In 2012, computer science researchers from the University of Toronto revisited an old Artifical Intelligence (AI) concept—neural networks—and rejuvenated it with the ‘AlexNet' architecture. This image recognition model leveraged a layered approach: input data passed through deep convolutional neural networks (CNN), each layer reassessing and refining the information.
By combining this hierarchical processing with a significant boost in computational power and data availability, AlexNet shattered records in image recognition tasks to date, with a top-5 error rate of 15.3% and an almost 40% reduction in overall error rate. This layered hierarchical assessment and assignment strategy marked the resurgence of neural networks as a dominant paradigm in AI research.
The approach was later crucial in the development of the attention architecture introduced in the 2017 paper ‘Attention is All You Need’, the title a reference to the song 'All You Need Is Love' by the Beatles. The paper paved the way for the emergence of generative pre-trained transformers (GPTs) - those Generative AI (Gen AI) chatbots that have profoundly impacted various domains—work, politics, media, art, and society at large—over the last few years, achieving remarkable success in intelligence benchmarks and becoming an increasingly significant part of our digital interactions.
Across the Atlantic around the same time, Mark Fisher—known for his k-punk alias and blog of the same name—was compiling his latest book 'Ghosts of My Life’, a title borrowed from a track off a 1993 EP of the same name by jungle act Rufige Kru. Emerging in the 90s, the British electronic music style jungle captured a dark, frenetic, and turbulent atmosphere, evoking the soundscapes of dystopic science-fiction. Think Blade Runner, Terminator, Predator—a vivid fusion of noir, decaying time stretched sonic landscapes, and a predatory futurism perpetually threatening the horizon.
On jungle, Fisher wrote "For about five years, between 1992 and 1997, jungle sounded like the future rushing in. During that period, jungle was involved in a process of ceaseless reinvention, dismantling and reassembling itself – and assumptions about sound – before our amazed ears.”
In the essay, Fisher recalls his experience on first listening to Rufige Kru's ‘Ghosts of my Life', and recognising its timestretched sampling of 'Ghosts’ by 1980s new romantic band Japan. "Time had folded in on itself. One of my earliest pop fixations had returned, vindicated, in an unexpected context” he wrote.
Continuing on this theme and almost as its counterpoint, Fisher's 'The Slow Cancellation of the Future' interrogates the temporal dislocation of contemporary culture, particularly in music. The essay laments how the future has been indefinitely deferred, caught in what Fisher calls the “relentless gravitational pull” of a recycled cultural past.
Fisher observes, “The ‘jumbling up of time,’ the montaging of earlier eras, has ceased to be remarkable; it is now so prevalent that it is no longer even noticed.” This temporal flattening, he suggests, has eroded our ability to cultivate new moments or subcultural shifts. Instead, we exist within a recursive loop where past forms are perpetually remixed causing paralysis and a foreclosing of the future. The result is a culture of nostalgia, endlessly cycling through its own echoes, where the future has become little more than a faint, ghostly afterimage of what has already been.
One example Fisher cited was the Blur vs. Oasis rivalry and the Britpop explosion of the 90s, which evoked a romanticised version of a bygone Beatlemania era—time mythologised as more 'authentic' or culturally rich. It’s poignant to reflect on this in 2024 as Blur and Oasis announce a string of sellout reunion shows and world tours. We’re still riding a train to future nostalgia, back on 74, constantly rebooting convolutionally identified and stratified cultural artifacts and memories. The stagnation of the increasingly dynamic-priced box office, exclusively fixated on established properties; a reluctance to risk and a dependence on the pastiche of past.
Now back to AI. For the uninitiated and at the very simplest, GPTs like ChatGPT and its competitors are essentially predictive text algorithms, which output the next token in a chain of words based on word probabilities from training data. What Stephen Wolfram calls a "reasonable continuation of whatever text it’s got so far, where by 'reasonable' we mean “what one might expect someone to write after seeing what people have written on billions of webpages, etc."
Earlier models drew from a training library comprising approximately 570 gigabytes of filtered text data, equating to around 300 billion words, predominantly sourced from the Common Crawl, a vast collection of information aggregated from millions of websites. Spanning hundreds of billions of words, the training data offers a comprehensive, though temporally limited, snapshot of human knowledge. It is an enormous, eclectic chrestomathy, a modern Borgesian Digi-Library of Babel, brimming with both profundity, practicality, triviality and memes.
This reflects a blind paradox: while the model is powered by unprecedented data breadth, it is also perpetuating a vicious cycle of linguistic repetition—a perpetual anachronistic stasis of language...if the history of language were from web pages linked from Reddit articles with more than three upvotes.
Large language models (LLMs) transform, through predictive parameters and vast compute, the past knowledge, reflections, and inane comment trolls of contemporary culture qua the world wide web. It forms an intricate yet impassioned pattern—an elaborate pastiche, a sophisticated stochastic parrot for what has come before. While the technology’s capacity for generating humanlike language is remarkable, it ultimately falls short of creating anything genuinely novel. Instead, it mimics, reassembles, remixes, rehashes and reprocesses the content it has absorbed in a determined structure with a ’temperature’ setting of predetermined randomness as a coagulant.
Fisher was acutely aware of this dynamic even before his untimely death and the emergence of Gen AI. In a YouTube lecture discussing his essay, he remarked, 'what the internet provides is an oppressive weight of the past,’ as if it had already begun to collapse under its own content. The emergence and exploitation of the contemporary commons of the internet neatly coincides with when the future definitively disappeared.
This is the latent contradiction of Gen AI: innovation that is rooted in repetitive patterns, leading to a potentially more pronounced form of future stasis. With Gen AI, the internet becomes further flooded with generated synthetic content—text, music, visuals—that feels eerily familiar, as though we've seen it before, as if it could be real, but might also just be ‘AI'. It's not so much an uncanny valley as it is a vast canyon—an overwhelming, planet-sized fugue that traps us in a perpetual loop. In the work of art in the age of metasemantic production, we find ourselves tethered to past artefacts as the means of production, struggling to break free and envision a truly novel future.
AI is profoundly reshaping our cultural landscape, but the questions it raises about originality, authenticity, and the repetition of past forms are not so much profound as they are perfunctory. The interplay between AI's capabilities and the frenetic cultural theories and histories that have come before may help better understand impacts and potentially guide improvements. Instead of resigning ourselves to an increasingly dead internet, there is an urge for theoretical frameworks and cultural awareness to steer the development of AI.
Fisher viewed the emergence of jungle in the 90s UK rave scene as a moment of radical cultural originality. Similarly, AI holds the promise of unprecedented innovation but exists within a paradox: while it can and will efficiently streamline the administrivia of everyday life, it must also aspire to create a truly reimagined future not yet pre-rehearsed. By recognising and assimilating patterns of cultural and social repetition, AI has the potential not only to mirror these rhythms but to transcend their stasis.
Words: Vicious
Illustrations: Conrad Armstrong
Comments