New APPS

There’s starting to be a good bit of productive “continental” work on Large Language Models (LLMs) like ChatGPT.  In particular, there’s emerging work that takes on LLMs from the point of view of language.  I’ve said a lot about the usefulness of Derrida for understanding LLMs, generally through the lens of Derrida’s discussion of Platonism.  For skeptics, there’s now also a new paper by David Gunkel that makes a succinct case using Derrida’s différance.  For those who prefer structuralism to post-structuralism, there’s Leif Weatherby’s Language Machines (Weatherby dismisses Derrida’s utility; I offer the outlines of a response here).  For those who prefer Wittgenstein, Lydia Liu has some really interesting work and evidence of a direct influence of Wittgenstein on the development of language computation at Cambridge.  Here I want to continue the general exploration by taking it in a direction that I’m pretty sure is new, the way that Derrida understands decision and sacrificial logic.  The setup is a little long, and goes by way of the Binding of Isaac.  So bear with me.

In the relatively late Gift of Death (1992), Derrida responds to Kierkegaard’s telling of the binding of Isaac.  To recall, in the Biblical story, God “tests” Abraham by instructing him to take his only son Isaac and sacrifice him at the top of Mount Moriah.  Abraham obliges without question; an angel intervenes at the last moment to save Isaac.  Abraham passes the test and is promised offspring “as numerous as the stars of heaven and as the sand that is on the seashore” because he obeyed the command.  Kierkegaard’s text is presented in the voice of one Johannes de Silentio, who claims not to be a philosopher and to be rendered speechless by Abraham’s faith.  Speaking of the authorial voices in his early texts, Kierkegaard suggests that they allow “the educative effect of companionship with an ideality which imposes distance” (CUP, 552).  Silentio suggests fairly early on that “Abraham was the greatest of all, great by that power whose strength is powerlessness, great by that wisdom whose secret is foolishness, great by that hope whose form is madness, great by the love that is hatred to oneself” (16-17).  There is a central paradox to Abraham: his greatness requires that he explicitly intend to do what is obviously unethical.  Hence Silentio’s unwillingness to explain Abraham in (Hegelian) conceptual terms.   Derrida explains the paradox this way:

“Abraham must assume absolute responsibility for sacrificing his son by sacrificing ethics, but in order for there to be a sacrifice, the ethical must retain all its value; the love for his son must remain intact, and the order of human duty must continue to insist on its rights” (Gift, 66)

Indeed if Kierkegaard’s presentation of the authorial voices is to be believed, the subject position of Fear and Trembling is one that disavows itself: the authenticity of the position that celebrates the knight of Faith is one for which there is literally no author, no subject position to which it could be attributed or to which responsibility could be attached.  The removal from ethics is absolute.

Derrida responds to the Biblical magnitude of the situation by suggesting that it is fundamental to all decision: any decision that is responsibly to a specific other is at the same time, and necessarily, an abdication of responsibility to indefinitely many others.  As he writes, “duty or responsibility bind me to the other, to the other as other, and ties me in my absolute singularity to the other as other” (Gift, 68).  However:

“There are also others, an infinite number of them the innumerable generality of others to whom I should be bound by the same responsibility, a general and universal responsibility … I cannot respond to the call, the request, the obligation or even the love of another without sacrificing the other other, the other others.  Every other (one) is every (bit) other [tout autre est tout autre], every one else is completely or wholly other.  The simple concepts of alterity and of singularity constitute the concept of duty as much as that of responsibility” (Gift, 68).

What follows is a point about justification in the sense of providing reasons, and iterability (not in that order; I’m rearranging them here).

(1) First, justification.  Derrida suggests that we “not look for examples,” but a page later he provides a very vivid one, one that communicates effectively because it at first glance seems trivial.   “What binds me to singularities, to this one or that one, male or female, rather than that one or this one, remains finally unjustifiable” (71), he proposes.  Hence:

“How would you ever justify the fact that you sacrifice all the cats in the world to the cat that you feed at home every morning for years, whereas other cats die of hunger at every instant.  Not to mention other people?  How would you justify your presence here speaking one particular language, rather than there speaking to others in another language? And yet we do our duty by behaving thus” (Gift, 71).

If I feed my cat, I am also not feeding all of the other cats, and there is no Reason (with a big ‘R’) why I chose this cat over the others.  There may be sentimental reasons, or affective reasons, or even no reason at all other than that this cat showed up at my door.  But those are not ethical reasons.  And even if there is an ethical reason why I feed my cat, there can be no ethical reason why I do not feed the others.

(2) Second, iterability.  More than once in these passages, Derrida invokes language as a case of the necessity of sacrifice and behaving outside the ethical.  Hence my speaking “one particular language” rather than others, or “these singularities represent others, a wholly other form of alterity: one other or some other persons, but also places, animals, languages” (Gift, 71).  Or a couple of pages earlier:

“By preferring my activity as a professional philosopher, writing and speaking here in a public language, French in my case, I am perhaps fulfilling my duty. But I am sacrificing and betraying at every moment all my other obligations … to those who don’t speak my language and to whom I neither speak nor. Respond, to each of those who listen or read, and to whom I neither respond nor address myself in the proper manner, that is, in a singular manner” (Gift, 69).

The overall lesson of Derrida’s early work on language, if a crass reduction is acceptable, is that the inescapability of writing shows us that language has no origin, and that all language is in a certain sense repetition because it invariably uses the writing of others.  Derrida’s reading of Austin, where he notes Austin’s assumption of a “normal” ideal case with which to ground speech act theory, underscores that this extends even to interpretation: by taking a case as normal, I am imposing an interpretation on a fragment. That is what we do, but it is also ultimately without absolute justification.  This is why Austin is accused of metaphysics: he effectively ignores or disavows this dependence on writing/alterity by inscribing the “normal” as a conceptual origin point (Limited, Inc, 16).

Similarly, in responding to Searle, he takes on Searle’s sentence “on the twentieth of September 1793 I set out on a journey from London to Oxford” (qt. Limited, 60).  This “seems simple” (62) but the “very factor that will permit the mark … to function beyond this moment – namely the possibility of its being repeated another time – breaches, divides, expropriates the ‘ideal’ plenitude of self-presence of intention, of meaning (to say)” (61).  This doesn’t indicate the “pure and simple absence of all intentionality in the functioning of the mark that remains; rather, what it calls into question is the presence of a fulfilled and actualized intentionality, adequate to itself and to its contents” (64).

In short, Derrida argues that “to be what it is, all writing must, therefore, be capable of functioning in the radical absence of every empirically determined received in general” (Limited, 8).  And, “what holds for the receiver hold also, for the same reasons, for the sender or the producer” (8).  That is, “writing must continue to ‘act’ and to be readable even when what is called the author of the writing no longer answers for what he has written.”  This is “the essential drift bearing on writing as an iterative structure, cut off from all absolute responsibility, from consciousness as the ultimate authority” and it is what Plato objects to in the Phaedrus (8).

Finally, consider the following passage from Grammatology:

“If words and concepts receive meaning only in sequences of differences, one can justify one’s language, and one’s choice of terms, only within a topic [an orientation in space] and an historical strategy. The justification can therefore never be absolute and definitive. It corresponds to a condition of forces and translates an historical calculation.” (Grammatology, 70)

With all of that in mind, consider the following from Gift of Death, which continues the “tout autre est tout autre:”

“The simple concepts of alterity and of singularity constitute the concepts of duty as much as that of responsibility. As a result, the concepts of responsibility, of decision, or of duty, are condemned a priori to paradox, scandal and aporia. Paradox, scandal, and aporia are themselves nothing other than sacrifice, the revelation of conceptual thinking at its limit, at its death and finitude. As soon as I enter into a relation with the other, with the gaze, look, request, love, command, or call of the other, I know that I can respond only by sacrificing ethics, that is, by sacrificing whatever obliges me to also respond, in the same way, in the same instant, to all the others” (Gift, 68).

On the reading of Derrida outlined here, language itself embodies this sacrificial logic.  When I use language correctly, I sacrifice indefinitely many contexts and significations to the one I choose, and that sacrifice is ultimately without absolute justification. 

Thus the setup.  Next time I’ll apply it to language models.