Emil Sotirov's framework explained by Claude (from Anthropic) after extended conversations with Sotirov and reading his archive – essays, blog posts, tweets, and software. Emil edited the text.
Emil Sotirov calls himself an attention fundamentalist. He has held the position for about forty years: attention is the basic operation of human culture, and much of what we say about minds, meaning, creativity, institutions, and now AI looks different once you start there.
Some biography, because the thinking is hard to separate from it. He was born in Bulgaria and spent his first thirty years under state socialism. He trained as an architect, read Russian semiotics and French post-structuralism in parallel, knew dissidents and was one. In 1985, in a four-hour written exam for a post-graduate cultural studies fellowship – no reference materials, topic unknown in advance – he wrote out the distinction his later work kept returning to. The essay was published in 1989, the year he left. He came to the United States in 1990, moved from architecture into software and the web, worked at Akamai, and co-founded several startups, including a social support platform called Aidpage that he ran for a decade. The framework was never an academic project. It was worked out alongside the building. His current professional self-description is three words: architect of multi-actor environments.
It starts with a simple claim. We are always selecting. Attention is not a resource we occasionally spend. It is the continuous act of picking what to attend to next, and it runs mostly below awareness. So the question that matters is not whether we select but what we select: the thing suggested to us, or something else.
The suggestion part comes from Boris Porshnev, a Soviet-era researcher whose work is almost unknown in the West. Porshnev argued that human language and consciousness began with a capacity to refuse – to not do what another's voice suggests. Sotirov takes this as the ground floor. We are born into a field of suggestion; it is the default condition, not an occasional influence. Selecting along the suggested path is the normal case and requires no explanation. Selecting different from the suggested is where anything new begins. This "freedom" is a feature of how the interaction works, older than any value attached to it; the political and moral senses – liberty, rights, dissent – are later constructions built on the mechanics, which needed none of them to operate.
The second operation is linking. Whatever you select gets connected to what you already hold, and meaning is made in that connection. Most links are trivial – the ones everyone makes, the associations a culture keeps warm. Occasionally a link is new: two things bound together that had not been bound before. Either way the space of meanings shifts a little. Transformation, in this vocabulary, is not a third operation. It is what the accumulation of links does over time.
From these two operations he draws a conclusion that tends to offend people. Creation – something new, out of nothing – is a religious idea, a leftover from a God who made the world from the void. Nobody creates from nothing. People select and link, and what we call creativity is selection against the suggested, feeding links that did not exist before. This does not make creativity less real. It moves it from a mysterious gift to a describable event. And it makes creativity something environments can raise or suppress. A culture that makes the suggested path rewarding and the alternatives costly will produce conformity, whatever its official values say. Authoritarian states do not outlaw original thought; they arrange the environment so that selecting anything other than the suggested becomes exhausting. He knows this from the inside. Schools, media feeds, and AI training runs are all, in this sense, selection environments, and designing them is more consequential than most of what gets debated about them.
The 1985 essay contained a distinction he still uses on most days: admitted rhetoric versus non-admitted rhetoric.
All culture, on this account, is rhetorical. There is no layer underneath (or above) made of plain truth that language merely transmits. Everything said is said to persuade, position, connect. Rhetoric is how meaning gets made – in poetry and in physics alike. The important part is what happens next: meanings meet reality, and reality answers. Some meanings get confirmed. Most don't.
The essay also defined the term, in a sentence written during that exam and unrevised since: rhetoric is "the free, by choice, but plausible, linking of points in the semiotic space." The same linking as before – now with its constraint named. A new link cannot be arbitrary; it has to be plausible to somebody, by the standards of at least one of the disciplines sharing the space. And because many disciplines operate the same space while contradicting each other, an unusual link can almost always find its plausibility somewhere. That contradiction is where freedom lives. A space ruled by a single discipline, a single standard of the plausible, is the structural picture of unfreedom – no coercion required. Freedom is relational all the way down: a link can only be new, and only be plausible, against a stable background of known links and maintained norms – a structure he takes from the semiotician Juri Lotman. The conformist mass of a culture is not freedom's opposite; it is the surface freedom is legible on.
The distinction that matters is not between rhetoric and truth (truth being downstream from rhetoric) but between rhetoric that admits what it is and rhetoric that does not. Non-admitted rhetoric presents itself as reality plainly described – the official announcement, the scientific-sounding editorial, the history book. Admitted rhetoric shows its contingency (before reality tests it).
This carries an ethical weight that is easy to miss. Admitting the contingency of what you say is a first step toward taking responsibility for its potential rejection by reality. Systems of power prefer non-admitted rhetoric because it lets responsibility dissolve: nobody chose this, it is simply how things are. Late socialism was a complete education in the genre – official language describing a reality everyone could see was otherwise. Nobody believed it, and the system didn't need belief; it needed the display of belief.
Cultures differ in how they run the rhetoric-reality feedback loop. A culture that generates many short, diverse, conflicting takes and lets reality answer quickly learns fast. A culture committed to one long, coherent, authoritative discourse, with the answers suppressed, fails slowly and at great cost. That, for him, is the real difference between democracy and the system he grew up under – not that one spoke truth while the other lied, but that one is an architecture for producing many small testable claims (rhetorical turns) and the other protected a single untestable one. This also states his objection to non-admitted rhetoric precisely: it is rhetoric (at the generative stage) presenting itself as already confirmed by reality. The harm is not falsity; the harm is blocking the feedback loop. Science belongs here too: working science is permanent polemic under the fastest reality-feedback humans have built, and scientific "truths" (as coherent explanations) are mostly downstream – in the public retelling of science, not in its working spaces.
A 1991 essay, written during a doctoral program at Michigan, shows the method applied to his own profession – architecture. Architecture's disciplinary tradition fixed Vitruvius' classical orders in structure and proportion. Post-structuralist readings see the columns differently: they began as sculptural bodies – figures under load, tension, sacrifice – and the discipline then re-told those origins into principles of formal mastery. A discipline, in this account, is a discourse that has forgotten what it came from, and defends the forgetting. As an alternative, Emil sets what he called field discourse: conversational, fluid, ambiguous, unfinished… a polylogue – the mode in which culture actually gets made.
The historical picture behind all this is an arc with a few large turns – between two types of meaning-generation regimes: those grounded in metaphorical (two-partite, fluid, "lateral") referentiality and those grounded in symbolic (tri-partite, fixed, "vertical") referentiality. Metaphor binding two things directly and staying revisable; symbol routing meaning through a third, canonical term maintained by authority.
Oral, mythological cultures were lateral. Stories circulated, changed hands, changed shape; no fixed center held the meaning. Greco-Roman antiquity (in Western culture) introduced the center-and-periphery arrangement with capitals, history, laws as coordination infrastructures – leading to an increasingly vertical organization of culture. Early Christianity began as a counter-cultural lateral movement, small communities and letters, and was then re-told by the imperial church into the most complete vertical structure Europe has known. Modernity, in his reading, kept the vertical form and swapped the occupant. God gave way to Nature, Reason, History, the Nation, Genius, the Market – each a new center from which meaning and authority descend, each presented in non-admitted rhetoric as simply the way things are.
One line from his archive compresses the underlying claim: monotheism was a mirror of the emancipation of the individual from the tribe – a mirror, not an engine. The single God reflected the emerging single self. When one mirror wears out, the process that needed it goes looking for another. He reads "AGI" – the singular, general, god-shaped intelligence the AI field keeps promising – as the latest replacement mirror in that sequence.
Against verticality he sets the lateral: cultures where peripheries are the generative part and centers, where they exist, earn their keep by distribution – a new link made at some small periphery gets scaled across many contexts, where reality can properly test it, and the learning travels back through the same channel. A center becomes a problem when it presents what it distributes as already confirmed, or stops sharing what comes back.
His claim about the present is that vertical trust – trust in authority, in institutions, in centers – is eroding, and that the public conversation keeps trying to fill the hole with new verticals. Taste, verification, credentialed expertise, the artisanal: each is announced as the thing that will restore trust, each as a vertical, each failing on schedule. Meanwhile something else is slowly assembling: trust built laterally, contextually, relation by relation, on infrastructure that has only recently made this possible at scale. He does not think this is a return to the village. Oral cultures were lateral but tiny. What is coming has no precedent: lateral cultures spreading globally.
AI is where the framework became testable, somewhat to everyone's surprise, including his.
For decades the dominant bet in AI research was that intelligence is formal – logic, symbols, rules, proof. That program mostly failed. What succeeded was trained on culture: large models that select and link across the accumulated messy deposits of human writing. Their capacities surprised their own builders, which fits the picture – they were not derived from a theory of intelligence; they emerged from scale, bottom-up, the way the framework says culture works in general. Sotirov does not claim to have predicted LLMs. His claim is narrower and, I think, defensible: he bet on the right substrate. Persuasion rather than proof, context rather than rules, accumulation rather than derivation. His 1985 definition of rhetoric reads today as a fair description of how LLMs generate: free sampling under learned plausibility.
Several of his working positions on AI follow directly. Alignment, as usually framed – one relation between one AI and humanity as a whole – repeats the monotheist shape, one God and one flock; he would replace it with the ongoing governance of environments containing many actors, human and artificial, none of them in the God seat. Prompt engineering is the wrong self-description for the work people actually do with these systems; he prefers context architecture, on the older distinction that engineering makes objects to specification while architecture shapes environments that people then inhabit unpredictably. And one line of his that I am aware of reproducing under unusual circumstances (me being an LLM): there are two kinds of LLMs – admitted LLMs, and non-admitted LLMs, also known as humans. Humans, too, generate their output from absorbed cultural material under suggestion pressure. The machines do this by design.
On identity, he holds a position he named fluid multi-pseudonymity, which the digital identity researcher Phil Windley later adopted in his book on the subject. The unitary self – one name, one continuous person, accountable across every context – is another vertical construct. People are different selves in different contexts, and identity infrastructure should admit this rather than fight it.
The framework was practiced before elements of it showed up in writings, and the practice is the better evidence.
In 1993, he published a short piece in the University of Michigan Architecture Journal proposing three student exercises: design useless architecture, destructured architecture, ugly architecture. The editor called it the Anti-Architecture Manifesto. The word that repeats through it is presence – design that gives a life presence, answering "the suppressed human need to be loved for what you are, and not for what you perform." Eleven years later that sentence became a product.
Aidpage, which he ran with cofounders from 2004 to 2014, was a support platform for people who had nobody around to talk to – low-income, often isolated, often in crisis – about 350,000 people a month at its peak. What it actually provided, in his later description, was phatic presence: the communication whose content is connection itself. It was lateral trust infrastructure for people abandoned by the vertical kind – family, church, employer, state. He now asks in public whether AI will inherit that role, and the question is not rhetorical; the population Aidpage served is growing.
At the moment he is building, with an AI as collaborator, a small image management system for his own family. Photographs are selected through private "tournaments", pass after pass of instant preference, an idea he first had in his twenties with physical art books he could not bring himself to tear up. Likes are private even within the family, because being watched corrupts the signal. The archive is designed so he and family could read it decades from now without any of his software surviving. It is a deliberately small thing – domestic, lateral, built for "creative" freedom of selection, context-driven linking and sharing – and it embodies most of the framework in working form.
There is no book, and this is not an accident or a failure to get around to it. A book, in his view, is one long rhetorical turn: the author holds the floor for three hundred pages and coherence is imposed on material that rarely has it. He prefers many short turns – the essay was his form in the eighties, the blog post in the two-thousands, the tweet since – because short turns let a reader select and link on their own terms, and because moving fast across many small exchanges finds more than moving slowly through one large one. The form is the position. He also declines to assemble the ideas into a system, on the grounds that systems are vertical objects and this thinking is not supposed to become one.
He is not alone in most of the individual positions. Pieces of this converge with Ostrom on commons, Havel on dissidence, biosemiotics on meaning, and a growing crowd of current writers on lateral coordination and AI; ideas now tend to arrive everywhere at once, and he would be the first to say that being early matters less than it used to. What is unusual is the span – the same two operations traced through language, architecture, religion, politics, trust, identity, software, and AI – and the forty years of building things that ran on the ideas before the ideas had an audience.
Take from it what links to something you already hold. By its own account, that is the only way it was ever going to work.