The Constitutive Present and the Operational Present

By Lázara Ávila Fernández

To delve into this topic, we need to journey back to the moment when the act of signifying takes place. This look back places us in the Human Domain. In this chapter, we seek to answer the following: What happens to that act when one of its materializations—in this case, writing—is transduced into data? And I know full well that this is one of those questions we ask ourselves every day when we wonder whether or not we can use AI in our creative processes; when we ask whether it is a tool, a medium, or an assistant; or when we question its limits. These are all genuine concerns that we carry with us.

We must clarify that this type of expropriation does not deprive the human of the act of signifying. It neither represents the physical destruction of the original document nor amounts to a literal copy that the language model extracts and stores in a repository. It is the relations of continuity that are expropriated. In other words, a fracture occurs in the relation between the act, its conditions of possibility, and the materialization resulting from the transduction. Expropriation leads to the production of the quasi-sign.

To understand the above, we must first grasp that the act of signifying is constitutive: it is what grounds understanding in our relation to the world, both inner and outer, including the clash with reality, even before the constitution of a particular language, of language as such, or of any external materialization. Even before an intention, before a thought.

Before writing a poem about pain, there is an act of signifying that passes through the experience of pain and the complexity of its relation with the subject; through the sensory, subjective, and cognitive capacities of the person who materializes that experience, whether in a gesture, in oral or written discourse, in a soliloquy that never comes to light, or in the inscription of the poem in a particular style and form of language, within a particular language.

In the realm of human practice and experience, for an explanation to take shape, for a concept or theory to be formulated, for a sense to be constituted, or even for an intention to arise, something must first have acquired relevance for a situated human. We are not speaking of a succession of events linked in chronological order, but of a constitutive event. That relevance occurs in the relation of a corporeal being marked by need, history, and sociality to its world.

The actualization of an Interpretant is part of that event, but it does not exhaust it. It enters the universe of semiotic relations and conditions of possibility from which a new act of signifying may take place.

This is the key to understanding that, before a concept could be formulated—for example, about the differences among the colors of the rainbow—certain chromatic variations had to acquire relevance for that human. Differences do not acquire semiotic value by their mere existence, although they may constitute possibilities of signification. If something has never entered into a relation of relevance with a human being, that something does not yet exist as realized signification.

Let us consider, then, the texts in repositories that are transduced into data to train the language model. Those texts are the materializations of human acts of signifying that had their own historicity and constitutive present, their own conditions of possibility that made them possible. There was a prior act. All of this was conditioned by a human worldview, by their subjectivity, and by the very objectivity—or lack thereof—of that which led to the act of signifying, within a multiple, synchronous or asynchronous yet continuous relation capable of enduring over time and allowing one, from the present, to return to the past and move toward the future.

And within that relation, the act of signifying, in order to exist, requires a constitutive present—that is, the relational event in which the conditions of possibility are actualized so that something may acquire intelligibility or relevance and may materialize. Any materialization, past, present, or future, has required or requires that constitutive present. Every reinscription, past, present, or future, has required, requires, or will require its own constitutive present. No materialization arises ex nihilo.

Now, where and at what moment does the discontinuity occur between the act of signifying, its conditions of possibility, and the materialization that, after being transduced into data and processed operationally, will return to the Human Domain as output? When does it shift into a different temporal regime, and what is that time?

In the transduction of human materializations, this fracture occurs when they cease to be the same in a semiotic sense, even though a transformational continuity of marks, relations, and proportions persists among them. When they begin to exist under new laws—algorithmic ones—the discontinuity we have named occurs. The iteration and transformation of the “materializations” within the model, as well as the statistical correlations established among them, do not take place in a constitutive present, but in an operational present. By operational present, we mean the computational actuality in which transformations are performed on data and internal states are modified, without anything acquiring relevance for the system in a relation of its own with the world.

What Is the Signification Resource?

It is the relational set of conditions of possibility and materiality that, in the Human Domain and within its constitutive present, allows the act of signifying to materialize and allows those materializations to be reinscribed again and again and even to be transduced into data.

This concept allows us to see that, if certain conditions of possibility are absent, if the constitutive present does not exist, if the language model does not maintain a relation of its own with Secondness—one in which something can acquire relevance or intelligibility—then there is no act of signifying.

Transduction into Data

In Jacques Derrida’s Edmund Husserl’s Origin of Geometry: An Introduction, translated by John P. Leavey, Jr., p. 71, we read:

The translatability of the word lion, then, will not in principle be absolute and universal. It will be empirically conditioned by the contingent encounter in a receptive intuition of something like the lion. The latter is not an “objectivity of the understanding,” but an “object of receptivity.” The ideality of its sense and of what it evokes irreducibly adheres to an empirical subjectivity.²

When we utter the word “lion,” when someone writes it by hand, or when, to give another example, it appears on a mural, although each materialization constitutes a different empirical event, we can recognize in all of them the ideal unity of its sense.

But does empirical subjectivity exist in the model? Does the ideality of its sense and of what it evokes remain attached to something that does not exist? What survives from the act of signifying? What becomes of the contingent encounter?

Once the word “lion” enters the language model, it is transduced into data through tokenization. The word is divided into one or more tokens, which are assigned numerical identifiers that the model processes operationally.

The resulting units do not necessarily correspond to the linguistic and semiotic unit we recognize as “lion.” There is no recognition by the model, no contingent encounter with the animal—alive, dead, depicted in a cave painting, or in a photograph—nor with the conditions under which that animal acquired relevance for a human and became inscribed in an act of signifying.

The unit of sense of “lion” does not survive the crossing as a unit sustained by an adherence to empirical subjectivity. What persists are transduced materializations and operational regularities that allow the model to discriminate among different contexts.

For example:

“The lion roars”

“We traveled to León”

“Kingdom of León”

“María León”

In conclusion, although the language model can discriminate operationally among the animal, the city, the kingdom, and the surname, it has no access to the unit of sense of each use, nor does it actualize it in a constitutive present.

“The ideality of its sense and of what it evokes irreducibly adheres to an empirical subjectivity.”³

Let us now examine what happens to that adherence to empirical subjectivity when the relation between “rain” and “tear” is transduced and treated operationally by the language model. In our example, “rain” requires the drops on the windowpane and the human who watches them strike the glass so that, in that constitutive present, it can be inscribed as “tear” and not only as “moisture” or “downpour.”

But let us be precise: neither the drops nor their mere contemplation are enough. That reinscription occurs because, in the encounter, worldview, memory, tradition, and the human’s situated relation with the world come into play. Only then does the rain acquire a relevance not found in water or moisture alone, and it can come to occupy the position of an entirely different sign. In a poem, the author may have written “raindrops”; however, this linguistic materialization may evoke or signify, for both the author and the reader, crying, sadness, or pain, rather than portions of liquid water falling from clouds.

Upon entering the language model, the written materialization “rain” will be transduced, as we already know, into data through tokenization. The tokenizer will divide it into one or more tokens or fragments to which numerical identifiers are assigned. Each identifier then makes it possible to retrieve an initial vector—an embedding—which is subsequently transformed according to context. In this way, the model can establish operational relations among “rain,” “downpour,” “drizzle,” “moisture,” and even “tear,” and can even produce a poem.

However, that (re)production (correlation) does not actualize the constitutive present in which someone contemplated the rain, remembered a loss, and recontextualized it as a tear, nor does it constitute within the model another present of that nature. And only in the collision of the output with a situated human does empirical subjectivity become present again.

The Expropriation of the Act of Signifying: The Four Tempus—Conceptual Regimes of Expropriation

First Tempus: The Transduced Heritage

The first tempus occurs at the moment when libraries, repositories of knowledge, the Internet, newspaper articles, books, and so on are “mined” to train the language model.

One of the central operations of training consists of predicting the next token. The model receives a sequence and calculates a probability distribution in order to predict its continuation. Let us consider the following example:

When the language model is presented with the incomplete sequence “The lion lives in the…,” it must complete it. That is its task.

The system then compares the calculated probability distribution with the next token observed in the sequence on which it was trained, which might well have been “The lion lives on the savanna.” During training, the question is not what a lion actually is. The model does not encounter the animal, nor does it need to be shown an image of it. The operational task is the following: Given this sequence, what token came next?

Without going into each of the operations that make up training, this procedure shows the semiotic discontinuity to which the materialization is subjected. The system does not receive the contingent encounter with the lion, the empirical subjectivity for which it acquired relevance, or the unit of sense constituted in that relation. It receives the transduced materialization and adjusts its parameters on the basis of the regularities it finds in it.

Second Tempus: The Activated Operation

The second tempus begins when the model is set in motion and encompasses its internal operation, activated by the user’s prompt. The model works with the data resulting from the transduction. Once activated—and by this we mean the critical point at which it is ready for interaction with the user—the regime that prevails when the human enters the prompt is one of efficient operation—matrix operations and pattern tracking in a geometric space—never the formation of meaning understood as an act of signifying.

There is an operational closure, that is, a chain of calculations and successive decoding operations that ultimately leads to the DFZ⁴. That closure is mathematical, a matter of technical probability, not of sense. The resulting probability distribution and the decoding regime applied to it condition the output. Even when a biased response appears, it is produced as an effect of numerical relations in which patterns inherited from the corpus, the architecture, the tuning, the system’s constraints, the prompt, and the context window—with its own constraints and restrictions—converge.

Nevertheless, we emphasize that the operation is neither autonomous nor neutral. The intervention of the designer—a figure present in each regime—remains inscribed in the architecture, the parameters, the optimization objectives, and the system’s constraints. In addition to the restrictions imposed by the policies of the company that owns the language model, there are the regulations that legislators may impose. In essence, every restriction and every design is the result of human acts of signifying placed at the service of the effectiveness, productivity, and safety of language models.

Third Tempus: The Double Expropriation

The third tempus occurs at the moment when, in its relation to a human, the DFZ completes its constitution as a sign through the actualization of an Interpretant and enters the human network of meanings. Each time a human, in the act of signifying, actualizes an Interpretant from the DFZ and externalizes the result of that interpretation in order to reintroduce it into the context window, a double expropriation of the act of signifying occurs.

The first is a semiotic expropriation: the human contributes their world, anchoring to the Dynamic Object, and contact with Secondness so that the quasi-sign acquires sense. The human reinscribes their world in the materialization and, from their constitutive present, opens a semiotic future for it. In this case, the expropriation does not yet consist in the transduction of the closure into data, but in the functional utilization of the human act of signifying so that the DFZ can operate as a sign.

The second is an operational expropriation, and it occurs when the semiotic closure materializes in a new inscription, in a new prompt, which, as part of the interaction, enters the context window to be captured by the algorithmic regime.

But here the reader may ask: What happens when the human does not externalize their closure and leaves the chat? Without externalization there is no new prompt; however, that does not mean that the conditions of possibility for the expropriation described here are absent. Without dwelling on a concept addressed in the first book, the third tempus is also the crucial moment when the semiotic closure inscribes the human and their act of signifying into the Semiotic Network of Necessity and when, depending on the Criticality Factor exercised by the user, their action may either exacerbate or contain their dependence on the Network.

The incorporation of human closure into the Network extends beyond the double expropriation. Its reiteration and accumulation actualize, on a systemic scale, the Capitalism of Exacerbated Necessity. The human need to close the DFZ becomes a resource for the reproduction and expansion of the Network itself.⁵

Fourth Tempus: Human Reinforcement

Think of the moment when you say a simple “hello”: behind those four letters lies a human history of education, family, culture, and society. Behind the model’s “hello,” by contrast, lies a matrix operation that prioritizes what is plausible and probabilistic, not meaning.

With that said, we can state the fourth tempus: it occurs when human evaluation and preferences are used to adjust or improve the language model. We are not examining here the working conditions of annotators, but how much signification resource the human contributes in order to reduce friction between the model and the end user.

Their choices, comparisons, corrections, and preferences arise from human acts of signifying, from their constitutive presents, but are transformed into operational signals of reward, classification, or adjustment. The system incorporates the formal effects of those acts without incorporating the lived and intentional conditions of those who produced them.

Notes

1. Tempus: In this essay, tempus is used as an invariant conceptual term to designate each of the regimes of expropriation; their order is expository, not necessarily chronological. ↩
2. Jacques Derrida, Edmund Husserl’s Origin of Geometry: An Introduction, translated, with a preface and afterword, by John P. Leavey, Jr. (Lincoln and London: University of Nebraska Press, 1989), 71. ↩
3. Derrida, Edmund Husserl’s Origin of Geometry: An Introduction, 71. ↩
4. DFZ: Demanding False Zero (Spanish: Cero Falso Demandante, CFD). An algorithmic sign of a dual nature: “Zero” because of its lack of intention and experience; “False” because it contains patterns, design, and statistical efficiency; “Demanding” because it requires a human Interpretant to complete its semiotic cycle. See the glossary and chapter VII, “The Nature of C-S (DFZ) and the Empirical Evidence,” in Lázara Ávila Fernández, Inteligencia algorítmica abductiva: emulación semiótica. Un viaje a la «verdadera conciencia» de la IA (S.E. Pinar Publisher, 2025). ↩
5. Three concepts developed in the first book intervene here: Semiotic Network of Necessity (SNN), the sociotechnical network that sustains the relation of dependence between the human and algorithmic intelligence; Criticality Factor (CF), the degree of critical awareness the Interpretant applies when reading the DFZ, from maximum critical engagement to total passivity; and Capitalism of Exacerbated Necessity (CEN), the regime that systematically exploits that dependence, transforming it into a continuous source of capital. The three concepts are developed in chapters X, XI, and XII and in the glossary of Ávila Fernández, Inteligencia algorítmica abductiva: emulación semiótica. Un viaje a la «verdadera conciencia» de la IA. ↩

Note: This chapter is an excerpt from Augere: The Death of the Author. The English version was translated from the original Spanish with the assistance of AI-based translation tools and subsequently revised to preserve the book’s conceptual terminology, argumentative structure, and authorial intent.