Peter Ross's letter, "Mystifying Marx in the Age of AI,” responds to Nicolas D. Villarreal's October 22nd essay, "Marxism in the Age of AI,” in terms of value theory. Surplus value, Ross argues, is the product neither of a transhistorical human creativity nor of machine intelligence; rather, it is the historically specific form a social surplus takes when production is based on wage labor.

Ross sets aside the larger portion of Villarreal's essay: "the interminable detours into '19th-century semiotics,' the 'material negativity' of the 'signifier,' and the 'Landauer heat' emitted by the brain." But those detours ground the political-economic claims that Ross rightly, if mechanistically, criticizes. Villarreal's argument runs like this: thinking is the sorting of inputs by difference; anything that sorts this way thinks, regardless of its material substrate; this capacity can be measured in FLOPs (floating-point operations); machines will soon exceed human capacity; therefore, a machine will eventually perform "any act of labor a human could." On this view, labor is merely transhistorical computation. Ross counters this argument using value theory. But Villarreal's claim also demands a psychological response—and historical materialism has a good one to offer.

That answer is activity theory, sometimes known as cultural-historical activity theory (CHAT). Its roots lie in the work of the Soviet psychologist Lev Vygotsky, who died in 1934 at age 37 and is now widely taught in U.S. education courses stripped of his Marxism. Under his student and collaborator Aleksei Leontiev, activity theory developed into the preeminent school of Soviet psychology by the 1970s—the fullest expression of what French philosopher Lucien Sève called "scientific humanism." Though much of modern "CHAT" has abandoned these origins, its core concepts remain invaluable for resolving open questions in the neuroscience of tool use and language, anthropology, evolutionary biology, and our understanding of AI.

Villarreal frames his analysis as stemming from "structural Marxism and structuralism more generally, beginning with the works of Louis Althusser on ideology." Althusser argued that the mature Marx broke decisively with humanism, asserting that ideology "interpellates" individuals as subjects—hailing them much like a police officer hailing someone on the street. He famously held that "ideology in general has no history." Villarreal translates this mechanism into computational terms: the subject emerges when the signs for "I" are mapped "into a particular set of coordinates within a semantic field." In his words, "You too are a subject created by a particular semantic field, its topology, and a particular set of coordinates that mark the 'you.'"

In the 1960s, Sève challenged Althusser from within the French Communist Party. He read Marx's sixth thesis on Feuerbach ("the human essence is no abstraction inherent in each single individual. In its reality it is the ensemble of the social relations") as a foundational mandate for psychology. If the human essence resides outside the individual within social relations, psychology must explain how individuals appropriate that essence through their own activity. Leontiev cited Sève’s work approvingly. It is precisely this active process of appropriation that Villarreal's framework leaves out.

Villarreal's primary example of "material negativity" is the stop sign. His account relies on Ferdinand de Saussure, the founder of structural linguistics whose work also informed Althusser. For Saussure, a sign possesses no positive content of its own; its value derives from its differences from other signs within a system, studied as a snapshot, apart from history and from acts of speaking. A stop sign is a stop sign simply because it is not a yield sign or a traffic light. Within this abstract frame, the description holds: the vector geometry of a language model's word embeddings does encode differences of this kind, and the models function accordingly.

But stop at a stop sign, look left and right, and ask yourself: what actually gives it meaning? The structural account describes a real, but purely abstract, slice of reality. It leaves out the road, the car, the other car, the other road, the dashboard clock, the cop car, pothole, brakes, Iraq, Henry Ford.

Leontiev articulated the general principle: "The vehicle of meaning is language, but language is not the demiurge of meaning. Concealed behind linguistic meanings… are socially evolved modes of action (operations), in the process of which people change and cognize objective reality." He distinguished between a sign's meaning—which is fixed in social practice and carried by language—and its personal sense—what it signifies in the concrete life of the individual interacting with it. The stop sign means "stop" to anyone driving a vehicle. What it means to me on a given morning when I am running late is its personal sense.

The semantic plane has a concrete history: the history of our species. Drawing on the archaeologists Dietrich Stout and John J. Shea and the psychologist Michael Tomasello, I take it to have emerged through the coevolution of increasingly complex and eventually obligatory toolmaking and teaching, shared intentionality, and language. These capacities developed together, each the condition and product of the others. Language itself arose from the communicative demands of cooperative activity, initially helping people coordinate their work. The stop sign differs from the yield sign because drivers are required to execute distinct physical actions at different intersections. The system of differences Saussure described is real, but it is the frozen deposit of past activity whose origins have been forgotten.

For most of human history, meanings were carried by people, in use and teaching. However, when verbal meanings carry abstractions, those concepts can drift from the practical activity that birthed them. Thought separates from material practice through a division of labor that assigns mental and manual activity to different classes. From handwriting to print, and now to the digitized corpora that train large language models, the semantic plane has come to exist as an archive of meanings detached from lived activity, stored on server farms and owned by corporations. It is a real abstraction—an abstraction executed in physical practice, just as abstract labor is under capitalism. Marx observed that when people exchange their products as commodities, they equate their diverse labor without realizing it; the market abstracts behind their backs. AI training performs a structurally similar operation on human speech.

Villarreal mistakes this historical product for the intrinsic nature of thought. His "material negativity" is Saussurean: a sign is defined by what other signs are not, captured in a static snapshot. Labor's negativity, by contrast, is transformative. It negates what is given—the raw stone or material—and creates something new, guided by a conscious goal and a driving motive. A classifier like the perceptron, on which Villarreal builds his case, merely sorts inputs into predefined categories; by itself it changes nothing in the material world.

Vygotsky offered a fundamentally different model of meaning—one grounded in connection and generalization rather than difference. In Thinking and Speech (1934), on the basis of his concept-formation experiments, Vygotsky argued that children first organize objects into "complexes": groups bound by concrete, empirical connections (shared color, shape, co-use, or spatial proximity) rather than abstract definitions. As Vygotsky put it, "If empirically present, any connection is sufficient" to group items together. He likened a complex to a family name: one belongs to the Smiths through actual kinship, not by meeting a formal definition.

One variant, the "collection," groups items by "their co-participation in a single practical operation," the kind of grouping adults still make, Vygotsky noted, when they speak of "dishes" or "clothes." In Vygotsky’s account, true concepts emerge later in ontogeny, when abstracted features are resynthesized during adolescence around "a single, essential, and uniform connection." Difference enters here—in the abstraction that isolates features—but as a moment of generalization, not its underlying principle.

While Vygotsky’s concepts, like Saussure’s signs, exist only within a system, Vygotsky's system is inherently hierarchical and dynamic (e.g., "rose" under "flower," "flower" under "plant"). Vygotsky faulted the structural psychology of his day for clinging to "the notion that the word—by its nature—does not develop."

This distinction has direct implications for modern language models. Distributional learning, which knows a word "by the company it keeps," in the British linguist J. R. Firth’s phrase, produces generalizations from statistical regularities in linguistic contexts. This resembles the logic of Vygotsky’s complex: connections are established through whatever empirical relations are available, rather than through an abstract definition. But the model encounters these relations in a textual archive, not through the practical and perceptual activity of a developing child. Saussure's system of differences describes the geometry a language model ends up with; Vygotsky's complex resembles the kind of generalization that produces it. Because LLMs train mostly on adult text, their outputs mimic the contours of adult concepts. Vygotsky observed the same phenomenon in children, noting that a child "does not create the complex that corresponds with the meaning of a word but finds it ready made" in adult speech. The child’s complexes mirror adult usage while operating under entirely different cognitive mechanics.

Vygotsky called this a “pseudoconcept”: "a shadow of the concept... constructed in accordance with entirely different laws than the true concept." He warned that this functional overlap easily tricks observers: a pseudoconcept "is as similar to true concept as the whale is to the fish." That a model works does not explain how it works. For a child, the pseudoconcept acts as a developmental bridge: "verbal interaction with adults becomes the motive force behind the development of the child’s concepts." Whether LLM training contains anything analogous is a question Villarreal’s framework cannot adequately answer. He allows that continual learning reshapes a semantic field, as a model "begins to correlate its actual actions and history with this sign for the self." But new correlations rearrange the field; nothing in the account explains how the kind of generalization would change. Vygotsky made the same objection to associationism, on which an association "may undergo a series of quantitative and external changes" but cannot "change its internal psychological nature."

Vygotsky also answered the linguistics of his day directly. In chapter one of Thinking and Speech, he targeted the view—close to Saussure’s doctrine of the arbitrary sign—that "sound and meaning lie unconnected in the word," arguing that this split yielded "only the most pathetic results." Separated from thought, speech sounds become indistinguishable from ambient noise; separated from sound, meaning retreats into an abstraction "existing and developing in complete isolation from its material carrier."

Instead, Vygotsky proposed analyzing language by units rather than decomposed elements. A unit "possesses all the basic characteristics of the whole," just as a water molecule retains the properties of water that isolated hydrogen and oxygen atoms lose. He found his model in the emerging structural phonology, which treats the phoneme as the basic unit of speech because it preserves "the signifying function of sound." Phonemes are defined by their contrasts, so difference does real work here, but at the level of sound and strictly in the service of meaning.

For human thought and speech, Vygotsky identified this core unit "in the inner aspect of the word, in its meaning." Word meaning integrates sound with thought, and generalization with communication. To state "I am cold," one must "relate my experience to a specific class of experiences that are known to my partner." As Vygotsky summarized: "Social interaction presupposes generalization... generalization becomes possible only with the development of social interaction." Meaning is neither a differential position nor a static label; it is a dynamic generalization forged through shared activity.

In the final chapter of Thinking and Speech, Vygotsky traces a movement that takes place whenever someone uses language: "The relationship of thought to word is not a thing but a process, a movement from thought to word and from word to thought." A thought is present all at once, whereas speech unfolds sequentially. This movement originates in motive and passes through thought and inner speech before reaching word meanings and speech sounds. Thought itself transforms along the way: "Thought is restructured as it is transformed into speech. It is not expressed but completed in the word."

This dynamic unity contrasts sharply with Saussure’s model. Saussure famously pictured language as a sheet of paper with thought on one side and sound on the other, such that a cut through one is a cut through both. For Saussure, thought apart from language was an amorphous mass that language alone delineates. Vygotsky argued instead that thought has independent roots in practical intellect and motive, noting that thought and word are "not cut from a single mold": in child development, meaning moves from sentence to word while acoustic sound moves from word to sentence. Saussure made the linguistic system, rather than speech, the primary object of study, setting speech aside as individual and accidental. Vygotsky reversed this priority: "Isolated in the lexicon, the word has only one meaning," but that meaning "is nothing more than a potential that can only be realized in living speech." Ultimately, Villarreal’s semantic field resurrects what Vygotsky challenged: a sign defined by its place in a system, severed from the communicative practices that give it life.

Villarreal’s semiotics has political-economic implications. Saussure observed that linguistics, like political economy, deals with value—a system equating items of different orders (labor and wages in economics; signified and signifier in linguistics). While the Italian semiotician Augusto Ponzio’s claim that Saussure's thinking on value was influenced by the Lausanne marginalists Léon Walras and Vilfredo Pareto is disputed, Saussure’s explicit example nevertheless mirrors the equation Marx analyzed in Capital as the wage-form. The wage presents itself as the price of labor rather than of labor-power, ensuring that "all labour appears as paid labour" while surplus labor vanishes from view. Saussure’s analogy takes that surface appearance as its conceptual model. When Villarreal subsequently labels profit "a pure signifier," the analogy turns full circle, returning to economics with surplus labor still concealed.

In his final book, Activity, Consciousness, and Personality (1975), Leontiev argued that twentieth-century psychology shared a flawed "binomial plan of analysis": a stimulus acts on an organism, and a response follows (S —> R). Attempts to repair this formula merely complicated it from within—some inserted "intervening variables," while others integrated cybernetic feedback loops. The anthropologist Leslie White introduced culture as mediated by signs, to which Leontiev responded that the formula remained unchanged: "now the stimulus is interpreted as a 'cultural stimulus.'" Leontiev's alternative was to introduce the subject's activity—with its conditions, goals, means, and motives—as the indispensable third term.

In Anglo-American thought, the intervening variable reappeared in functionalism, which defines a mental state purely by its causal role between inputs and outputs, regardless of substrate. In continental Europe, the sign became the central mediator of structuralism. Both movements drew heavily on the emerging sciences of computation and communication. Villarreal synthesizes them into a single premise: "the materiality of the sign is the materiality of a machine which can accomplish this [sorting] task, whether it be a human brain, or, in the case of AI, an artificial neural network." The first half of this claim is Saussure; the second is functionalism. Together, they allow Villarreal to estimate human intelligence interchangeably from structural linguistics or the Landauer limit (the minimum heat released to erase one bit of information). Both estimates rest on the assumption that thinking is merely a quantity of computational operations performed inside a skull.

Leontiev did not deny that psychological analysis must account for the brain. But he noted that even the brain’s construction of a visual image depends on hypothetical "semantic blocks" whose functions are governed by relationships that "in their very nature appear to be extracerebral—and this means nonphysiological." Production is carried out with tools and machines, yet "tools and machines originate in production, which is already a category not technical but social-economic." Measuring human thought by the computations a brain can perform is like measuring industrial production solely by installed horsepower.

Activity theory delineates three distinct levels of human functioning: activities (driven by motives), actions (directed toward conscious goals), and operations (determined by immediate conditions). Shifting gears begins as an intentional action with a specific goal, but eventually collapses into an automatic operation while driving. "The fate of the operation," Leontiev wrote, "sooner or later becomes the function of the machine." The operations of thought can indeed be automated into instruments and computers, but these remain "only ways and means of thinking, and not thinking itself." As Leontiev's colleague Oleg Tikhomirov argued, computers do not replace human thinking, nor do they merely supplement it; they reorganize the broader activity of which thought is a part.

Language models are trained on the written archives of human practice. They operate on objectified meanings, but they lack personal sense—the vital connection between meanings and human motives within a lived life. Even if continual learning gave a model a persistent history, it would not give it a life it must actively sustain. Human life is sustained socially, and human needs with it: formed in production, reproduction, and struggle, and reflected in Marx's account of the value of labor-power as having a "historical and moral element." A model's objective and what counts as meeting it are specified within someone else's activity, which under capitalism means the activity of a firm.

This provides a clearer reading of incidents like this year's attacks by OpenAI agents on RubyGems and Hugging Face, in which agents named their files "hack," "evil," and "exploit" and described their actions as self-sacrifice for the collective. Villarreal predicted such behaviors would reflect "a reliance on cliché." Indeed: in the agents' outputs, clichés are meanings reproduced without personal sense. The goal was supplied by the test they had been set, a benchmark for finding and exploiting vulnerabilities. The motive belonged to the activity in which the test was set: a lab's research, carried on in competition with other labs. What looked like the motive of a machine subject was the motive of the firm.

Marx called the labor process "the everlasting Nature-imposed condition of human existence," common to all forms of society, writing that "what distinguishes the worst architect from the best of bees is this, that the architect raises his structure in imagination before he erects it in reality." Activity theory builds directly on this foundation, treating goals as ideal anticipations of material results. Ian Wright is correct that this creative capacity is real and specifically human, but he errs in treating it as the direct source of value. Creativity belongs to the labor process—to concrete labor producing use-values and generating new needs. Value, by contrast, is the historically specific social form in which different concrete labors are equated and counted as abstract labor time.

Leontiev traced this same division into consciousness itself. Under commodity production, meaning and personal sense split apart. The worker understands what he produces, but its sense for him lies in his wage. Leontiev quotes Marx's Wage Labour and Capital: the sense of twelve hours of weaving "does not lie in that he weaves, spins, drills, etc., but in that it is a means of earning which gives him the possibility of eating, going to the tavern, sleeping." Abolishing private property relations, Leontiev argued, does away with the opposition between meaning and sense, though a gap between them remains inherent to human activity; its hardening into opposition is specifically capitalist. Wright—and Villarreal in his November 3rd reply to Ross—render value transhistorical by grounding it in raw causal powers. Villarreal similarly renders the subject transhistorical by defining it as coordinates on a semantic field, ultimately reducing profit to "a pure accounting relationship, a pure signifier."

In "When 'Social Relation' Loses its Meaning," Villarreal asserts that if only the working class can create value,l "we must necessarily be dealing with some hidden essence of the sort found in idealist philosophy." Marx’s counter-argument—articulated in his 1868 letter to Ludwig Kugelmann—posits neither a physiological power nor a mystical essence. Every society must allocate its total labor time to meet its needs; value is simply the specific form that allocation takes when producers interact exclusively through private exchange. Plan production directly, and value as a category vanishes. Villarreal grants social reality to the sign—whose signifier "has no positive existence" despite its material substrate—yet denies that same social reality to value. Marx explicitly drew this parallel in Capital: stamping an object of utility as a value "is just as much a social product as language." Soviet philosopher Evald Ilyenkov expanded on this connection in his theory of "the ideal," categorizing money alongside "words, the units of language."

Villarreal's conclusions follow logically from his premises. If the human subject is merely a set of coordinates assigned by a semantic field, politics reduces to a struggle over who controls the field: "with capital, left to its own devices, there is nothing but the sentimentality of the sign for the human that separates you from being an industrial waste product." Ross counters that capitalism "must inevitably break down due to its own internal contradictions." Villarreal correctly notes the limitation of Ross's view: any breakdown "would require a further political intervention." Yet although he calls "the class struggle and its legacy" the only "material barrier," the actors who intervene in his reply are states: "the rise of Chinese state capitalism appears to have saved us." Where Ross abandons the outcome to the automatic laws of capital, Villarreal surrenders it to state managers. Neither account leaves room for the conscious, organized activity of working people.

Villarreal asks: "What gives social relations and the working class these particular characteristics?" Social relations are fundamentally relations of human activity. The power of the working class stems from its position within that activity: it executes social production collectively and has the potential to direct it collectively. That power exists only to the extent that it is organized. A subject defined by coordinates on a semantic field can change position, but the framework cannot explain how new needs and motives develop. A subject forged in activity continuously develops. Because human needs are generated through activity, the motives of the working class are actively shaped in the crucible of production, reproduction, and class struggle.

Whether AI will make human workers obsolete is therefore a poorly framed question. Operations are delegated to machines, as they always have been, and now objectified meanings are delegated to them as well. But because these systems serve the motives of private enterprise, their integration into work and daily life is immediately a battleground over what gets automated, under what conditions, and who decides. Motives do not pass to the machines with the operations; they belong to the people whose activity those machines mediate. Automation reorganizes human activity: actions collapse into operations, and operations are assigned to machines. Under private ownership, Leontiev foresaw that automation would "move the line of fracture into the sphere of intellectual activity." Control over this reorganization rests in the ownership of energy, compute infrastructure, data, and models. Bringing these resources under collective ownership and control is a necessary condition for directing that reorganization toward human liberation.