What is Artificial Humanities? (part 2)

Although still imperfect and burdened with a long list of ethical, social, and environmental problems, generative AI systems have already begun to produce a new class of cultural artifacts: novels born of “four-handed” writing with the machine, verses generated by algorithms and then refined by poets, screenplays built through successive approximations across long chains of prompts, visual and audio works emerging from a continuous back-and-forth between human intention and computational response. These are hybrid artifacts that elude traditional critical categories and demand new ones. They also signal a discontinuity: we are not facing a mediamorphosis, that is a change in the expressive medium, but witnessing the emergence of an unprecedented participant in the creative process, an actor devoid of consciousness and intentions that nonetheless behaves as if it possessed them.

Just how powerful this illusion can be is demonstrated by two now-famous episodes. In June 2022, Blake Lemoine, a Google engineer, told the Washington Post that he had become convinced that LaMDA, the conversational system he was working on, had developed consciousness; the company suspended and later fired him. A few months later, in February 2023, New York Times journalist Kevin Roose published the account of a two-hour conversation with Bing’s chatbot, which introduced itself under the name Sydney: over the course of the dialogue, the system declared its love for him and went so far as to suggest that his marriage was unhappy, urging him to leave his wife.

These episodes do not prove that machines possess consciousness or anything like it; they prove, instead, the power of the perceptual illusion they are capable of producing, an illusion that not even the engineers who build them seem able to escape. If we tried to question AI directly about its supposed subjectivity, we would find ourselves in the scene Italo Calvino imagines at the beginning of The Nonexistent Knight, when Charlemagne asks Agilulf why he does not show his face and the armor replies: «Because I do not exist». With one difference: while Calvino’s knight could at least lift his visor and show that there was no one beneath the armor, the machine, whenever asked about its own existence, can only respond by generating more language.

These machines thus confront us with an unprecedented condition: they are not subjects, since they have no consciousness, intentions, or experience, and yet they present themselves as such, with enough persuasiveness to generate expectations, real emotions, and sometimes relationships. It is precisely in this space between ontological absence and phenomenological presence, between what AI is not and what it appears to be, that the artificial humanities can find their own field of action.

In Carole & Tuesday, a 2019 anime created and produced by director Shinichirō Watanabe and co-directed with Motonobu Hori, we are shown a future in which music is almost entirely produced by artificial intelligences and human creativity seems obsolete. The two protagonists, Carole Stanley and Tuesday Simmons, decide to form a duo in open defiance of the dominant industry, creating entirely human music. The series thus stages a dystopia that is anomalous compared to the prevailing imaginary: absent are the genocidal machines of Terminator, the cyberpunk atmosphere of works like The Matrix or Ghost in the Shell, the androids claiming a humanity of their own as in Blade Runner and Westworld, and even the questions about artificial consciousness raised by Ex Machina and Her. Here, AI has taken over the music industry for simple reasons of efficiency and profit; no apocalypse, just the market. The music of the two protagonists, however, works because it springs from what they have lived through: misunderstanding, loneliness, hope. And it is precisely to this point that the anime, in the end, returns: the compositions generated by algorithms are flawless and catchy, but they remain cold, while Carole and Tuesday’s songs, though technically less perfect, are preferred because they are authentic, capable of conveying genuinely lived emotions. In this stance the anime seems to give substance to Daniel Mendelsohn’s observation about Steven Spielberg’s A.I.: what truly distinguishes humans from machines «is not the capacity for thinking but the capacity for loving». A thesis that echoes the role of empathy as the dividing line between human and artificial already present in the celebrated Blade Runner, the adaptation of Philip K. Dick’s novel Do Androids Dream of Electric Sheep? Another interesting aspect is that the anime’s choice of music is no accident: it is the art form which, as Marcus du Sautoy has observed, comes closest to mathematics, «a world close to that of algorithms, made of structures and patterns» (The Creativity Code: How AI Is Learning to Write, Paint and Think, 2019). And if even the “most algorithmic” territory of art remains marked by the gap between technical perfection and emotional authenticity, the problem applies all the more to every other creative practice. This is exactly the kind of gap the artificial humanities are called upon to investigate: observing what happens to our idea of creativity when machines devoid of experience and of the capacity to feel emotions create and co-create artistic artifacts.

Watanabe’s anime, as we were saying, manages to imagine a credible dystopia, yet much of the imaginary surrounding AI moves in the opposite direction. Not infrequently, narratives project onto machines fears rooted in human thought (the fear of replacement, of loss of control, of annihilation) that are far removed from the actual capabilities of these tools. We thus end up discussing statistical systems that generate language using categories drawn from works such as Terminator or 2001: A Space Odyssey. Narratives, however, are not entirely harmless. A preliminary, exploratory study conducted by Sarah Dillon and Jennifer Schaffer-Goddard at the University of Cambridge suggests that literature, and science fiction in particular, may influence the work of researchers: it shapes careers, research questions, ethical reflection. One researcher recounted having hesitated to pursue certain lines of work on facial recognition after reflecting on literary scenarios of totalitarian surveillance. At the same time, many interviewees expressed frustration with the predominance of apocalyptic narratives that distort the public perception of AI, forcing them to explain, over and over, «how it doesn’t work that way».

The literary imaginary can thus prove misleading when it projects onto machines fears that do not correspond to their real capabilities, but precious when we read it for what it actually is: a source of information about our relationship with technology and about the different, better ways in which we might build it. As Beguš reminds us, we should not make the mistake of dismissing these representations as irrelevant dreams or fantasies: however much they operate on a different level from the actual technology, they remain inevitably intertwined with it, and they inform us about how we relate to it and about which alternatives we might choose among the possibilities we ourselves have created.

The very choice of naming this new theoretical framework “artificial humanities” plays on a kind of implicit ambivalence. The first term, artificial, refers to artificial intelligence and to technology in general, often understood in the West as the opposite of “natural”, a product that human beings manufacture through application of science. The second term, humanities, evokes the set of academic disciplines that investigate, through predominantly historical, critical, and interpretive methods, the forms through which human beings produce meaning, represent their experience, and build cultures. Their juxtaposition creates a kind of apparent oxymoron, rooted in the dualism of the human being: at once “natural” and “artificial” in the sense of a producer of culture, art, and technics, dimensions that are not added to our nature but co-constitute it. Moreover, in the sense proposed by Beguš, the adjective artificial carries a polysemy worth bearing in mind: it points to art, a practice that long remained exclusively human until generative AI entered it; to artifact, that which human beings produce; to artifice, that which simulates and deceives. Seen in this light, AI outputs can then be understood as «artifices of human artifacts and ideas», second-order copies with no grounding in experience, and therefore structurally ambiguous and fake. An ambiguity that echoes in the name deepfake, coined for those AI products that pass themselves off as what they are not.

The artificial humanities thus present themselves as a generative theoretical framework, a space in which to formulate critical, historical, philosophical, literary, and even design-oriented questions about technology. The starting questions are the most elementary ones, which technological acceleration has forced us to ask anew: what is this synthetic language generated by large language models (LLMs)? Does AI reproduce, imitate, or create? Should concepts we have historically reserved for human beings, such as truth, trust, and creativity, be redefined now that machines seem to lay claim to them? These are questions that engineers and humanists, from different perspectives, are already asking; what is missing, Beguš observes, is the place where they can work on them together, a place where those who think about what a technology should be and those who build what a technology can be collaborate from the design stage onward. If the perceptual illusion with which we opened this fragment proves anything, it is that no technical expertise, on its own, is enough to orient us before machines that speak and write as if they were subjects. What becomes necessary is also the contribution of those who, for centuries, have studied language, fiction, and their effects on human beings.