Is a Derridean Critique of Generative AI Possible? Writing Machines and Logocentrism without Subject

by Víctor Betriu Yáñez

The arrival of generative AI seems to have renewed interest in the work of Jacques Derrida. Today, his views on language and exteriorization are deployed to refuse the naivety of the classic anthropocentric position on language, authorship and creativity. Unsurprisingly, such a position was quickly adopted by early critics of ChatGPT for fear that delegating distinctly human activities of signification to automated machines would unleash ethical-political issues such as massive spread of misinformation, the inability to distinguish sense from nonsense, or the reinforcement of the status quo through discriminatory biases. For example, Emily Bender, with the well-known image of the stochastic parrot,^[1] defended a view of language according to which writing can only be considered potentially truthful if it corresponds to the communication of a subject’s intention.^[2] If there is no subject behind words, then there is no meaning but merely signs, which, despite being empty, appear to us as carrying meaning. This leads to the problematic claim that AI-generated text is an illusion which cannot possibly attain the status of writing, insofar as it corresponds to the mere stitching together of linguistic forms devoid of content.^[3] These linguistic forms cannot attain the status of a sign because they do not represent anything, given that there is no consciousness behind that could embed thoughts in them. The strategy of this argument consists in redrawing the opposition between the human and the machine when new technologies happen to blur it. It is then to be expected that anyone with a deconstructive sensitivity and a concern with the possibility of hospitable, non-binary thinking would be on the alert for any opportunity to blur existing clean-cut oppositions and to attack conservative strategies of metaphysical reconstitution.

David J. Gunkel may have been the first with such a sensitivity to bristle in the face of this argument.^[4] He not only noticed with sagacity that large language models (LLMs) are illustrative of the fundamental Derridean insight that neither speech nor thought precedes writing. He also noticed that critics like Bender were taking the side of what Derrida called logocentrism: any reaffirmation of an oppositional way of thinking that privileges and justifies an understanding of reality through the imposition of an ultimate foundation that bestows upon the meaning of being a fixed and self-present identity. Such oppositional thinking results in forms of violence because it is instituted through a conceptual operation which subjects and represses one side of the opposition in order to impose the purity and fullness that any identity needs (yet lacks) to keep afloat. Yet, although logocentrism refers to the ethnocentric way of thinking that constitutes our world of sense and our unconscious (‘all the more when one does not suspect it’^[5]) and is thus related to the systematic organization of our conceptuality by Western metaphysics in hierarchized oppositions, such as masculine/feminine, human/animal, organism/machine, life/death, soul/body or being/becoming, logocentrism is particularly bound to the discussion about language, writing and meaning that today surrounds generative AI. This is because the identification of this Western rationality emerges out of the deconstruction of the opposition between speech and writing.

In his early works, Derrida carefully shows how writing is systematically repressed and subjected to speech in the history of Western metaphysics through a common-sensical, self-evident and instrumentalist conception of writing that constitutes speech as a purer form of signification due to its immediate relation to truth and the logos as the rational foundation of what is. A logocentric view of writing then refers to the reduction of writing to a means of communication, as a vehicle that carries a preexisting discrete unit which has been embedded in written marks. In short, logocentrism treats writing as the representation of a representation, as the last link of the series according to which one first thinks, then speaks and finally writes what one had first thought. Considering this, one might better understand why Gunkel would consider Bender’s position logocentric: once writing is reduced as the representation of speech or as the means carrying a thought that can allegedly be attained after decoding the intent or wanting-to-say embedded in a given linguistic expression, writing becomes subjected to speech and the transparent medium of the voice (consciousness). With this operation, only a text created by a human can really signify and thus have the status of authentic text.

Like Bender, Gunkel is aware of the several societal challenges that large language models pose. But, unlike Bender, instead of preparing a critical analysis of this technology, Gunkel foregrounds that, despite such challenges, large language models bring the opportunity to break with the logocentric understanding of writing. He writes:

LLMs do not threaten writing, the figure of the author, or the concept of truth. They only threaten a particular and limited understanding of what these ideas represent—one that is itself not some naturally occurring phenomenon but the product of a particular culture and philosophical tradition. Instead of being (mis)understood as signs of the apocalypse or the end of writing, LLMs reveal the terminal limits of the author function, participate in a deconstruction of its organizing principles, and open the opportunity to think and write differently.^[6]

It would then seem that Gunkel puts forward a desirable stance insofar as he makes use of the language of deconstruction to reveal an alleged bright side of LLMs. In contrast, Bender’s position appears undesirable for the naivety of her philosophical presuppositions. However, I will try to show in what follows that Gunkel’s position is equally naïve or insufficient for not being able to unfold the critical potential of Derridean insights against a way of thinking that risks supporting the technological oligopoly orchestrating popular LLMs such as ChatGPT. Convinced that a different use of deconstruction could make Derrida’s philosophy an element of conceptual resistance to the economic goals of Big Tech and their anthropological, existential, political and environmental consequences, my aim is to lay out the possibilities for a third stance: one that is both deconstructive and critical, at the same time and inseparably; one that is critical for being deconstructive. To do so, I will highlight central Derridean insights that appear to be in contradiction with Gunkel’s view. But first let us outline the latter.

For Gunkel, if one looks at ChatGPT through Derridean glasses, one can but witness the heralding of the end of logocentrism. Insofar as large language models evince that writing is possible without a consciousness or mind behind it, insofar as large language models ‘write without speaking’,^[7] these technologies materialize the core insight of deconstruction, announcing the chance to abandon the anthropocentric paradigm of creation which takes the human mind and intention as the center and only form of truthful linguistic production. As Gunkel writes: ‘the decentering of intention, the primacy of the signifier, the endless deferral of meaning—now finds material expression in the operations of LLMs’.^[8] Or, as he puts it elsewhere: ‘the LLM … materializes the arche-writing that Derrida described — not writing as mere “words on the page” but writing as the fundamental structure of signification itself’.^[9] From this perspective, generative AI systems not only signal and confirm the death of the author but also function according to or put into practice the dynamic movement of differing and deferring that makes meaning possible: ‘LLMs operationalize and perform this différance’.^[10] This suggestion that generative AI deconstructs the alleged stability of our philosophemes through ‘the unraveling of Western metaphysical assumptions that no longer hold and perhaps were already losing their grip on our thinking about writing and writing about thinking from the very beginning’^[11] shows, according to Gunkel, the opportunity of a new beginning: the inauguration of a new paradigm of creation and signification based on a collaborationist model of human-machine interaction for enhanced production.

So, if the strategy behind Bender’s argument is to reenforce an opposition that new technologies seem to call into question, the strategy behind Gunkel’s argument is to defend and endorse the technology as the opportunity for a blurring that would inaugurate a more emancipatory form of expression, interpretation and creation, free from the authority of the self-present subject. In his words: ‘LLMs may well signal the end of the author, but this isn’t a loss to be lamented. In fact, these machines can be liberating: They free both writers and readers from the authoritarian control and influence of this thing we call the “author”’.^[12] If the former strategy is about reclaiming that buoy which gives us (a sense of) security and certainty in a sea of simulations to keep the sovereignty of the I in light of deceptive experiences, the latter is about promoting the idea that not only a shift in conceptuality is to be attained technologically but that it is technologies serving the economic interests of Big Tech which can bid farewell to metaphysics. To put it differently, while the former fights the fear of the end of human writing with logocentrism, the latter celebrates the end of human writing with the language of deconstruction.

Gunkel is not alone in this Derridean endorsement of generative AI. Alžbeta Kuchtová holds a similar position by foregrounding the importance of ‘the erasure of the I in the text’^[13] in the Derridean perspective. Like Gunkel, Kuchtová thinks that ‘Derrida’s theoretical framework of exteriorization is applicable to recent developments in machine learning and is useful for explaining it’.^[14] They both see in Derrida an explanatory potential to understand the age of AI and to rethink creation in the absence of a conceptual opposition between the human and the machine. In addition, they both share a critical stance against AI critics who reproduce a conservative idea of creation based on anthropocentric authenticity. In fact, it is interesting to see how little they use the term logocentrism or its cognates, given that they merely use it to disqualify the perspective of the critics and never to critique the technology itself.^[15] Their questioning attitude regarding technology is never suspicious but confiding. Gunkel and Kuchtová do not consider the possibility that generative AI illustrates a form of logocentrism as well; they do not consider the possibility that both AI critics and generative AI operate on logocentric grounds.

In light of these interpretations, it would seem as though the possibility of a Derridean critique of generative AI was an oxymoron. Derrida’s philosophy appears to be the legitimizing discourse of AI without any possibility of critique. One then might be ready to claim that ChatGPT is a deconstructive technology and any resistance to it conservative logocentrism. But why should we accept this? Instead of witnessing, as Gunkel diagnoses, a technological deconstruction of our own metaphysical assumptions concerning writing, could we not be witnessing the neoliberal appropriation of poststructuralist principles that neutralize their oppositional force or salience towards any given order of sense? Does the marketability of generative AI services not rely strongly on this blurring between the human and the machine (as a strategy of identification) so that the user can think the machine can come to substitute their capacity for self-expression? What if, contrary to appearances, generative AI reinforces logocentrism by promoting an idea of writing that, despite having no transcendental subject, is still instrumental and expressivist of a divine totality that is always mathematizable? What if we have substituted the transcendental subject by a finite network of weighted relations among data? What if this non-logocentric, collaborationist panacea is complicit with the obliteration of the weak force or residual intentionality that remains after the subject has relinquished the sovereignty of its ‘I’?

Even if we are willing to welcome a way of thinking that acknowledges the impossibility of full intention, of the absolute authority of the author, of the sovereignty of the centered or self-present subject—even if we need to think of new forms of writing from this groundless ground that Derrida’s philosophy helped us articulate, we should not be afraid nor forget to defend the singularity of every affected body that gives itself in writing to affect and transform oneself and others. It is neither metaphysical nor conservative to critique the obliteration of all intentionality in light of the technological homogenization of our relationship to the world and ourselves through the loss of writing skills, self-expression and linguistic inventiveness. The problem this renewed interest in Derrida’s works makes us face is how to inherit deconstruction in a stage of our digital age, where emerging technologies appear to be synonymous with neoliberal logics of profit making at the cost of not only our inventive, expressive, communicative and intellectual capacities, but also of the exploitation of the Global South and the environment. For we risk presenting Derrida’s philosophy as a justification to AI and, subsequently, to its arguably indissociable business enterprise led by the tech giants and to the transhumanist ideology hiding behind collaborationist accounts of the human-AI relationship which only serves to increase the marketability of these technologies.^[16] And this translates more simply to the question of whether we can maintain the critical attitude of Bender without adopting her logocentrism while we adopt the deconstructive sensitivity of Gunkel and Kuchtová without giving up on critique. The question is then whether it is possible to stand by a deconstructive and critical stance (in case one would even think they can be uncoupled) concerning generative AI.

Thus, alternatively to the idea that Derrida is alluring today for the potential of his works to explain what large language models do, I venture here the thesis that Derrida is alluring to critique generative AI through the notion of logocentrism and the unconditional care for the act of writing (even after generalizing it). In order to respond in the affirmative to the possibility of a Derridean critique of generative AI despite extant interpretations, in what follows I briefly bring to our attention a couple of insights from Derrida’s philosophy that could serve as starting points for such a critique. The thesis which binds these insights together is the idea that, even if generative AI seemingly interrupts logocentrism, it actually reinforces it.

I. Writing is essential to thinking

We have seen that, from the perspective of logocentric metaphysics, writing is understood instrumentally, as a means through which preexisting thoughts can be made to travel in space and time. This implies that writing is inessential or secondary to thinking. Consequently, speech is the primordial representation thought requires to be exteriorized in the presumed self-conscious presence of the speaker. Derrida notes that this view attributes to speech and not to writing a close relationship to truth: ‘Plato said of writing that it was an orphan or a bastard, as opposed to speech, the legitimate and high-born son of the “father of logos”’.^[17] In his early works, Derrida takes on the challenge of showing that this alleged purity of logos and speech, as well as their intimate solidarity, stem from an unfounded metaphysical belief whose function is to repress the actual contamination of logos and speech by writing. The characteristics that this tradition tried to circumscribe or quarantine within writing, such as being the signifier of another signifier or exceeding and severing from the animating intention supposedly instituting it, are constitutive of anything that functions as a mark, whether it appears as thought, speech, or text in the traditional sense. These traits associated with writing are tied to iterability (i.e., the possibility of being repeated); Derrida reveals that thought and speech are no less dependent on iterability. Not only is there no thinking outside of language but thinking and speech always already contain the possibility of being written down insofar as they can be re-thought and re-spoken. A non-logocentric understanding of writing then not only breaks with the idea that thought precedes speech and speech precedes writing, so that the idea of writing without speaking becomes perfectly intuitive. A non-logocentric understanding of writing foregrounds the fundamental insight that writing is the condition of possibility of both thinking and speech. While for logocentric metaphysics writing is inessential to thinking, after Derrida’s deconstruction, writing can only appear to us as ‘essential’ to thinking. Thought is not communicated through writing but rather takes place in writing.

Of course, in extracting from the act of writing the structure of iterability, Derrida undergoes a conceptual operation of ontological generalization which elevates writing from an ontic phenomenon as mere marks on a piece of paper to a quasi-transcendental structure of signification. This is the reason why Rodolphe Gasché can write that ‘writing has little or nothing to do with the act of writing’^[18] and Gunkel and Kuchtová can be so unconcerned about the end of human writing. Whereas these views are correct, I wonder whether we have ended up privileging the ontological structure of writing over the ontic act of writing. Should we not aim at a re-evaluation of the latter in light of the consequences of the former? Once writing is not understood instrumentally, once we accept, in Derrida’s words, that ‘[w]riting is not simply a “composition” … [that] writing is not simply a way of positing or posing things together’,^[19] writing almost fuses with existence itself, with the unavoidable act of leaving traces behind in the passing of time. From this generalization of writing, the world appears to us as a text to be read, as an interwoven fabric of past traces which were never present, to the extent that seeing is reading and touching writing. This perspective reveals an inaugural and transformative character of writing that must be defended. This is not to say that we must rescue writing from its historical subordination to speech because writing actually has a better relationship to truth that we would like to restitute. About this, Derrida is very clear:

it has never been a question of opposing a graphocentrism to a logocentrism, nor, in general, any center to any other center. Of Grammatology is not a defense and illustration of grammatology. And even less a rehabilitation of what has always been called writing. It is not a question of returning to writing its rights, its superiority or its dignity. … [N]othing would be more ridiculously mystifying than such an ethical or axiological reversal, returning a prerogative or some elder’s right to writing.^[20]

And yet, Derrida plainly acknowledges:

writing to me is the essential performance or act. I am unable to dissociate thinking, teaching, and writing. That’s why I had to try to transform and to extend the concept of writing, which is not simply ‘writing down’ something.^[21]

In the age of generative AI, the act of writing is made equivalent to the act of prompting and to the act of ‘posing things together’ probabilistically. The user makes a request in the form of a command and the AI model spits back a statistical recombination of existing elements that produce an unseen unity which is perceived as fitting with what the user would have liked to write. This perception is a self-deception that precisely stems from our logocentric way of thinking. We as users believe that our thoughts exist in our mind and are retrieved upon demand. AI models then come to materialize this presupposition and make such retrieval effortless and immediate. Writing thus becomes a matter of assessing whether the output aligns with one’s ideas. However, as Anne Alombert puts it, ‘it is impossible to know what one would have wanted to write before having written it. If it is necessary to write, it is precisely to discover what one has to say. On the contrary, through the given command, it is the exercise of writing and, with it, the time of thinking, which are short-circuited’.^[22] And this logocentric discourse is not only implicitly held by users. Alombert has perspicaciously noted that CEOs propelling these generative AI models rely on the same discourse: ‘Sam Altman [CEO of OpenAI] affirms a metaphysical conception of creation, according to which the artwork preexists “in the mind” or “in the spirit” of the creator before engaging in its realization’.^[23] Hence, LLMs need not necessarily be interpreted as illustrating deconstruction insofar as they write without speaking. LLMs can be said to rely on the reinforcement of logocentrism to ensure the massive adoption of these technologies for users to entrust most of their signifying activities without resistance.

II. Logocentrism can live on without a subject

We have seen that, for Gunkel and Kuchtová, generative AI seems to announce the end of logocentrism and the possibility of unauthorized writing, of a writing that is not subjected to the intentions of a subject. This interruption of logocentrism is indicated by the possibility of staging meaningful significations without the authority of a transcendental subject that has access to truth. However, this presupposes the conviction that logocentrism cannot survive without the figure of the transcendental subject. And yet, Derrida sees in Leibniz’s characteristica universalis (the project for a universal language that could enhance philosophical thinking in expressing the totality of human thought through combinatorial operations) the confirmation of a logocentrism of another type. Giovanni Menegalle, studying this intriguing double regime of logocentrism, has called this overlooked Derridean model of logocentrism a ‘logocentrism without a subject’.^[24] Derrida writes in Of Grammatology:

In spite of all the differences that separate the projects of universal language [langue] or writing at this time (notably with respect to history and language [langage]), the concept of the simple absolute is always necessarily and indispensably involved. It would be easy to show that it always leads back to an infinitist theology and to the logos or the infinite understanding of God. That is why appearances to the contrary, and in spite of all the seduction that it can legitimately exercise on our epoch, the Leibnizian project of a universal characteristic that is not essentially phonetic does not interrupt logocentrism in any way. On the contrary, universal logic confirms logocentrism, is produced within it and with its help, exactly like the Hegelian critique to which it will be subjected. It is the complicity of these two contradictory movements that we target here. There is a profound unity, within a certain historical epoch, among infinitist theology, logocentrism, and a certain technicism. The originary and pre- or meta-phonetic writing that we are attempting to think here leads nothing less than an ‘overtaking’ of speech by the machine.^[25]

The essential feature of this logocentrism without subject is then the intimate solidarity between the existence of a totalizing God which can be rationally understood through mechanical calculation and formal language made of non-phonetic signs. In this form of logocentrism, the figure of the transcendental subject is substituted with the figure of a transcendent God. As Menegalle puts it: ‘At stake here is a version of logocentrism that short-circuits subjective presence via an alliance between onto-theology and universal technology—a parallel regime of logocentrism founded not in a transcendental subject but a transcendent God’.^[26] This ‘“overtaking” of speech by the machine’ suggests that breaking the representational link between writing and speech is insufficient to herald the end of logocentrism. Menegalle writes: ‘Derrida’s aim in Of Grammatology is to show that this decoupling of writing from the voice is insufficient to deliver it from logocentrism due to its anchoring in an “infinitist theology” through which any finite thing, sign or phenomenon, is taken to express God’s infinite intellect or logos’.^[27] From this we can draw a parallel with Gunkel’s claim that, insofar as LLMs write without speaking, ‘LLMs are structuralist machines — they are practical actualizations of structural linguistic theory, where words have meaning not by reference to things but by referring and deferring to other words — and they thereby disrupt the standard operating presumptions of classical (Aristotelian) semiotics’.^[28] One could then say that LLMs can be considered to be structuralist insofar as they put in the place of the subject a finite network of mathematically weighted relations among data. From this perspective, the outputs of LLMs may be seen as expressions of the underlying formal constants that would account for the set of relations constituting the network of data.

However, Derrida sees in structuralism the type of logocentrism that is inaugurated with Leibniz. In ‘Force and Signification’, Derrida accuses the structuralist Jean Rousset of being too Leibnizian in thinking that, ‘confronted with a literary work, one should always be able to find a line, no matter how complex, that accounts for the unity, the totality of its movement, and all the points it must traverse’.^[29] For Derrida, as Menegalle shows, ‘structuralism … is haunted by the ghost of a subject it refuses to acknowledge’.^[30] Structuralism, while operating under the idea of the death of the author, bestows upon a network of relations among formal linguistic elements the status of the subject which would ensure the possibility of a total reading. The philosophical presupposition here is that there is a transcendental place outside of writing in which meaning waits to be found or retrieved. In the same way one would expect to find a structure explaining the meaning of a text regardless of the author’s will, one would expect to be able to prompt the adequate form to an already existing meaning. Derrida writes, in line with the previous insight about the essential role of writing for thinking and against the Leibnizian and structuralist ideas of writing as the expression of a divine intellect or underlying formal constants:

To write is not only to know that through writing, through the extremities of style, the best will not necessarily transpire … It is also to be incapable of making meaning absolutely precede writing: it is thus to lower meaning while simultaneously elevating inscription. … To write is to know that what has not yet been produced within literality has no other dwelling place, does not await us as prescription in some topos ouranios, or some divine understanding. Meaning must await being said or written in order to inhabit itself, and in order to become, by differing from itself, what it is: meaning.^[31]

Hence, Derrida’s critique of structuralism shows that Gunkel’s idea that LLMs as structuralist machines announce the interruption of logocentrism is self-contradictory given that logocentrism is constitutive of structuralism. If LLMs can indeed be considered to be structuralist machines, or Leibnizian for that matter, then they are necessarily reliant on a logocentrism which can very well live on without a subject. What do we celebrate when we celebrate the death of the author? Death, or the phantasmatic return of the author?

* * *

Distilling these insights from Derrida’s philosophy helps us see that logocentrism is not solely on the side of the strategy of re-drawing the boundary between the human and the machine through a metaphysical, instrumentalist and anthropocentric view of language and creativity. Logocentrism also dwells in writing machines; it is perhaps constitutive of them. The possibility of affirming with Derrida that, contrary to appearances, generative AI does not interrupt logocentrism but rather confirms and reinforces it, shows that his philosophy is alluring today for its critical potential and not so much a source of legitimation for AI innovations. Promoting logocentrism, generative AI promotes an instrumentalist understanding of writing as well as the tacit belief in the transcendental status of probabilistic calculation of formal constants under existing data as an expression of truth.

If this matters, it is not only because the oligopoly of truth and of the sensible by the process of platformization linked to commercial AI chatbots risks fortifying the given order of sense, which, despite teeming with newness, remains inhospitable as a realm of the same. It is also because the loss of writing skills has detrimental effects on one’s capacity for (critical) thinking about oneself and the world, and thus, for transforming and emancipating oneself from the logics which constitute us through language even when we do not suspect it. Writing is both a form of self-making and self-destruction but never a vehicle to communicate a self which would remain immutable. To acknowledge that in writing one discovers one’s ideas, that one discovers oneself, also requires us to acknowledge that in writing one destroys oneself, killing the subject one is in the moment of writing to then metamorphose into something else. We must acknowledge that generative AI cannot become the instrument for our self-expression, because it substitutes the chance of a self-relation possibly leading to invention, self-discovery and self-destruction, for the pragmatic and efficient production of the predictable. This is what the two views scoped above (Bender’s critique without deconstruction and Gunkel’s use of deconstruction without critique) cannot account for.

Besides, we must recall that logocentrism is indissociable from the privilege of the voice or consciousness as a sign of the potential to access truth (logophonocentrism), from the patriarchal and androcentric logic in the quest for truth which privileges the masculine in the constitution of meaning (phallogocentrism), or from the (symbolic) carnivorism that undervalues the life of the other (animal) to quench oneself. In light of this, one might say that logophonocentrism is reproduced in the interface design of these AI chatbots which generally stage textual outputs in a temporalized way to evoke the ‘overtaking of speech by the machine’; that phallogocentrism is seen in the reinforcement of algorithmic biases and the tendency of homogenization which privileges the normalized and standardized significations; and that carnophallogocentrism corresponds to the logic behind the exploitation of workers in the Global South to train the models, the huge environmental impact these technologies have been shown to have or the intellectual theft which is inseparable from the required size of models to attain linguistic accuracy.

Finally, acknowledging that LLMs like ChatGPT reinforce logocentrism instead of halting it provides us with an explanation for the swift and massive acceptance of this technology upon days of being released, its continued growth in use as well as the popularized dissemination of its application to different services and institutions. In a sense, our texts were ChatGPT-able before ChatGPT arrived. This is why we received this technology with open arms. A historical process of linguistic and stylistic homogenization that necessarily results in a degree of sterility for thinking has taken place along with the neoliberalization of universities and other educational institutions, where texts have been reified and commodified as papers or where school writing exercises, presupposing normalcy, objectivity and lack of singularity, have become tedious tasks. When the average user of a technology like ChatGPT ranges from students to teachers, from full professors to PhDs, this should simply make us reflect about the type of writing we are conforming ourselves to. Today, the imperative is that to know how to write is to know how to prompt. In this sense, no resistance to AI or defense of human writing will be substantial unless we are able to rethink the structural significance of writing for thinking and world-making.

The challenge left for us is then how to defend that radical otherness anyone bears within, which appears necessary to contend against any closing off a given order of sense. How could Derrida help us defend one’s singularity without relapsing in a defense of the self-present subject? A starting point would be to recognize that, even if Derrida deconstructs the concept of intentionality, he never gets rid of it entirely. He writes: ‘the category of intention will not disappear; it will have its place, but from this place it will no longer be able to govern the entire scene and the entire system of utterances.’^[32] And elsewhere: ‘I must first recall that at no time does Sec [‘Signature Event Context’] invoke the absence, pure and simple, of intentionality. Nor is there any break, simple or radical, with intentionality. What the text questions is not intention or intentionality but their telos, which orients and organizes the movement and the possibility of a fulfillment, realization, and actualization in a plenitude that would be present to and identical with itself’.^[33] The challenge would then be to conceptualize that intentionality which rests after deconstructing it, in order to understand its role in the act of writing despite its inherent destinerrance. Throughout his work, from ‘Force and Signification’ to Rogues: Two Essays on Reason, Derrida uses in a more or less enigmatic way the notion of force to name this residual intentionality. He writes: ‘one would seek in vain a concept in phenomenology which would permit the conceptualization of intensity or force. The conceptualization not only of direction but of power (sic) [puissance: potency], not only the in but the tension of intentionality’.^[34] Furthermore, this force is a ‘force of weakness’^[35] or a ‘weak force’^[36]: ‘This vulnerable force, this force without power, opens up unconditionally to what or who comes and comes to affect it. The coming of this event exceeds the condition of mastery and the conventionally accepted authority of what is called the “performative.”’^[37] This residual intentionality is the force of an I who has divested itself from its ipseity, the force of a subject who has relinquished its identitarian sovereignty, the force of an irreducible singularity with the potent weakness that is able to resist the given order of sense without promoting metaphysics, as impossible as this may be. It is the force of a blind subject, a subject without I, who, unable to see, unable to be dazzled by the light of the logos, now touches the world with violent tenderness.^[38] To affirm life through endless critique and openness towards radical alterity, we might have to understand and defend this weak force that remains.

1. Emily M. Bender, Timnit Gebru, Angelina McMillan-Major and Shmargaret Shmitchell, ‘On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?’, FAccT ’21: Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, March 3–10, 2021, 610-623. They define the stochastic parrot as follows: ‘Contrary to how it may seem when we observe its output, an LM is a system for haphazardly stitching together sequences of linguistic forms it has observed in its vast training data, according to probabilistic information about how they combine, but without any reference to meaning: a stochastic parrot’ (616–17). ↑
2. Emily M. Bender and Alexander Koller, ‘Climbing towards NLU: On Meaning, Form, and Understanding in the Age of Data’, Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 2020, 5185–5198. They define meaning and understanding as follows: ‘The speaker has a certain communicative intent i, and chooses an expression e with a standing meaning s which is fit to express i in the current communicative situation. Upon hearing e, the listener then reconstructs s and uses their own knowledge of the communicative situation and their hypotheses about the speaker’s state of mind and intention in an attempt to deduce i’. (5187). We see here how they treat language as a codified convention of symbols that one chooses as a vehicle to represent one’s thoughts taking place outside of language. ↑
3. ‘The problem is, if one side of the communication does not have meaning, then the comprehension of the implicit meaning is an illusion arising from our singular human understanding of language (independent of the model)’ (Bender et al., ‘On the Dangers of Stochastic Parrots’, 616). ↑
4. David J. Gunkel, ‘AI Signals the Death of the Author’, Noema, June 4, 2025, https://www.noemamag.com/ai-signals-the-death-of-the-author/. Gunkel writes: ‘something … has been missed in the frenzy over the technological significance of LLMs: They are philosophically significant. What we now have are things that write without speaking, a proliferation of texts that do not have, nor are beholden to, the authoritative voice of an author, and statements whose truth cannot be anchored in and assured by a prior intention to say something.‘From one perspective—a perspective that remains bound to the usual ways of thinking— this can only be seen as a threat and crisis, for it challenges our very understanding of what writing is, the state of literature and the meaning of truth or the means of speaking the truth. But from another, it is an opportunity to think beyond the limitations of Western metaphysics and its hegemony’. ↑
5. Jacques Derrida, Of Grammatology, 40^th anniversary edition translated by Gayatri Chakravorti Spivak (Baltimore, Johns Hopkins University Press, 2016), 25. ↑
6. Gunkel, ‘AI Signals the Death of the Author’. ↑
7. Mark Coeckelbergh and David J. Gunkel, ‘ChatGPT: deconstructing the debate and moving it forward’, AI & Society, 39 (2023), 2221-2231. They write: ‘The fundamental challenge (or the opportunity) with LLMs, like ChatGPT or Google’s Bard, is that these algorithms write without speaking, i.e., without having access to (the) logos and without a living voice’ (2226). ↑
8. David J. Gunkel, ‘Prompted by Me. Generated by ChatGPT’, Human-Machine Communication, 10 (2025), 33. ↑
9. Gunkel, ‘Prompted by Me. Generated by ChatGPT’, 34. ↑
10. Gunkel, ‘Prompted by Me. Generated by ChatGPT’, 37. ↑
11. Gunkel, ‘Prompted by Me. Generated by ChatGPT’, 37-38. ↑
12. Gunkel, ‘AI Signals the Death of the Author’. ↑
13. Alžbeta Kuchtová, ‘The Incalculability of the Generated Text’, Philosophy & Technology, 37 (2024), article number 25, 10. ↑
14. Kuchtová, ‘The Incalculability of the Generated Text’, 4. ↑
15. Cf. Kuchtová, ‘The Incalculability of the Generated Text’, 6-7. Cf. Coeckelbergh and Gunkel, 2023, p. 2225. ↑
16. Cf. Frederic Jameson, Archaeologies of the Future: The Desire Called Utopia and Other Science Fictions, (London, New York: Verso Books, 2005). A similar argument was raised by Frederic Jameson in relation to the philosophies of the decentered subject and their risk of complicity with capitalist logics unless we learn how to read and fight their appropriation. Jameson writes: ‘This is indeed the ambiguity of postmodernism as a philosophy, that its progressive endorsement of anti-essentialist multiplicity and perspectivism also replicates the very rhetoric of the late-capitalist marketplace as such’ (163). And later, he continues: ‘today the presence to hand of the computer has blurred the economic issues, allowing one to assume that decentralization can now magically be achieved by the new technology, and thus flattening out and defusing the contradiction which Utopian solutions were once called into being to resolve, at least in the imagination. … But I hope some readers will want to take the position that postmodernism in economics is not at all the same as postmodernism in thinking or in philosophy; and that a principled rejection of the old “centered subject” (whether in psychology or in ethics) ought not to be discredited by the replication of its form in globalization, in business and in finance. This is an awkward historical situation, and it is by no means always cheap invective and mud-slinging to argue, as some of us have from time to time, that such replication is exceedingly suspicious and testifies to the way in which postmodern or decentered thinking and art reinforce the new social and economic forms of late capitalism more than they undermine it’ (164–65). ↑
17. Jacques Derrida, Positions, translated by Alan Bass (Chicago, The University of Chicago Press, 1981), 12. ↑
18. Rodolphe Gasché, The Tain of the Mirror: Derrida and the Philosophy of Reflection (Cambridge, Harvard University Press, 1986), 274. ↑
19. Gary A. Olson, ‘Jacques Derrida on Rhetoric and Composition: An Interview’, Journal of Advanced Composition, 10:1 (1990), 8. ↑
20. Derrida, Positions, 12. ↑
21. Olson, ‘Jacques Derrida on Rhetoric and Composition: An Interview’, 4. ↑
22. Anne Alombert, ‘Intelligences artificielles : vers une automatisation de la pensée ?’, Innovations : Le blog de Le Réseau de Recherche sur l’Innovation, 10 Juillet, 2023. Translation is my own. ↑
23. Anne Alombert, De la bêtise artificielle (Paris, Éditions Allia, 2025), 65. ↑
24. Giovanni Menegalle, ‘Two Regimes of Logocentrism’, Angelaki: Journal of the Theoretical Humanities, 28:6 (2023), 55. ↑
25. Derrida, Of Grammatology, 85. ↑
26. Menegalle, ‘Two Regimes of Logocentrism’, 50-51. ↑
27. Menegalle, ‘Two Regimes of Logocentrism’, 58. ↑
28. Gunkel, ‘AI Signals the Death of the Author’. ↑
29. Jacques Derrida, ‘Force and Signification’ in Writing and Difference, translated by Alan Bass (London and New York, Routledge, 2001), 21. ↑
30. Menegalle, ‘Two Regimes of Logocentrism’, 58. ↑
31. Derrida, ‘Force and Signification’, 11. ↑
32. Jacques Derrida, ‘Signature Event Context’ in Margins of Philosophy, translated by Alan Bass (Sussex, The Harvester Press, 1982), 326. ↑
33. Jacques Derrida, Limited Inc (Evanston, Northwestern University Press, 1988), 56. ↑
34. Derrida, ‘Force and Signification’, 32. ↑
35. Derrida, ‘Force and Signification’, 32. ↑
36. Jacques Derrida, Rogues: Two Essays on Reason, translated by Pascale-Anne Brault and Michael Naas (Stanford, Stanford University Press, 2005), xiv. ↑
37. Derrida, Rogues, xiv. ↑
38. Perhaps it is Hélène Cixous’s notion of ‘écriture féminine’ one of the attempts to articulate this vulnerable force that, putting the body in writing, resists the phallogocentric order of writing. Cf. Hélène Cixous, ‘The Laugh of the Medusa’, translated by Keith Cohen and Paula Cohen, Signs, 1:4 (1976), 875-893. ↑