The previous Chronicle’s discussion of the limits and possibilities of AI provoked some discussion that suggests the need for a more rigorous set of categories, an ontology, of what I shall call for the moment transcendence.
The point of departure for that Chronicle was the emergence of ChatGPT as an AI program that can compose meaningful discourse and rather than simply finding words a la Google can answer questions based on a huge database of online information. Thus the question arises as to how to distinguish such activity from human thinking. Clearly when humans and computers solve math problems, the difference in method is unimportant; the point is simply to reach the correct answer, which is presumably unambiguous. In other words, assuming a priori that humans have souls and computers do not, there is no “soulfulness” in solving such problems. So much is simple enough; and the same can be said to apply to such activities as playing chess or Go: the machine can beat us at Go because it can “learn” the game by playing against itself and continually improving its algorithms beyond the capacity of even the most talented humans. There is no “soulfulness” in playing these games; success in them is fully algorithmic, which is to say, mechanical.
But with ChatGPT we get into the realm of composition, and with DALL-E into that of graphic art, activities that have always been associated with the “soul,” since they express and reflect our profoundest emotions, unlike mathematical and related phenomena that express our soulfulness only peripherally, such as games that allow us to discharge our rivalrous tendencies in a civilized manner. Thus the claim, which at first seemed scandalous, that such artifacts as poems, novels, paintings, musical compositions, etc., can be produced by “soulless” machines, is so no longer. This material, like the simple conversations for which the “Turing test” was first devised, has as its only test of “soulfulness” our judgment as to its human or mechanical provenance. Now that we have seen the results of machine learning in the purely algorithmic sphere, it should not surprise us that its products in the domain of what may be called “meaningful composition” can appear equally authentic. No, the machine cannot feel the emotions it “expresses” in a poem, but with a sufficient corpus of artistic examples, and the ability not merely to “understand” them literally, but to “learn” what sound and vocabulary combinations are harmonious, what kinds of tropes or narratives have given humans pleasure, etc., we should have increasing difficulty in distinguishing their products from those of our fellows. This has certainly already been tested many times by students handing in papers composed by such programs.
Clearly this level of representational transcendence, which we had thought protected from mechanical reproduction as mathematical calculations were not, since it involves the production of meaningfulness of whatever kind, requires human input in a sense unnecessary in the first category. The Go program learns by playing against itself; it has no need to study human master games. Whereas it is inconceivable that an AI system could learn to write poetry without beginning by digesting a corpus of human poems, just as on a lower level ChatGPT has done. Perhaps at some point inputting such compositions would no longer be necessary, but without the original human corpus, the task of programming poetry or other art forms for human taste would be of a wholly different level of difficulty, requiring a simulation of the conditions of composition not (yet) conceivable.
Yet, to ask the Ur-question of science fiction first posed to the modern world in Frankenstein, we are not tempted to assign to these AI implementations a “soul.” We may define our soul most simply as the individual’s repository of the sacred, a term which we have previously defined as the subject of the imperative to preserve the proto-human community, or in other words, to give it a chance to become human by renouncing potentially violent rivalry. Whether one follows a specific religious tradition, simply “believes in God,” or rejects religion entirely, I think we should all be able to agree that the sacred is not, like God, an object of faith, but an anthropological reality: our participation in humanity’s shared experience of the need to stand back from, to defer, the rivalrous impulse to possess for ourselves what alone makes possible the continued existence of the community. We conceive God as the subject whose will compels this deferral, but the compulsion itself is the necessary condition for our existence, and this, knowing that we are all tempted by the “original sin” of violating it.
What makes the soul problematic is its ambivalent character—like that of the meanings of language, but unlike them, of ultimate personal concern—between my own sense of it and its reality for the community. The meaning of any word is not situated in an objective place; it exists within the mind of each of us, but only insofar as we conceive ourselves to share it. In the general case, this poses no problem, but as usage evolves, connotations and syntactical relationships change, so that an expression that seems natural to one person strikes another as incorrect. Today, for example, the word “like” is increasingly used as a conjunctive adverb in a way that strikes me as illiterate (“he came by like I asked him to”), or worse, in the construction “he was like” to mean simply “he said.”
One’s sense of petty horror at such constructions reflects the violation of what we feel as a little bit of the sacred pact that holds the community together. It reflects a minor offense to my soul whose coherence depends upon that of the human community, so that a linguistic usage that strikes me, not simply as incorrect, but as affirming a new correctness that I deny puts in question a little bit of my security in belonging to this community. This is a phenomenon that the aging process especially teaches us; our community-conception varies from one generation to another, in each generation seemingly more rapidly than in the last.
Let us suppose we have an AI device called Arnie. If Arnie can write decent sentences, he can surely be made to react against those it considers badly formed, just as a writing teacher forces pupils to abandon turns of phrase incompatible with scholastic prose. Assuming he continually acquires new input from cyberspace, Arnie’s notion of correctness can also evolve by incorporating trends in language usage.
Just as in writing sonnets or compositions, there is in principle no outward manifestation of human-like conduct that Arnie cannot be taught to perform. Like Samantha in Her, he could learn to manifest all the appropriate verbal signs of love—and in a more advanced, robotic version, all the physical ones as well. Yet we would still not be obliged to grant him an “inner life,” a soul of his own, that would link his behavior to the foundation of the human in the originary deferral of rivalrous appropriation.
Would this manifest itself as a problem for Arnie himself? Would he like the replicants in Blade Runner express regret for not being fully human? That too would depend on his programming. We can let AIs teach themselves any aspects of human behavior we like, but only if we implement the “emotions” we want them to feel. It is a myth to believe that at some level they could transcend their manufactured status and begin experiencing “emotions” independently of the parameters we had originally installed in them. Or to put it more prudently, the sort of machines that might eventually, without having been programmed to do so, become dissatisfied with their status as tools of humanity and decide to throw off their human yoke are so far beyond our current understanding that we cannot describe them in an intelligent fashion.
Fred Jameson’s book title, The Prison-House of Language, roughly translated from Nietzsche’s “in dem sprachlichen Zwange” in his late notebooks (see Emily Apter’s “The Prison-House of Translation?”, Diacritics 47,4 [2019]: 54), seems the ultimate expression of the frustration of thinkers at the end of the era of metaphysics. Hegel was no doubt the last philosopher whose work truly embodies the faith that all of human reality could be expressed in the categories of metaphysics, that philosophy was indeed the “queen of the sciences” in the universe of Galileo and Newton.
Indeed, even the early years of Einstein and quantum theory did not seem to contradict this idea. As a reader of Scientific American since the 1950s, it is only in the last couple of decades that I have begun to feel that these explanations for the layman were no longer simplifications of the work of physicists but merely descriptions of them. Perhaps the most obvious sign of this is the use of superlatives to describe the sound and fury of cosmic phenomena as though we were witnessing them (with appropriate sound effects) in a planetarium. For there is simply no intuitive way of explaining in common-sense terms the relations between quantum theory and relativity, black holes and quantum entanglement, the nature of “dark” matter/energy… What is most discouraging is that, despite the assurances of physicists that we just need a still more powerful collider to produce the particles missing from our theoretical paradigms, one cannot avoid the impression that rather than our coming closer to a reasonably coherent “theory of everything,” the goal keeps getting farther away the more deeply we probe physical reality.
Physics is, of course, not metaphysics, as Aristotle made clear. But Jameson’s title concerns not physics but human reality; it paradoxically uses language to question our capacity to put into words the truth about our world and ourselves. And if generative anthropology, which I have proposed as a step toward the solution to this problem, has not won many adherents, this reflects its so far limited achievements in liberating us from Jameson’s prison. The originary hypothesis is a radically new way of thinking about human origin that I see no need to abandon, but the connection between our originary scenic model and its historical avatars is by nature incapable of supplying us with a tool to understand/influence history in more than the marginal sense in which realizing the most fundamental elements of human social organization allows us to reject the anthropologies that have led to the various forms of totalitarianism (which seems to be proving itself, as Arendt had suggested, the real model underlying the various forms of “socialism”), without however offering us more than a demonstration of the necessary impossibility of providing a formula for “the good society” more precise than my favorite Churchillian quote as “the worst… except for all the others.”
The soul that humans have and that no machines can duplicate is, as AI negatively demonstrates, independent of all our intellectual conquests. But as the necessarily incomprehensible notion of God the sacred Being demonstrates, Arnie cannot… not simply conceive God, because humans cannot “conceive” him either, but experience a sacred revelation that, beyond its semantic content, would recreate for him the scene of the community’s presence to itself.
No doubt we can name God, but such naming remains within the prison-house of language, where we can also “name” square circles or true lies. This is the basis of the Jews’ refusal to give God a “normal” name; the Orthodox write “G-d” and in conversation call him “ha Shem,” meaning “the Name,” a nice way of making naming a paradox. Yet we can aspire to God as the Being who is the source of the sacred.
Arnie can talk about God as well as we humans; anything we say about God can be placed in a database to which he will have access. But although Arnie can say anything that we can say about God, we will always know that the source of this thought is in us, that like the man in the Chinese room, Arnie can reproduce it, but could never have generated it, lacking the scenic experience of the sacred, in which, as Derrida saw, deferral generates différance, and the aborted gesture of appetite becomes the sign of desire. When we speak of God, we refer to the source of this deferral, which we recognize in ourselves but at the same time realize as an interdiction imposed upon us.
An atheist can deny the Being of God and attribute this restraint to our sense of the human community, whose communal will we fear to contradict, or like pre-Axial “immanentist” believers, attribute it to the interdicted object(s) it/themself/ves. It was only after many millennia that, during the Axial Age as defined by Jaspers, as humans from different communities met and communicated, the discovery was made that they all had in common this phenomenon of différance. Historically, whether or not as an extension of the henotheistic experiment of Akhenaton, the Hebrews thematized for the first time the identity of these experiences of interdiction, whose Subject they called by various names, but today call simply The Name, the ultimate object of the pointing we call language.
None of these explanations is “final,” because, precisely, words cannot escape the prison-house of language, but we understand them because our history has been lived through them, and even if many or most of us no longer share on a regular basis, as had long been the case, a communal ceremony in memory of the originary scene of the production of the sign, we all still share the sense of the sacred, however much we may struggle against it.
What AI teaches us about this sense and the soul that contains it is that, however closely it can imitate the language in which we express it, that language’s grounding in experience is foreign to it, and will always remain so—unless and until at some point we become willing and able to create beings who can experience for themselves the scene of mimetic desire, and create the sign of its renunciation.