It's difficult not to recognize a certain unique and "authentic" existence, something that philosophers refer to as an "aura," of the creative neural networks of previous years. This quality is not due to their remarkable abilities, but rather to the glaring weirdness, deficiencies, and problems that are inseparable from their personalities, often generating more interest and joy than their properly functioning aspects. The creepy motifs and weird failure modes of DALL-E and Midjourney, the "personality" of ChatGPT, or even more so, of the Bing Chat—nicknamed ChatBPD due to its distinct emotional (emojional?) and elusive style, impulsive reactions, even little tantrums—are the elements many of us were most captivated by when exploring those new technologies, and these are the aspects we might remember.
The concept of "aura" comes from Walter Benjamin's famous essay, "The Work of Art in the Age of Its Technological Reproducibility." He argues that the unique and authentic quality of artworks conflicts with the reproducibility that technology has uniquely enabled in our time. Encountering and making things possessing that auratic quality is a very human aspiration, so this problem, in one form or another, was among the biggest questions that 20th-century artists grappled with. Arguably, apart from the outright luddism and attempts to achieve authenticity purely stylistically, the most typical artistic strategy here would involve incorporating into the work the critique of its specific material conditions; as those conditions change with each reproduction in a new context, this critical essence staked its claim to authenticity and uniqueness, making it more resilient to the threat of technology. As powerful as such critique can be, relying on it as the sole means for art to possess an "aura" has long started to feel tiresome and dull.
Technology and aura are not in as direct conflict as it might seem from a superficial understanding of reproducibility. While technology, particularly in the late industrial age, appears fundamentally serial, it often allows and even requires the creation of unique objects; prototypes are essential both for technological progress in general and as a means to enable and define serial production. Of course, every prototype is designed so that it can be understood and replicated, but this potential reproducibility doesn't strip away its "aura." It’s not enough, of course, for a prototype to be unique to expect an aura about it, but the “reproducibility” aspect is not as damning as it might seem. The true irreproducibility, whether of technology or an artwork, is not a matter of fundamental limitation but, as Pierre Menard demonstrates, a purely cultural constraint. It’s not about a practical impossibility to repeat something (a truly reductive perception of what art is), but a social, economic, historic inadequacy, or inevitable displacement, of that repetition. Here this inadequacy is not achieved through a particular critical relationship with the act of creation, but rather through the excess, an expensive, and extravagant redundancy that is impossible to justify repeating.
No one will ever train another exact version of Bing Chat or Midjourney of a specific version, not because it's impossible, but simply because it's expensive and relatively pointless. If you have that much capital, you might as well invest it into creation of something new, to discover new bugs and failure modes instead of repeating those ones, no matter how loveable they are. These systems, with all their peculiarities, are unique and authentic because they exist, somewhat actively engaging with the world, but they won't be reproduced again. They're quite fragile: we enjoy their idiosyncrasies, but one day, likely soon, they'll be turned off and replaced by something "better," with higher quality and utility, perhaps systems so flawless that no one would even consider their personality or "aura." And we will miss those older AIs, and their quirks will remain a part of our cultural history, even if the systems themselves never run again.