] is where the same logic ends up.
In Matagi Drive, in 2026, Ochiai writes this landscape in the words of hunting and gathering. Under AI and the platforms, work comes close to gathering. “Labor is no longer production; it is an act of gathering, of meeting one’s quarry in a forest that capital has carefully laid out.” And lying in wait for someone to go up in flames on social media, and throwing stones, he called “a regressive form of hunting that lacks reverence for the god of the mountain and the humility of ‘being granted’ one’s quarry, and lays bare nothing but aggression.”[14] Picking up a clipped sentence and passing it around is the easiest gathering there is in this forest. The quarry is not brought down. It is not eaten either. It is only passed around, and converted into counts.
Put the two “I don’t know”s side by side. In April 2018, Ochiai wrote of the mackerel work, “Just ‘I don’t know what it was, but it was beautiful’ is fine.” In December of the same year, the article wrote, “I don’t know what the talk was about, but at any rate I understood that the pattern on a mackerel’s back has something to do with it.” Both set out from incomprehension. The former tied incomprehension to looking, and the latter tied incomprehension to laughter. The attention economy pays cash only for the latter. It does not let incomprehension turn into curiosity or into beauty; it turns it into ridicule and cashes it in. That is the corruption of this market.
Clip, laugh, gather attention. What in 2018 was the technique of one writer is now a mechanism. In its report of 2024, a study group of Japan’s Ministry of Internal Affairs and Communications explained, as what is generally called the “attention economy,” the economic model in which people’s attention and time become scarce relative to the volume of information supplied, and are traded as things of economic value; and it then wrote as follows.
For example, some information-transmission platform services have adopted mechanisms that grant the originator of a piece of content an economic incentive, through the sharing of advertising revenue and the like, in accordance with the number of views and so on of the content or of the digital advertisements displayed along with it. As a result, there have appeared those who, in order to obtain monetary compensation by stimulating the users on the receiving side, such as viewers, and gathering more attention and interest, circulate and spread on information-transmission platform services extreme titles and content, information not based on fact that has been produced from speculation alone, and the like…— Report of the study group of the Ministry of Internal Affairs and Communications, p. 33[55]
Look at the mechanism itself. X began sharing revenue in July 2023.[7] Under the terms as they stood near the end, one of the conditions for taking part was to “have at least 5M organic impressions within the last 3 months.”[56] That program ended on September 7, 2026, and was replaced by a new one, the Original Content Rewards Program.[56] Under the new program, too, the unit is the display count. Payouts are based on “qualified impressions.” These are the number of times at least 50% of a post was visible on the screen in the Home timeline of a paying subscriber (repeats by the same person are not counted). One of the conditions for taking part is to have been displayed 500,000 times, in the last 90 days, in the Home timelines of verified users.[57] YouTube Shorts has the same shape. The advertising revenue of videos that meet the criteria “is distributed based solely on the number of views of Shorts.”[58]
What is being counted is whether at least half was visible on the screen. Whether it was read, whether it was understood, whether it was correct: none of these is in the unit. If half of the post is visible, that is one. Even if it is not read as far as the predicate, that is one.
What increases under this unit has been studied. Brady and colleagues showed, in 560,000 Twitter posts about gun control, same-sex marriage, and climate change, that each additional moral-emotional word increased diffusion by 20%.[59] Rathje and colleagues showed, in 2.7 million Facebook and Twitter posts by news media and members of the U.S. Congress, that each additional word referring to the politically opposed side raised the odds of a post being shared by 67%, and pointed to the risk that social media is creating incentives that reward expressions of animosity.[60] When an expression of outrage gets “likes” and shares, that person becomes more likely to express outrage the next time too. Brady and colleagues report this as a result consistent with the principles of reinforcement learning.[61]
As for headlines: in experiments on the headlines of the news site Upworthy, for a headline of average length, each additional negative word raised the click-through rate by 2.3%.[62] Headlines that withhold the answer and make the reader read on are an application of the theory that the gap between what one knows and what one wants to know gives rise to curiosity,[63] and as such linguistics has studied their forms.[64][65] “Wait, This Was About Mackerel?” also has the form of a question that withholds its answer. The article answers that the talk “turned out to be about the pattern on a mackerel’s back.” But what that talk is about never comes out, even if one reads to the end. For what the article hands its reader is the understanding, “I don’t know what the talk was about, but.”
As for truth and falsehood: a study of 126,000 rumor cascades on Twitter reports that false news reached more people than true news did, and faster.[66] In the Noto Peninsula earthquake, according to X Corp., among the main posts containing false or mistaken information there were about 21,000 posts concerning “rescue requests.” Associate Professor Shibuya of the University of Tokyo points out that duplicated posts of rescue requests, thought to have been made for the purpose of farming view counts, were observed.[67] The ministry’s report writes that, under mechanisms that give originators an economic incentive in accordance with view counts, the risk of effects on society rises in times of disaster, with, for instance, “an increase in posts of so-called ‘impression farming’ made for the sake of the economic incentive.”[55] Clips, quotations with a sneer attached, headlines that withhold the answer, reposting at regular intervals. The forms seen in the article of 2018, and in the posts that have kept it going around, are every one of them forms that readily turn into counts under the present mechanism.
Remedies for this incentive have been written into the platforms’ terms. X’s new program excludes copied content, minimally modified content, aggregated content, and cross-platform reposts from its rewards, and states that it also excludes disinformation and misleading content, and posts that have a helpful Community Note.[57] YouTube excludes from monetization “clips of moments from your favorite show edited together with little or no narrative,” and writes in its policy that content “should be made for the enjoyment or education of viewers, rather than for the sole purpose of getting views.”[68] A mechanism that counts and pays by the count adds, in writing: do not make raising the count your sole purpose. This paper reads this one sentence of the terms as a written diagnosis of the mechanism.
And the mesh of this net does not fit the shape of clipping. X’s rules say, “Commentary is at the heart of X,” and count as original the posts that react to events, analyze them, and add one’s own perspective. Those to which one’s own contribution is minimal are not counted.[57] YouTube, too, keeps as eligible “using clips for a critical review.”[68] The article of 2018 runs to more than 10,000 Japanese characters, and fills the spaces between its quotations with its own prose. Its contribution is not minimal. In outward form it is commentary. The quoted words, too, were certainly spoken. Only the predicate is missing. Neither set of terms has a clause that names the place of the cut and calls it into question. Whether the two words “misleading content” in X’s rules cover it is not stated.
Three things break. The first is context. A quotation taken out of context can contaminate even the reading that comes after it has been put back (Section 5). Even if this paper puts the predicate back eight years later, the 1,018 posts do not go back. The second is the satisfaction of the receivers themselves. Milli and colleagues, in an audit comparing X’s recommendation system with a chronological display, showed that engagement-based ranking amplifies posts that are emotional and hostile toward the opposing group, and, moreover, that users do not prefer the political posts selected in that way.[69] What one ends up tapping and what one wants to see are not the same. The third is the stripes of the one who explains. This can be stated in the mackerel’s own logic (8.6).
What, then, is to be done? Leaving it alone is no countermeasure. The clip that was left alone is still going around eight years later (Section 6). In 2024 Ochiai wrote this.
What I think of as one of the solutions is for everyone involved in a piece of content “to be left with a bad aftertaste.” If that bad aftertaste comes stuck to it, and everyone who has a hand in it is left with a bad aftertaste, then I think this kind of page-view farming and outrage business will cease to be viable.— Yoichi Ochiai, post of August 28, 2024[70]
This paper offers the aftertaste as the facts that remain after the laugh. The predicate had been taken out. The book’s editor was right there. The equations of the stripes have existed since 1952. The mechanism that pays by the count lowered the price of laughter and raised the price of explanation. Aftertaste reprices them. The same facts remain for the one who clipped, for those who passed it around, and for the mechanism that paid for the counts. This paper is one such aftertaste.
So far we have looked at what was said on the program (Section 3), how stripes form (Section 4), and what the article and the mechanism did (Sections 5 to 7). What remains is the question itself, the one that sentence was pointing at. The core of this paper is here.
At the close of the book of 2018, Ochiai wrote: “Mackerel. The protective coloration of blue-backed fish is a visual mimicry that looks like the boundary surface between wave and wave; it is the natural world’s optical imitation of the sea and the sun.” And he went on to say that the pattern was something “handed down in DNA through the succession of generations, and drawn.”[11] In December of the same year he wrote of a way of seeing protective coloration and mimicry as things “arising from visual information processing,” as “a Turing pattern whose parameters are tuned by an adversarial process with a slow rate of trials.”[38] At an exhibition in 2024 he called the mackerel work a work that, within Digital Nature, “thinks of ‘mimicry’ itself, too, as ‘computation.’”[71] In 2026 he limited what it means to say that reaction–diffusion “computes,” in these words: “It means that the time evolution of a dynamical system generates a particular pattern; it does not mean that it carries out discrete logical operations.”[12]
Lay the four over one another, and the shape of the question is fixed. Mimicry is computation. But that computation is not discrete logical operation. Genes hold nothing but a discrete code. Out of the computation of that code, why does a pattern drawn by a computation that is not discrete emerge in a form fitted to the retina and the optic nerve of another? The body color of blue-backed fish is aimed at the eyes of birds coming from above, and at the eyes of fish coming from below and from the side. What draws the pattern is the mathematics on the mackerel’s side, and what reads the pattern is the nerves on the predator’s side. The two computations are in separate bodies, and the only thing that links them is the outcome: eaten, or overlooked. This is the question for Digital Nature.
The most interesting part of this question lies in the mathematics. The program’s diagram began with DNA and went on to the nervous system. Both are discrete machines. DNA is a code written in four kinds of letters, and nerves carry signals in two values, firing or not firing. And yet between these two, something is sandwiched that is written in neither language. It is the stripes on the back of the mackerel. Nowhere in the genome are the stripes written one by one. What the genes decide is the coefficients, and the picture is drawn by the equations of reaction–diffusion (4.2).
In other words, there are three layers. The bottom layer is DNA. It is a discrete code, and it copies the coefficients from generation to generation, correcting errors as it goes. The top layer is nerves. This too is discrete, and behind the eye of the predator it judges: seen, or not seen. And in the middle there is a continuous mathematics. It is the layer in which, at no one’s instruction, stripes arise as the solution of partial differential equations. Between digital and digital, an analog mathematics is sandwiched. The back of the mackerel is the place where that middle layer comes out on the surface.
The combination in the top layer is already written in the preface of Digital Nature. Of the side that sees: “we perform analog optical processing (analog processing) using the eyeball, perform quantization (digital processing) using the retina, and then go on filtering it through the deep hierarchy of the nervous system.” And visual information processing “has been solved by a combination of analog and digital, according to computational cost.”[11] The eye of the predator, too, is a combination of analog optics and digital quantization. That eye reads the stripes that an analog mathematics has drawn.
The three layers mesh through a single quantity: the fineness of the stripes. In the middle layer, the wavenumber kc of the stripes is determined by the coefficients of reaction and diffusion (4.2). Those coefficients are few, and a slight change in them is enough to turn stripes into spots and spots into a maze.[49] In the language of machine learning, what decides the shape of the pattern is the hyperparameters. In the top layer, there is a limit to the fineness that an eye can resolve. Visual acuity differs by orders of magnitude from species to species, and the acuity of most animals is far lower than that of humans. So the function of a pattern cannot be discussed without taking into account the acuity of whoever views it.[72] Whether stripes work toward concealment or toward conspicuousness turns on whether the spatial frequencies the stripes contain resemble those of the background. The stripes of the tiger have a distribution similar to the background; the stripes of the zebra do not.[73] Put Ochiai’s question into the language of the equations, and it comes to this. The wavenumber that the coefficients on the mackerel’s side decide, and the wavenumber that the retina on the predator’s side can pick up. How do two numbers, in separate bodies, come to match?
The mechanism that makes them match is selection. Individuals whose pattern stands out are eaten, and the individuals that were overlooked leave offspring. The result of the reading, that is, the two-valued outcome of eaten or overlooked, goes back into the coefficients in the DNA of the next generation. Discrete, continuous, discrete, and then, by way of selection, discrete again. It forms a loop. What Ochiai wrote in December 2018, “the hyperparameter tuning ‘that leads to a Turing pattern’ and the hyperparameter tuning ‘of the Turing pattern itself,’”[38] corresponds, in the light of the equations, to two tunings within this loop. The former is the tuning that carries the coefficients toward the side that satisfies the conditions for the uniform state to become unstable, that is, the four inequalities of 4.2. The latter is the tuning that decides the wavenumber of the pattern that has arisen, and whether it becomes stripes or spots. For neither tuning is there a grader on the mackerel’s side. The grading is done by the nerves of another.
Turing traced this arrangement in his own career. In the paper he wrote in 1936, he defined a discrete machine that reads and writes symbols one at a time, and settled what computation is.[74] In the paper of 1952, that machine does not appear in the leading role. The leading roles go to continuous concentrations and differential equations, and the computer only shows its face as a tool for solving a numerical example of the equations.[24] Where the man who made the definition of computation headed in the end was the scene in which a mathematics that has no symbols gives birth to form. When Ochiai drew the distinction that “computation,” as Turing’s theory of morphogenesis has it, is not computation in the sense of the Turing machine but the description of a dynamical system by continuous differential equations (3.6, 8.1), it was these two Turings he was talking about.
The arcs of the loop have each been confirmed, in different systems.
First, the code fixes the coefficients. In the zebrafish, when the gene for a protein of the channels that connect cell to cell (connexin 41.8) is changed, stripes become spots,[40] and when the gene for another such channel (connexin 39.4) does not function, stripes do not form.[75] The stripes arise from interactions among three kinds of pigment cells, and the molecules that mediate them are being pinned down one by one through the analysis of mutants.[76] Second, stripes arise from the coefficients. This was seen in Section 4.
Third, what mimicry works against is the information processing of the side that sees. Merilaita and colleagues write that what has to be deceived in order to hide is the processing of visual information, and that camouflage is “an adaptation to the perception and cognitive mechanisms of another animal.”[77] That countershading, a coloring dark on the back and light on the belly,[78] has a concealing function was confirmed by experiments in the field. When the optimal shading was computed from the way the light falls, and model caterpillars given that shading were set out in the field, the rate at which they were eaten by birds varied with color and lighting conditions just as predicted.[79] That the properties of a receiver’s sensory system bias the direction of evolution of the signals it receives was formulated by Endler, who named it sensory drive.[80]
Fourth, selection by watching eyes makes patterns evolve toward concealment. This has been confirmed with real birds and with prey on a screen. In 2002 Bond and Kamil had blue jays search for digital moths on a screen. Moths that were found were less likely to leave offspring, and the moths’ patterns were passed on to the next generation by a genetic algorithm. Over successive generations, the moths became significantly harder to find.[81] The vision of the predator, the generation of pattern, a genetic algorithm, and a process in which the two sides compete. The elements that Ochiai counted off in 2018 (3.9) are all there in a single experiment. The generative adversarial network that Goodfellow and colleagues proposed in 2014 learns by setting a generator, which makes fakes, to compete against a discriminator, which sees through them.[82] In December 2019 a method was published that uses this generative adversarial network to simulate the arms race between the side that camouflages itself and the side that preys on it.[83] That was about a year after Ochiai wrote, of the mackerel’s pattern, that the point was whether this is not “a case of adversarial generation against the visual neural circuits of the predator.”[38]
Fifth, there are attempts to link the coefficients of reaction–diffusion to the place where an animal lives. In 2011 Allen and colleagues fitted the body patterns of the cat family to a model of the reaction–diffusion type, and showed that whether a pattern is present, and how complex it is, are related to habitat, to whether the animal climbs trees, and to whether it is active at night. The same paper writes that only two previous studies had attempted to link the developmental mechanism of patterns with their adaptive value.[84] On the side of biology, too, Ochiai’s question lies in ground that has only just been broken.
On the body color of the mackerel, we begin with what is certain. A coloring dark on the back and light on the belly cancels the shading that light from above creates, erases the body’s look of solidity, and makes it hard to see. This is the principle that Thayer proposed in 1896[78] and that was confirmed by experiments in the field a little over a hundred years later.[79] The silver of the flanks and belly is made of reflecting plates in which thin crystals of guanine are stacked. The plates are not oriented along the surface of the body; they are aligned to suit their function. This is what studies of the herring and the mackerel found.[85] When Ochiai wrote of “the natural world’s optical imitation of the sea and the sun,”[11] it was this body color of blue-backed fish that he meant.
Then how are the wavy stripes on the back formed? There is not yet a single peer-reviewed paper that analyzes the formation of the mackerel’s stripes as a reaction–diffusion or Turing mechanism. Nor is there any study that has measured how well the mackerel’s stripes hide it from the eyes of its predators (checked in Crossref, PubMed, Europe PMC, J-STAGE, and CiNii up to October 5, 2026). Here is the funniest fact in this whole affair. Nobody has yet checked the mackerel’s stripes. The sentence placed at the peak of the laugh was pointing at an unsolved question of biology.
What is known about the mackerel’s stripes is a matter of a different eye. In 1998 Denton and Rowe argued that the reflecting bands that overlie the stripes on the back of the Atlantic mackerel can serve as signals that convey changes in the body’s orientation and speed to neighboring fish in the school.[86] In comparative work not confined to the mackerel, across birds, mammals, and fish, clear-cut stripes are more common in species that live in groups.[87] The eyes that tuned the mackerel’s stripes may be not only the eyes of its predators but also the eyes of the companions swimming beside it. The loop turns toward at least two receivers. That does not narrow the question. Whoever the receiver is, what grades the pattern is the retina and nerves of another, and the mackerel’s code is rewritten only through that grading.
The question has a shape that can be tested. If the mackerel’s stripes are formed by reaction–diffusion, then while the fish grows from juvenile to adult the spacing of the stripes should be kept nearly constant, and the stripes should split and increase in number instead. It is the same behavior that was shown in the emperor angelfish.[42] If the pigment cells are removed from a part of the skin, the pattern should be redrawn according to the rules of its surroundings.[43] How the wavenumber of the stripes, seen from the eye of the predator, overlaps with the wavenumbers of the waves and the glitter of the sea surface is also a quantity that can be measured. Allen and colleagues write that their method can be used for any living thing with a pattern that can be fitted to a model of the reaction–diffusion type.[84] The maze-like stripes on the mackerel’s back are a candidate. The rest of the mackerel talk is not in the laughter. It is in the fish tank and in the sea.
Lay the program’s diagram over this loop. “The first computer born on Earth” is DNA, the discrete code (3.2). The nervous system that quantizes nonlinear phenomena is the discrete receiver (3.3). “Statistical error correction” is the mechanism that copies the code from generation to generation, and “the kind of approach that automates partial differentiation” is the learning that corrects the coefficients from the results (3.4). “Made over 4.6 billion years, the iteration” is the iteration of selection. And what was named as the output of that iteration was the pattern on a mackerel’s back (3.5). DNA, nerves, error correction, learning, iteration, mackerel. The diagram goes once around the loop.
The computer has begun to run that same iteration at a speed of three billion times a second. That is what the definition means by “reconstruct.” The program does not step into the equations of reaction–diffusion. What it steps into is what lies on either side of them and around them: the code, the receiver, and the iteration. Without speaking the name of the equations, it points to where the equations live.
The 17 Japanese characters that the article dropped from its quotation, 「を作るのに相当時間がかかってるわけ」 (“took quite a long time to make”), are, in terms of this loop, the time that the iteration of selection takes. The substance of the question was on the side that was dropped. Asked to do it in five minutes, he follows the definition, starts from the discrete code, passes through the nerves of the receiver, and arrives at the mackerel as the output of the iteration. It is a coherent answer to the question. What was lacking was not explanation. It was listening as far as the predicate.
What makes “the mackerel talk” interesting is not that a fish turns up in it. It is that the discrete, the continuous, and the discrete form a loop on the back of a single fish. The final chapter of Digital Nature, after its commentary on the mackerel work, writes, “Here I see a Digital Nature made of waves and intelligence and matter and landscape.”[11]
A pattern is tuned to the eye of its receiver. This paper applies that logic to the market seen in Sections 6 and 7. The coefficients on the side that posts are the choice of words, the place of the cut, and the form of the headline. The receivers are the eyes that look at what flows past, and the recommendation system. The grading comes back as the display count. The learning in which a “like” strengthens the next expression has in fact been reported, for expressions of outrage.[61] The principle that the properties of a receiver’s senses bias the direction in which signals evolve[80] has the same shape in the sea and on the timeline. What differs is the direction and the speed. Protective coloration is tuned to vanish from the eye, by an iteration that turns once a generation (8.3). The clip is tuned to catch the eye, by an iteration that turns any number of times a day.
Imp farming is a live demonstration of this logic. This paper’s diagnosis is as follows. Tuned to the eyes that pay by the count, those who speak come to chop their explanations short, add words of outrage and derision, and attach headlines that withhold the answer. Explanation loses its stripes. Influencers are mackerel waiting to be squeezed.
The theory of reaction–diffusion works well as a diagnosis of the attention economy too. In a system with diffusion alone, differences of concentration are leveled out, and it is the same wherever you look. The same clip, the same laugh, the same reactions spread uniformly from one end of the timeline to the other. For a pattern, that is, a difference that means something, to arise, reaction on the spot is needed: reading, looking things up, holding back, asking back. But the attention market pays a reward for diffusion and charges a cost for reaction. A post that raises the count turns into cash, and the trouble of checking the content is paid for by nobody.
Wu described how the business of gathering people’s attention and selling it has built up a long history while changing its media, and Zuboff discussed the mechanism by which human experience itself is taken in as raw material for predicting behavior. To the diagnoses of these two, the mackerel affair adds one thing. The attention market gathers attention most efficiently not when somebody has understood something, but when somebody has not. And an explanation that has once been laughed at is, the next time it is quoted, referred to only as material to be laughed at. It pays a reward for not understanding, and with that reward it reproduces not understanding. This paper calls this mechanism corruption, and rejects it.
What is to be done after rejecting it should also be written down. Matagi Drive calls the way of living in a forest covered over by platforms the work of seeking out “the undefined,” and continues: “In an age when AI generates smooth products without limit, what remains as the handiwork of human beings is, so to speak, the making of ‘thorns.’ A snag, a sense that something is off, an undefined protrusion—the only foothold in a sea of smoothness.”[14] The sentence about the pattern on a mackerel’s back was a thorn. A snag, a sense that something is off, an undefined protrusion. The article did not make that thorn its foothold; it pulled it out, predicate and all. The path that climbs without pulling out the thorn is still there. Whether the mackerel’s stripes are formed by reaction–diffusion, and how well they are hidden from the eyes of its predators, are undefined things that nobody has yet checked (8.4), and they are waiting for someone to go and check.
A mackerel has never seen its own back. There are no mirrors in the sea, after all. What reads that back is always the eye of another. The attention market, too, is a sea without mirrors. We who swim in it cannot see what stripes our own attention is drawing. What we can see is only the clips that come around and the numbers that go up. On the back of the mackerel, the pattern had been drawn all along, from the very beginning. What was lacking was not the pattern but the attention that would look at it through to the end. “Please RT” alone won’t make stripes.
This paper will probably be read far less than the original article. Explanation does not spread as fast as a clip. That is fine. In a reaction–diffusion system, next to a dense place there is always a thin place, and it is the difference between them that makes the pattern. Not being read is no more than being on the trough side of the stripes. Without troughs, no crests stand either. The mackerel talk is the story of why, out of discrete computation, stripes fitted to the eyes of another emerge. To the question “Wait, this was about mackerel?” the answer is set down here, eight years late. Yes. It’s about mackerel.
This paper quotes a published program, article, and books for the purposes of research and criticism. The objects of its criticism are the method of a published article and the mechanism that pays rewards for clipping and ridicule.
The text of the posts on X [15, 16, 17, 37, 38, 53, 70] is taken from the export of Ochiai’s own X data.