The only argument I ever made within GA that received some attention and rebuttal was my critique of Eric Gans’s concept of the “internal scene of representation.” My reason for making this critique was to make space within GA for a performative, pragmaticist, extended mind and, now we might say, distributed consciousness, understanding of thinking. My critique was very specific, targeting only the claim that the internal scene was created on the originary scene itself, and allowing for the possibility (I would have said, reality) that something like an internal scene of representation emerges post-literacy and post-Axial Age.
Here, I wanted to insist on the originary scene as constituted exclusively by the transaction between each of its members and the center and, politically, to resist planting the ideological preconditions of “liberal democracy” and “market society” on the scene itself. But, since this critique was also part of my insistence that we are always constituted by the center, which is therefore always speaking through us, maybe this distinction is not that important, especially since I have also consistently maintained that what is happening on the internal scene of representation, insofar as it’s separable from other scenes, is essentially rehearsal for entering those other scenes.
This would mean that those with a rich “inner life” are simply writing and performing a wider range of scripts for possible scenes, each of them in one way or another an extension of other scenes, ultimately leading back to shared ones which in some way were experienced as unfinished. So, our inner life entails answering unanswered questions from previous scenes, justifying behavior that drew forth or could have drawn forth accusations, preparing speeches for when future opportunities arise, and so on. It’s really being off-stage between scenes, making sure we remember our lines (and the proper gestures, postures, expressions, etc.) as long as we allow for a wider range of improvisation than conventional drama permits. I’ve never explored this conception, though, not even as much as I’ve done here already, so I thought it would be worthwhile taking a post to do so.
It may be that the emergence and rapid proliferation of language-generating machinery has impelled me to return to this issue, which so far has only really existed as a possible counter to defenses of the “internal scene.” How different “rehearsal” is from the “internal scene of representation” would depend on what, exactly, one thinks is going on on that internal scene and how much autonomy it is granted relative to other scenes. Maybe it’s a question of degree or angle of vision: is Hannah Arendt’s notion of thinking as a sustained internal conversation more like the internal scene or more like rehearsal? My strong suspicion is that Arendt herself was trying to defend a strong notion of an internal scene of representation, one which would be independent enough to be capable of resisting totalitarian propagandistic designs that would obliterate morality and judgment.
And I should then be able to answer the objection that would follow to “rehearsal”—how does it not resubmerge us in a pre-literate world that, in an advanced technological order, makes us especially vulnerable to mimetic contagion and conformity (just wanting to play well the roles allotted to us)? My own sense, as I’ve indicated, is that the theoretical concept has been constructed to meet the political need and that means we’d either have to meet that need differently or reconfigure it. We could trust our own judgment on the status of our interior monologues, but the “tensegrity” of the concept itself will lie in how its effects are manifested within shared spaces.
Insisting that we are always in some exchange with the center upon some scene has the productive effect of making us revisit the question of “thinking”—what it is, and what we’re doing when we’re doing it. The question of machine “thinking,” and the reactive response to it that Leif Wetherby calls “remainder humanism” (invoking the very tropes of the human that much of 20 th century though dedicated itself to investigating) increases in the interest in the question. “The LLM just predicts the next token while humans…” What, exactly? Can you honestly examine your own trafficking with language and deny that you do a great deal of next token predicting yourself?
“LLMs don’t really think, they just engage in pattern recognition, while humans…”—again, what, exactly? Engage in logical operations? As if that isn’t computable, or as if there weren’t a whole tradition of complaint that reducing thought to logic itself betrayed the human? Are humans not massively pattern recognizing? How do you think you find your way around in the world? Is pattern recognition not thinking? If you say it isn’t, you just beg the question. It also seems to me ludicrous to say that pattern recognition can’t generate anything new, since patterns always mismatch and something has to be framed differently to make a fit.
That existing LLMs can’t bring objects in the world into their representations is an important consideration, but maybe just an issue of combining LLMs with robotics, which is to say, adding or refining perception. I’m working with the hypothesis that the defense of the internal space as a space of thought, reflection, free conversation, etc., just comes down to wanting to experience oneself—to present oneself before others—as someone who of course wouldn’t turn his neighbor over to the Nazis.
What, then, is thinking? As I’ve pointed out repeatedly, Anna Wierzbicka’s Natural Semantic Primes enable us to leave the terrain of philosophy altogether in answering this question. “Think” exists on a field with an array of other prime verbs involved in referring to intention, speech and action: say, want, know, feel, hear, see, do (we could widen the field to bring other words, but I’m not going to conduct a sustained analysis here, just a little demonstration). The argument is that these words exist in every language, and all non-prime words would ultimately be explicable in terms of the primes, which are not themselves explicable in simpler terms, so we all, as language users, know what these words mean, which is to say we know how to use them.
So, what is thinking? It’s something I say I am doing when I’m not, or not only, feeling; I say I think when I do not want to say I know; it’s what I say I’m doing, or what I say I am taking you to be doing, when I or you see things in some way; I think when I don’t do; thinking is what I tell someone they should do before saying something; I say I am thinking when just saying I want something is not good enough; and so on. (I should, as I often don’t, make it clear that Wierzbicka bears no responsibility for the directions in which I take her inquiries.) Not only do we not need an internal scene of representation to speak about thinking (or knowing, or wanting, etc.) in this way but it would seem to me highly artificial to try and force it into that (ultimately solipsistic) mold.
We could assign something like deferral tokens to each use of any of these words (and we will get around to that) and then a distribution of tokens across various uses of them—precisely because they are primes, we can say they don’t really have to mean any one thing—maybe the whole linguistic substrate can be thought of differently, as a kind of residue of the most elemental forms of declarative culture that serves as a substrate of all language use. “Think” is interesting in this regard because across its various uses it does seem to me there is always some element of delay and deferral, which we could internalize if we feel compelled to but could also speak of as a prolongation of the scene—“I don’t think so”; “I’m still thinking”; “think before you speak”; “what do you think about this?”; and dozens of others you could easily generate for yourself all seem to me (provisionally) to involve some prolongation of the scene, some delay of a decision, of a doing.
Rehearsing, then, is answering a different kind of question than “what is thinking?,” which might not even be a real question. To return to the political and moral stakes, we can ask a question: let’s say that, instead of preening oneself on being the kind of person totalitarian modes of thinking could never penetrate, we were to concern ourselves with how we might teach someone to resist propagandistic, which is to say repetitive, sloganeering, deliberately selective so as to render invisible counter-arguments, intrusions upon one’s own language making. The dominant method, now commonly called “critical thinking” (the genealogy of which, so far as I’ve discovered, begins with John Dewey, which should raise some questions), invariably involves trying to bolster that internal scene of representations with supposed methods, lists of questions, rules of evidence, etc., etc., all of which is doomed to fail miserably because which questions are relevant, which claims defy logic and with what implications, what are the rules of evidence, etc., are only learned on scenes where all these features of language take on historically specific forms (the rules of evidence, and therefore “critical thinking,” modeled on a 21 st century US courtroom would be unrecognizable to someone with a form of critical thinking modeled on an 18 th century French courtroom).
All you can do on the pedagogical scene I’m proposing for examination is prepare students, and even that not too well, to perform better on similar pedagogical scenes. And, in the end, I suspect someone like Arendt would concede that “thinking” cannot be taught, and there is no method for it and that therefore at best one could instigate or inspire it by performing thinking in others’ presence. But this already gets is pretty close to “rehearsal.” (Note that the one thing machines definitely can’t do, issue promises, is itself a performative speech act. And the relation between promising and thinking would be an interesting avenue to explore)
The most classic example of this way of inducing thinking is in Hamlet’s plan to “catch the conscience of the king” with a “play,” an example I’ve referred to occasionally over the years. I agree with the reading of the play that concludes that the real tragedy is that Hamlet abandons this approach, which we, but not Hamlet, can see is working from Claudius’s desire but inability to pray. You get someone thinking by representing them back to themselves, by inhabiting a role implicit in the scene which they give evidence of rehearsing for and thereby interfering in their rehearsal and scriptwriting process. In the process you do the same for yourself by taking direction from the other.
This breaks into the fantasy world, where one triumphs, is vindicated, speaks the lines one never got a chance to utter in real life, received the deserved approbation, avenges oneself on dastardly foes, etc., by forcing the rehearsal mode into anticipating a wide range of jagged, unpredictable scenes. Rather than punishing and shaming others for not demonstrably using your preferred “critical thinking” method you get in between their thinking, saying, knowing, doing, etc., and let them get in between yours. And it might very well be possible to scale rehearsal up computationally without damaging its “integrity,” because it doesn’t really have integrity but, rather, an ongoing state of play.
Monica Belevan, on her Covidean Aesthetics substack, has provided a reading of LLMs as essentially dramatistic, and one’s engagement with them through prompts is necessarily dramatic and improvisational. I’m currently very opposed to corporately produced LLMs but very much in favor of building language models based on purposefully curated databases for research projects that couldn’t have been conducted otherwise and I think the rehearsal model will help us to proceed without agonizing over whether we can now say that the machine is or is not, in fact, “thinking” because now we can accept that lines of code produce lines of scripts that we can co-design so as to force our rehearsals into previously unimagined scenes—scenes which prolong apparently closed scenes of our histories.