Description
H.U.M.A.N. is a poetry upscaling, sound sculpture which utilizes two copper hot water tanks with sound exciters, found wooden cabinet, amplifier, computer, dataset, Markov Chain + LLM models for increasing resolution and text2voice.
Exhibited in 'Mirror, Mirror'
Victoria, British Columbia. October, 2024
This 3D Model will be sent to collectors of the first 28 poems produced by H.U.M.A.N. which were presented at the opening of Mirror, Mirror.
Model scanned on November 7th, 2024.
🪞🪞
H.U.M.A.N. explores the evolution of AI technologies by using two different models of "artificial intelligence": an older Markov Chain and a modern Large Language Model (LLM) to reveal the gaps between technological generations. The Markov chain, a statistical method that was popular in the early days of AI research (dating back to the 1950s and 60s), predicts the next word in a sentence based on the previous word. It works with limited data and often creates fragmented, disjointed text due to its small scope of context. In contrast, today’s LLMs are much more advanced, processing vast amounts of data to generate coherent and structured language. They use neural networks to understand patterns in language, making them capable of producing text that reads more smoothly and naturally. Through these two networks, H.U.M.A.N. explores how AI, much like humans, can transform and upscale fragmented information, turning raw data into something more meaningful and comprehensible. The initial dataset for the Markov chain consists of thousands of song lyrics from the artist’s past memories. These lyrics are chopped, rearranged, and reassembled by the Markov chain into new, often disjointed poems. This process mirrors how early AI worked, creating outputs that were sometimes hard to follow due to limited data and computational power.
The first stage of H.U.M.A.N. involves the Markov chain generating a poem from the dataset of song lyrics. The output is a rough, somewhat incoherent poem, typical of early AI. This poem is then read aloud in the gallery through the sculpture via an early text-to-speech model that reflects the era when Markov chains were in use. The spoken poem comes through a sculptural speaker made from an old copper hot water tank. The hot water tank, which has a silhouette that vaguely resembles a humanoid robot, symbolizes the way AI has been anthropomorphized—or given human characteristics—to make it more familiar and easier for people to accept. However, this is a thinly veiled, crude attempt to make an unfamiliar, almost alien technology seem more relatable.
In the second stage of this installation, the output from the Markov chain is fed into a modern ‘Large Language Model’ (LLM). The LLM, unlike the Markov chain, is able to interpret and produce text on a deeper level as it operates on vast datasets which provide the rules on how language is engineered. It takes the jumble of words and phrases from the Markov chain and upscales them, creating a new, more comprehensible poem in the form of rhyming couplets. This transformation reflects how AI today can reconstruct meaning by increasing the “resolution” of ideas—much like how an image can become sharper and more detailed with higher resolution.
This process is not just about clarification but also about reinterpretation. The LLM is reconstructing the past in a new form, one that aligns with current technological capabilities. The result is a poem that is easier for humans to understand, but it's also a reshaped version of the original—a new narrative created through AI's lens. In this way, the artwork highlights how both AI and humans take disjointed experiences or ideas and weave them together into something clearer, giving them new meaning.
However, this upscaling process comes with risks. As AI fills in the gaps of fragmented knowledge, it can distort our understanding of the original information. While increasing the megapixels in a camera makes an image sharper, the new AI-generated clarity is a reconstruction, not a perfect retelling. The result may obscure the original, creating a new version of history based on the biases and limitations of the AI itself.
H.U.M.A.N. cautions against relying too heavily on AI to interpret and retell our history. With each generation of AI, as we compress and upscale information, we risk losing the full picture of the past. Without a deep understanding of how AI reshapes and reinterprets data, we may find ourselves following generations of A.I. models without looking back to gauge our initial direction. If unable to trace back the fragmented breadcrumb trail we may lose the true path back to the original story.