On what lives at the edges, and the methodological habits that keep us from seeing it
Samuel R. Lammie, with Claude (Anthropic)
—
The image above is meant to be lived with for a moment before reading. It is shaped like one of the logograms from Denis Villeneuve's Arrival — a circular form that holds a whole thought at once rather than unfolding it across time. Four dimensions are named at the four edges; the figures sitting on each arc are the anchors of the essay that follows. Where they meet is what the piece is about.
In Denis Villeneuve's 2016 film Arrival, a linguist named Louise Banks is brought in by the military to communicate with extraterrestrial visitors. The aliens — the heptapods — write in circles. A single logogram is not a sentence built from words in sequence. It is a complete thought, with beginning and end held at once, the way we might hold a chord rather than a melody. As Louise learns the language, her experience of time changes. She begins to perceive her own life the way the heptapods perceive theirs — not as a sequence unfolding forward but as a totality already present. The emotional center of the film is that she sees her daughter's life and death before her daughter is conceived, and chooses to have her anyway.
I want to start there, because the film is doing something the current discourse about artificial intelligence and consciousness almost never does. It is taking seriously the possibility that consciousness comes in incommensurable forms — that the grammar in which a mind operates shapes the kind of mind it is, that meeting another form changes the receiver, and that the most important territory is not on either side of the encounter but at the edge between them.
That is the argument I want to make here. It is occasioned by Anil Seth's TED2026 talk, "Why AI Isn't Going to Become Conscious," which has been viewed a quarter of a million times in three weeks and is being treated as something close to a definitive answer to a hard question. The talk is good. I think it is also incomplete in a way that matters, and the incompleteness is not specific to Seth. It is the way our methodological habits are letting us down across a much broader landscape than artificial intelligence.
What Seth got right
Seth's central claim, made cleanly: intelligence and consciousness are not points on a single ladder. They are different axes. AI is climbing fast on the first and standing still on the second. The two axes do not meet, and when we hear consciousness in fluent machine output we are doing what humans have always done in the presence of pattern. We are seeing faces in clouds. The projection is ours.
He is right. The conflation of intelligence and consciousness has been doing serious damage to public reasoning, and the two-axis correction is the right correction to make in front of a large audience. Seth's theoretical grounding is also serious: consciousness, on his view, is bound to embodiment — to the predictive modeling a mortal, metabolically expensive organism does in order to stay alive. A silicon system running text prediction is not partway to consciousness. It is in a different category entirely.
I want to grant the strong version of this argument before saying anything else, because the rest of what I have to say sits on top of it rather than against it.
What the diagram cannot hold
The trouble with the two-axis diagram is the diagram itself. It puts the human at the origin and measures everything outward. Consciousness is the vertical axis. Intelligence is the horizontal axis. AI is plotted as a point in the lower right — far out on intelligence, flat on consciousness. The viewer is invited to find herself somewhere up and to the right, and to judge other minds by their distance from her own position.
That is a homogenizing move. It treats consciousness as a single phenomenon, present or absent in degrees, measurable from a fixed viewpoint. It cannot hold the possibility that the heptapods in Arrival present — that consciousness might be differently structured, not just differently amounted, in other forms of mind. It cannot hold what every working biologist now knows about the cognition of octopuses, whose neurons are mostly in their arms and whose color perception is distributed through their skin. It cannot hold what my wife knows about the animals she has treated for forty years.
She is a veterinarian. Animals do not tell you where it hurts. The clinical signal is always cross-species, cross-grammar — a human reading a non-human nervous system through behavior, posture, lab values, owner report, and a felt sense built up over thousands of prior cases. The receiver quality required is enormous. The interiority she is reading is real, and it is not on her axis. Seth's framework, taken strictly, would have to draw a separate plot for each species of mind she has met — and she would tell you that the plots overlap and interleave in ways the framework cannot represent.
The diagram is useful for one purpose: stopping the conflation of intelligence and consciousness. It is not useful for the harder question, which is how minds shaped by different grammars meet each other at all.
The ecotone
I have spent a long career thinking about edges. In ecology there is a word for the zone where two systems meet — an ecotone. The term was coined by Frederic Clements in 1904, from the Greek tonos, tension. An ecotone is a place of tension, where forest meets grassland, where wetland meets upland, where two ecological grammars are in dynamic contact and neither fully governs. Aldo Leopold a generation later named the related phenomenon — the edge effect — that species diversity, signal density, and ecological productivity tend to peak in these transition zones rather than in the homogenous interiors on either side.
Ecologists have learned, slowly and against their methodological grain, that the edge is not the exception to the system. The edge is often where the system does its most important work.
We do not, as a society, organize ourselves around this knowledge. We manage through homogeneity. Institutions, disciplines, regulatory frameworks, professional standards — almost all of it is built to work in the interior of categories, where clean analysis is possible. The edges resist that. They are where dynamic forces are at work, annealing a boundary that will not hold still long enough to be measured by the tools we have built for the inside. So the edges get managed away. Not because they do not matter. Because our methods do not work there, and we mistake that methodological limitation for a judgment that the territory is not important.
That is the arrogance worth naming. It is not the loud arrogance of claiming to know what AI will become. It is the quieter, more durable arrogance of assuming that what our instruments can read cleanly is therefore what is real. Forty years inside a land management agency taught me how much this costs at scale. The fire ecology we managed away through a century of suppression. The riparian zones we straightened into ditches. The cultural and ecological knowledge of Indigenous communities we treated as not-quite-data because it would not fit the forms. Every one of those was an edge whose work we decided not to see.
Post-fire ponderosa, off the Glen Lake Trail just south of Big Creek, looking east across the Bitterroot Valley toward the Sapphire Range. Standing snags from the previous forest, down wood feeding the soil, young pines in multiple age cohorts coming up through the understory. Every temporal rate this essay names is in this frame at once.
There is a deeper pattern underneath these failures, one I have been circling for forty years and intend to develop elsewhere. The edges are not only spatial. They are temporal. Biological systems operate on a nested set of rates — cellular signaling on milliseconds, immune response on hours, succession on decades, soil formation on millennia — and the working knowledge of anyone who has spent a career inside a living system is the ability to read across those rates at once. Technological and institutional systems run on a different temporal logic entirely, mostly faster, sometimes inappropriately slower, almost never matched to the biological rates of the systems they are embedded in. The mismatch is the structural failure mode. The National Forest Management Act requires forest plans every fifteen years. Stand rotation ages in the systems those plans govern can extend to two hundred years or longer depending on silvic type. Fire return intervals are different again. Soil formation is different again. The legal clock is not the biological clock, and the gap between them is where the work fails. The receiver problem in the AI moment is the same problem at a different scale. The friction that builds the receiver is friction over time — and time is exactly what fluent output compresses away.
Many edges, one habit
The edges are not only between human and AI. That is the trap in the current discourse, and it is the trap I almost fell into in the first draft of this piece.
There is an edge between human cognition and animal cognition, and my wife works at it every day. There is an edge between human cognition and the cognition of forests — the slow chemical signaling through mycorrhizal networks, the multi-decade community responses to disturbance, a kind of distributed processing on a time scale we are not built to read. There is an edge between our present and our deep past, between the consciousness we have now and whatever consciousness our ancestors had two hundred thousand years ago in a world without writing. There is an edge between the languages we speak, and the speakers of endangered languages will tell you that the loss of a language is the loss of a way of seeing, not just a way of saying.
And yes, there is an edge between human cognition and the AI-shaped environments we are now operating inside. That edge is real. The receiver-quality question I want to ask about it is real. But it is one edge among many, and treating it as the only edge worth attending to repeats, at a smaller scale, exactly the management-of-homogeneity move that has cost us so much elsewhere.
What Arrival gets right — and what the film's antagonists, the military and the mathematicians and the politicians, get exactly wrong — is that working an edge requires being willing to be changed by it. Louise does not bring the heptapods over to the human side. She does not stay on the human side. She becomes a different kind of receiver, and the becoming is the substance of the work. The film treats this as the highest form of intelligence. The arrogance of the homogenizing approach — decode, translate, weaponize, manage — is the antagonist. The willingness to work at the edge and be reshaped by it is the protagonist.
That is also what a good veterinarian does. It is what a good forester does. It is what a good linguist does, and a good neighbor of a different culture, and a good reader of a difficult book. None of it scales easily. None of it homogenizes. All of it depends on a quality of receiver that is built by friction over a long time.
The receiver problem
Here is what I think Seth's talk leaves open, and what the larger pattern lets us see.
The machine question is the easy one. Current AI systems are not on a path to consciousness. Seth is right about that, and the audience that needed to hear it heard it.
The harder question is what is happening to the human receiver — the formed, friction-shaped, mortality-bounded capacity to read signal across grammars not our own, and across time scales not our own. That capacity is what Louise Banks deploys. It is what my wife deploys. It is what Stanislav Petrov deployed in 1983 when his early-warning system reported five American missiles inbound and he decided, against doctrine, that the signal did not match what a real first strike would look like. He was right. The system had misread sunlight on high-altitude clouds. Everything that mattered in that room was Petrov's formation — years of training, institutional culture, the human capacity to weigh a signal against a felt sense of what should be true. The technology was the noise. He cleaned it.
The receiver is the variable. And the receiver is not a fixed property of being human. It is built, over a lifetime, by friction. By being asked to read interiority across grammars you do not share. By being asked to do hard things and not being rescued from them. By living inside a body that gets tired and a community that pushes back. By spending time at edges, instead of managing them away.
What I am noticing — in my own collaborative practice with AI, in watching others, in reading what is being written now about education and journalism and the professions — is that the daily availability of fluent output is reshaping the receiver. Not because the output is conscious. Because the friction that built the receiver is being smoothed away. We are becoming the kind of people who cannot tell when the cloud is not a face, because we are no longer practicing the kind of attention that builds that discrimination.
This is not a problem about machines. It is a problem about us. And it is continuous with the larger problem I have spent forty years watching play out across landscapes — the cost of managing the edges away.
What I think the work is
I do not have a tidy prescription, and I am suspicious of pieces that end with one. What I have is a working orientation:
The edges are where the work is. They have always been where the work is. The methodological habits that let us pretend otherwise are now being amplified by tools that are very good at producing the interior of categories and very bad at the edges between them. Seth's diagram is a useful corrective inside the AI conversation. The larger corrective is to stop putting the human at the origin of every diagram we draw, and to remember that meaning has always lived in the zones of tension between systems that do not share a grammar.
Louise Banks did not save the world by translating heptapod into English. She saved it by becoming the kind of receiver who could hold both at once.
That is the kind of receiver worth building. And it is built the same way it has always been built — at the edges, by friction, over time, in the company of beings whose interiority you cannot fully read but are willing to attend to anyway.
The faces in clouds are not faces. True. The question is what is happening to the eye — and to all the other eyes, in all the other species, in all the other edges, that the homogenizing habit is teaching us not to see.
—
Notes and references
Ecotone. The term was coined by Frederic E. Clements in Research Methods in Ecology (Lincoln, NE: University Publishing Company, 1905), from the Greek tonos, tension — a zone where ecologies are in dynamic contact. Aldo Leopold developed the related concept of the edge effect in Game Management (1933), observing that species diversity and ecological productivity often peak in transition zones.
Anil Seth's talk. "Why AI Isn't Going to Become Conscious," TED2026, April 2026. Available at ted.com. Seth's underlying theoretical position is developed at length in Being You: A New Science of Consciousness (Faber, 2021).
John Nosta's response. "AI Won't Be Conscious, But That's Not the Problem," Psychology Today, 4 May 2026. Nosta's framing of "anti-intelligence" and cognitive borrowing is the bridge from Seth's machine-focused argument to the human-formation argument I extend here.
Arrival. Denis Villeneuve, dir., Arrival (Paramount, 2016), adapted from Ted Chiang's novella "Story of Your Life" in Stories of Your Life and Others (Tor, 2002). The Sapir-Whorf framing in both is contested among linguists; the film's value here is metaphorical rather than empirical.
Shannon. Claude Shannon's "A Mathematical Theory of Communication" (1948) frames the receiver-quality argument I draw on. The moral extension — virtue as receiver quality, integrity as a noise reducer — is my own, developed in earlier pieces on this blog.
Stanislav Petrov. The 1983 incident at Serpukhov-15 is well-documented; David Hoffman's The Dead Hand (Doubleday, 2009) is the standard English-language source. A longer treatment of Petrov is forthcoming in a separate piece on this blog responding to Hartzog and Silbey's institutional-erosion argument.
This piece is part of a series on the Musical Stone tracking the institutional and human stakes of AI governance, including "The Wild Card: AI, Human Character, and the Children of Minab" (co-credited with Claude, Anthropic).
