Welcome to my newsletter about Playing Software. I am using this space to think as I type about AI, play, and other important topics. I’m still trying to figure out what on earth I’m doing, so expect turbulences.
I was a young and impressionable literature student when one of my best teachers started his class with that sentence. I loved Borges' literature, its precise yet lyrical account of clockwork magic in the a bookish worlds, so I felt immediately attacked. I was ready to counter-argue, to defend Borges against this claim. However, Borges is a liar. A great author, too, don't get me wrong. I still admire his works, maybe a bit less than back then. And maybe liar is not the best description of the problem with Borges, but it does some of the provocation work needed to think about it. Just in case, let me oversimplify Borges' literary universe. His works are full of recursive engagements with the history of literature, like the famous Pierre Menard, Author of the Quixote, where he tells the story of, well, Pierre Menard, who writes a line-for-line identical book to Cervante's Don Quijote.
ChatGPT got this one right …
… but didn’t quite get the joke with this one.
This short story is a playful reflection on authorship, textual analysis, and the weight of the history of literature in everything we write (including newsletters). Of course, Borges argues, we cannot read Menard's Quijote without taking into consideration XX Century events, even if the text is the same as that written by Cervantes in 1605. Texts grow from texts, interpretations from interpretations. Borges's work also creates and comments on fictional literature, like in Tlön, Uqbar, Orbis Tertius, in which actual literary references are enmeshed with plausibly sounding yet totally imaginary texts that together form the plausible story of forgotten regions of the world, other planets, and Berkeley's idealism. For Borges, all written texts are possible, all books have been written and can be found in The Library of Babel (which is also online), because all literature could be understood to be the pleasing result of permutations. Finally, Borges' stories are obsessed with worlds of mirrors, with models that may contain and reflect a world, and with worlds that are part of, or contained by, different models. There are few true things in Borges' worlds, everything is fiction, or text, or a model or a reflection in a mirror. Borges' worlds are clever games of reflections, where there is a truth but it evades us as we try to catch it with the limited capacities of language1. For the sake of my argument, let me simplify even more: Borges lies because his texts are the skilful assemblage of other texts, real or not, put at the service of a replica or simulacrum of actual texts, which then are given meaning or value by the way they resonate with other, existing texts. Borges' texts are mirror halls of western, white literature. Borges does not write about the world, he writes about a model of the world that comes from books, from a universe of texts the preceded him. Generative AIs are liars. Their output is a recombination of a model of texts or images that inform them. These systems do not create, they do not compose: they recombine all the texts or images that were part of their datasets into texts or images that sound and look new but carry no history, no understanding, no contribution, no dialogue. The output of Generative AI is not an original text or image, is a Borgian mirror of a model of the world (Borges would probably be amused). Generative AI tells us nothing about the world, and everything about their model of the world, which is presented, and often interpreted, as the world itself. The model is not the territory, and the output is a well-crafted lie. Staying with Borges, it would be tempting to think of these Generative AI systems as equivalents to The Aleph, a mythical space/object from which everything can be seen. Generative AI systems have seen many things, they aspire to creating everything, every image, every text. But they are not reflections of the universe, they are distortions of the model built on data they grew from. If anything, Generative AI systems are prisms from which we can see internet culture: the aesthetics of Deviant Art, the prose of Reddit. They are an Aleph of digital mundanity. Actually, generative AI systems are not even liars, they are impersonating a liar. They are constructed to sound like a good liar would sound, like an excellent pretense. But they are not a distillation of the art and the images they have been trained on. To lie, they would need to understand the notion of truth. But these things don’t understand anything, they are just calculating probability and lying to us about being able to lie to us. ChatGPT has a model of what human writing can be, but it doesn’t know what writing is. They have a model, but they don't know the territory2.
Borges, for all his lying, is working with the territory. His mirrors reflect a particular understanding of the books that existed. His recursive models provide commentary on being human. Borges lies but he is a dialogue with the texts that preceded him, and that followed him. Generative AI systems are just "tattered Ruins of that Map, inhabited by Animals and Beggars”. (where is play? you may be asking. In the next newsletter, play will come back, pretending it has never left).
There is more to Borges, of course. He may be a liar, but he is not a sophist. There is a (romantic) thread in his stories that deals with loneliness, love, and a profound joyful and cruel fascination with people and the way they relate to others. This joy is often subdued in the presence of the importance of the text and Literature itself. So it’s on us to stop treating these systems as territories, as places where truth can happen. They are models with certain uses, excellent for playing, but confusing them for a territory is buying into their lies. |