By Lázara Ávila Fernández

This inflation of the sign “language” is the inflation of the sign itself, absolute inflation, inflation itself. Yet, by one of its aspects or shadows, it is itself still a sign. This crisis is also a symptom. It indicates, as if in spite of itself, that a historico-metaphysical epoch must finally determine as language the totality of its problematic horizon. —Jacques Derrida, Of Grammatology.

Derrida speaks here of the semantic overextension of the term “language” in the human sciences of his time. He observes that the concept of “language” has been inflated to the point where it no longer designates a faculty or a system among others, but the total horizon of knowledge. Everything—the unconscious, society, literature, art, philosophy itself—is declared to be “language.” At the same time, he underscores that this fact is a symptom of an epoch that must ultimately determine the totality of its problematic horizon as language. Reading this sentence might serve as a reminder of what is happening right now. Today, an even more excessive variant is gaining ground, and its nature is being overvalued in every field of human knowledge and action. We are referring to the output.

But what makes us see the quasi-sign as a symptom of a larger crisis? Think for a second of something as simple as needing to find a recipe. Now, to meet that need, you will most likely turn to a language model before an Internet search engine, before a friend, before your grandmother.

And if, for example, you searched on Google, you would notice that the first response generally comes from Google AI Overview. Today, language models are asked about almost everything, and the output is accepted with little or no resistance. Think, likewise, that you need a marketing plan for your company. You will provide a little context, assign a role to the language model, and within seconds the result, better or worse, will be circulating on your social media. And what of more complex fields such as education, health, defense, or philosophy? There is such an excess, such an overflow in the use of the quasi-sign that this society may already be unable to imagine itself without its support. One could say that the crisis goes much further because, in most cases, we accept the quasi-sign as a legitimate representative of the act of signifying when, as we have explained, that act has been expropriated.

But where is the fatal error in that perception, in seeing the machine as a generator of signs rather than quasi-signs? In us or in the algorithm?

The Inflationary Illusion

Let us think for a moment about how we reconstruct a face in the clouds when, in reality, there is only a visible mass formed by millions of tiny droplets of water, ice crystals, or a mixture of both, suspended in the atmosphere; yet the brain deceives us into seeing a face. Try reading a text on a digital screen that contains spelling errors, and you will see how the brain supplies what is missing.

We call the phenomenon underlying the human perception of a mind in the machine—the attribution of the act of signifying to the language model—the inflationary illusion.

The inflationary illusion is the cognitive mechanism that hides from the recipient the true nature of the output and, correspondingly, that of the language model. It is not an intellectual error or a lack of human capacity; it is a cognitive category error. And it occurs because the brain cannot avoid reading thought where there is structure—in this case, algorithmic syntax. It is the same cognitive faculty that allows us to read other humans or even the language of other species, but which betrays us when faced with the quasi-sign.

The human breathes their own thought into the quasi-sign, anchors it to their Dynamic Object and to their constitutive present, and in that dynamic, without intending it, the fact that a loan has taken place dissolves from view—and not just any loan: the human has lent their most precious asset, the act of signifying. The inversion in the process and in the perception of it is such that it intensifies the phenomenon.

The Possibility of Historical Forgetting in the Human Domain

Derrida, glossing Husserl in Edmund Husserl’s Origin of Geometry: An Introduction, writes:

We also see that, for the same reason, forgetfulness will never be radical, however profound it may be, and sense can always, in principle and de jure, be reactivated.

For Husserl, forgetting presupposes a prior intentional acquisition. In other words, something was present to a consciousness, became sedimented, and passed into a state of latency from which it can be reactivated.

But what happens in the Algorithmic Domain? Were the sedimentations with which it operates ever present to an algorithmic consciousness, one that does not exist?

In algorithmic systems there may be information loss, parametric interference, erasure, inaccessibility, or even what is technically termed catastrophic forgetting. None of these phenomena is equivalent to the intentional forgetting described by Husserl. They are modifications of an operation, not of an ego. The model does not forget in the Husserlian sense.

Earlier on that same page, Derrida argues:

Forgetfulness of truth itself will thus be nothing but the failure of an act and the abdication of a responsibility, a lapse more than a defeat—and this forgetfulness can be made to appear in person only on the basis of an intentional history.

Language models lack not only an ego but also responsibility. The inflationary illusion not infrequently leads us to shift that responsibility onto the model. And the language model is not responsible.

Note: This text 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.