3. AI Is Us

To this day, there is no “artificial intelligence,” if we take it as a human-made autonomous agent showing understanding and comprehension of the world. Were we to stay closer to the Latin etymology and interpret intelligence as the faculty of selecting and linking relevant data (inter-legere), then computer science has reached that goal for quite some time. There is a lot of room between these two conceptions of intelligence. The initials AI, in between a password and a code name (two letters standing for a noun that is not often fully uttered), register the ambivalence. In this essay, AI mainly refers to nonliving machines supporting and emulating mental operations, to the theories permitting them, and, by implication, to human interactions with them.1 There are obvious, massive differences between the kinds of AI that have been, or could be, developed. One remarkable point of divergence is about the degree of specificity, from the potentially powerful software designed for a domain, such as playing chess, to the so-called large language model (LLM) that does not really model human language but is capable of varied output, or to the putative artificial general intelligence that would draw inferences from separate faculties.2 Another concurring and noticeable difference in the constitution of AI comes from the introduction of self-learning algorithms. Would it be left completely to its own devices, the machine autodidact would still be programmed to self-train in order to mine the data it was fed. The generative AI that recently prompted reactions of enthusiasm, awe, fear, or discontent couples self-learning algorithms mining human-made data with a myriad of ad hoc interventions by engineers according to guidelines shaping content production (“reinforcement learning” and “fine-tuning”). There are still other kinds of internal divergences, especially with regard to future research. For example, would a generative system endowed with the capacity of additional learning in real time be a game changer? Would a larger introduction of randomness radically modify the calculation process? Would a higher level of coupling between a computer and a living brain introduce another kind of AI? I am not seeking to consider each variety or to promote one type because it would be either closer to a biological model or the most powerful. I recognize such differences, those I mentioned along with many others, but I believe that such engineering problems are not whereon we should focus our attention. The first thing to do is to reflect further on the degrees of separation between AI and “us” and avoid wholly setting apart one from the other.

AI is not not-us. We are intertwined with it in several ways. First of all, the “creature” was originally designed by humans, with all their cognitive, technical, social, and political limitations. As machines are now able to code, one can imagine future artificial entities programmed, and eventually built, by AI. This would not be enough to consider “we” had nothing to do with such a production, in the same way the science-fiction representation of the evil robot escaping human control—a tale currently recycled by marketing visionaries such as Yuval Noah Harari—would suffice to exonerate Homo sapiens from its own responsibility.3 Would the mechanical brain display agency, then our own plans, fantasies, hopes, and fears would be inscribed within it, to such a degree that humans could recognize themselves in this dissimilar other. More precisely, the development of AI has been strictly joined to an inquiry about the living mind. While cognitive science originally benefited from the progress of computerization, the whole field of artificial intelligence never stopped relying on experimental psychology for its own tenets. Automated content generators offer us a certain vision of ourselves that looks passably familiar, which would not happen if these machines were operating on a completely different plane. In addition, the large language models and any other comparable technical strategy are, in the first place, a function of their training set. The billions of parameters that the industry is vaunting are learned from immense collections of human-made artifacts (texts, pictures, sounds, movies, etc.). There would be no GPT output of any sort without the records of the collective achievements of our species predating actual AI. Finally, “we” are, and will be, collaborating so much with our mechanical extensions that “we” can only be transformed by our “transformers.” Besides the feedback circuit that allows users to reinject their assessment into the machine by selecting the best answer or the best version they get on their screen, computerized content and processes make their reentry in human minds, which cannot be without any consequence.4 We can only admit that we are already in the machine—or that it is in us.

Thus, we cannot content ourselves with a simple opposition between the humanities and computerized technology. We are rather dealing with different protractions of the human. The strife I named in the beginning of the book is not the fight of the discursive disciplines against the capitalistic and tech forces of information science or against objective rationality. Furthermore, within the corpus of the humanities themselves, we would have no difficulty finding enshrined conceptions of the human mind or self that are fully compatible with the oblique portrait of ourselves generative AI is showing us. From a humanities perspective, the reality and the discourse of artificial intelligence cannot be perceived as an outside enemy, or as a foreign ally, or as a transparent transcription of the philosophical. If we fall prey to that parallax error, we shall be unable to articulate our own position, either defending a camp whose raison d’être is social (and therefore epistemologically extrinsic) or ending de facto the very singularity of our scholarship. Through the implementation of theories, AI reveals what was not concealed, and everything we do that we can outsource to the machine is in effect secondary in the life of thinking.

A postscript. Plato referred to “some ancient discord between philosophy and the literary.”5 Immanuel Kant wrote an essay titled The Conflict of the Faculties on the differences of position vis-à-vis knowledge between the faculty of philosophy and that of medicine, law, and theology. The strife I am naming, for being easy to assimilate to societal blocks (the tech industry, groups of university departments), is, however, not solely associated with distinct discursive and epistemic registers (Plato’s discord) or with institutions (Kant’s conflict). I insist on the transversality of a conception that seeks to anchor the possibilities of the mind in (senseless) predictability, an idea well represented in the engineering sector but that is overly powerful within scholarship itself, where it has been argued over and over again. Thus, the strife could be added to the discord and the conflict that were previously qualified. As for the engaged fight that consists in mapping general political coordinates onto the world of research, it forms another sort of agonistic organization that might only partly correspond to the internal quarrel I am mentioning. Then, as much as anticapitalism or antifascism would like to come in handy right now in opposing AI surveillance, they would miss the mark if they were leading to yet another restriction of the possible along the line of systematic determinism.6