Aesthetics of Spectrality – Part 1: AI and Ghosts, the Case of the Series “On This Day… 1776”

 

1. The Return of Ghosts 

I don’t know whether it’s just my impression, but I find very strange the AI-generated images in Darren Aronofsky’s series "On This Day… 1776", produced to commemorate the 250th anniversary of U.S. independence and released on Time’s YouTube channel. 

Of course, AI-generated images will continue to evolve, but in this series one has more the impression of watching strange figures moving before our eyes than of humans meant to embody George Washington or Benjamin Franklin. Put differently, the vacant gazes, the camera movements, and the overall atmosphere produce an effect of spectrality. 

An effect that is not specific to this series, since many AI-generated videos- and some of them are very beautiful - give the feeling of seeing a world haunted by spectes. 

And without wanting to advance an overly hasty thesis about AI, I almost feel that the very paradigm of this technology has something to do with spectrality. 

Whether it is in old photographs that are brought back to life; in those famous videos where one can take a selfie with one’s favorite stars; in those that plunge us back into a city’s past; or in applications that allow us to adopt the aesthetic of a well-known film, such as the filter inspired by Studio Ghibli - everything in the world of AI seems to refer to the notion of the revenant. 

A notion that designates that which returns, insists, persists without being fully there, and which inevitably brings to mind the reflections of two great thinkers of spectrality: the art historian Aby Warburg and the philosopher Jacques Derrida. 

The first, Warburg, with his notion of Nachleben (survivance), which highlights the ghostly return of forms originating in Antiquity at the very heart of Renaissance art. 

And the second, Jacques Derrida, whose philosophy, in its later period, took the name hauntology. Without going into the details of his often very complex thought, I would emphasize here that if the figure of the spectre became central to it, it is because, unlike ontology - that is, a thinking of Being as full presence - Derrida always granted great importance to the notion of “traces,” that is, to what is neither entirely present nor entirely absent. 

And AI-generated videos, especially when they seek to mimic reality, belong precisely to this neither entirely present nor entirely absent. 

But if, in cinema - the ultimate art of the return of ghosts, according to Derrida - or in Warburgian survivance, we are dealing with the trace of something that once was present, it is legitimate to ask what kind of ghosts we are dealing with in these AI videos that look so realistic, but not quite. 

2. The Return of Which Ghost? 

If this question may seem theoretical in the case of the beautiful, mesmerizing videos generated by AI, it becomes clearly existential when it concerns videos that touch on our History, as is the case with "On This Day… 1776." 

Of course, technically speaking, those who take an interest in AI know that the spectres we see in this film - and in so many others - bear the trace of all the data used to train them. From this comes the impression, when facing AI-generated images, of witnessing the “survivance” of images we have already seen. 

But this “survivance” does not concern images alone. Those who use ChatGPT have certainly noticed that, in the texts generated by this tool, we witness the return of words, expressions, and even punctuation marks that had fallen into disappearance, having long ceased to be used. 

And, speaking of ChatGPT, those who use it must have noticed certain strange behaviors, close to amnesia, such as when, in the middle of an exchange, a response suddenly appears that relates to an earlier conversation. 

This gives rise to the strange feeling that, beyond the traces of our data, what is spectral in AI is its very behavior, even its identity. 

This is, in any case, the thesis put forward by computer scientist Andrej Karpathy - formerly of Tesla and OpenAI - in his text “Animals vs. Ghosts.” 

In this text, he explains that research into artificial intelligence is closer to the summoning of ghosts than to the training of animals: “today’s frontier LLM research is not about building animals. It is about summoning ghosts.” 

Without going into the details of his argument, I would note that the major difference he identifies between educating an animal and training an AI lies in the fact that, in the case of an animal, the process involves guidance aimed at bringing it to maturity, whereas with AI we are dealing more with constant supervision, intended to prevent the “out-of-context” survivance of learned information. 

The example Karpathy gives is that of the zebra, which, twelve hours after its birth, is already capable of running. That is a clear evidence of innate abilities, something AI obviously still lacks. 

Imaginative or not, this theory nevertheless resonates with the feeling I experienced while watching On This Day… 1776, since, more than seeing the spectre of the Founding Fathers of the United States, I had the impression of seeing that of a machine in the process of claiming our History for itself. 

This produces that “uncanny” feeling—to use the term Freud employs to describe haunting—of being dispossessed of our own History, since the process of identification, so essential to cinema, struggles to take hold when faced with characters whose gaze seems to grant access to the depths of a machine rather than a human soul. 

3. Towards Erasure? 

And, speaking of gaze and the feeling of dispossession, I cannot conclude without mentioning the famous photograph of Balzac taken by Louis-Auguste Bisson. In this photo, Balzac makes a point of not looking at the camera’s lens, convinced that photography possessed a magical power, capable, in his view, of stealing a layer of our soul. 

Even if his intuition may seem archaic, I think it largely explains why, when faced with a photograph or a film, we feel as if we are confronting a ghost. But if, in photography or cinema, only a part of us is “stolen,” what might happen to us when, as can be the case with AI, all our data is copied? 

I am well aware that the field opened by this question probably belongs to science fiction. Yet it seems to me that it is already being raised in subtext in Hollywood, where several actors, such as Matthew McConaughey, are seeking to copyright their image to protect themselves from the excesses of AI. 

And, speaking of science fiction, this very question of image copyright lies at the heart of Ari Folman’s film "The Congress," adapted from a science fiction novel by Stanislas Lem. 

Despite moments that can feel a bit heavy-handed, the film remains strikingly relevant, with its story of an actress, played by Robin Wright, who, after agreeing to be scanned so that her image could be used at will, ends up disappearing - in the metaphorical sense - since she can no longer work. 

And, without wanting to be overly alarmist, I wonder whether, in the future, when three-quarters of the images in circulation will be generated by AI and that our History will be told through these images, we may ultimately disappear metaphorically, becoming nothing but traces, just as Robin Wright does in "The Congress."