Knowing Your Creativity, Part 2: Speaking Finnegan’s Wake
Researchers trained an AI exclusively on Joyce’s novel. The objective was to map the inner realm where, perhaps, literary creativity happens.
In my previous article on creativity, I discussed how intellectual restlessness drives artists to experiment, challenge common sensibilities, and advance art forms. Often, historic breakthroughs come when creatives find ways to express things previously considered inexpressible. In music, think of how avant-garde composer John Cage used oddly “prepared” musical instruments and even silence, in one famous composition, to broaden the boundaries of musical expression. You might also consider how some forms of rock music and jazz were regarded as noise when first introduced. (Some still are.) It’s all about a) inventing new languages of expression which, in turn, b) redefine art.
In literature, there was possibly no greater boundary buster than James Joyce, who among other innovations helped advance stream of consciousness writing to the forefront of modern fiction, most particularly in his novel Ulysses. Joyce’s most ambitious work was Finnegan’s Wake. For lovers of literary fiction, Finnegan’s Wake is Mount Everest — something to one day perhaps scale, but never conquer. In writing the book, Joyce used English as a sort of base color on a palette on which he dabbed a host of other languages to develop, through blending and alteration, his own signature idiom.
Joyce, you might say, found English inadequate for expressing the expansive range of cultural and historic references, narrative elements, sensual stimuli, and linked associations that fill the human mind (or his, at least). I believe the ability to trace those vectors from one idea to another is essential to creativity. All great writers seem to have a facility for identifying affinities between things most would never consider to be related, which is why analogy, simile, metaphor, and allegory are fundamental to expressive writing. Great writers can perhaps trace those vectors in the mind between like things more easily than your average Joe.
Joyce took it upon himself to go to extreme lengths in this endeavor, virtually inventing a new language form to suit his purposes. To do so, he mined the space where thoughts take shape, attempting to govern that internal realm where language develops and resides in all humans.
I can’t claim to have had much luck in tackling Finnegan’s Wake, which is why I was excited to learn about a research project that employed Joyce’s inscrutable opus to train an AI model. The aim, however, was not an interpretation of the novel; it was an effort to better understand that very realm of ineffability that Joyce explored, what the researchers term a “latent space,” where thoughts, language, and creativity come into being.
Nina Beguš is a postdoctoral researcher at U Cal Berkeley who, along with Gašper Beguš, studies an emerging field which she refers to as “artificial humanities,” work that “sits at the intersection of fiction, the humanities, and emerging technologies.” Along with other collaborators, they recently published a scientific paper about their latent space experiment that is better explained for us non-scientists in an interview she did with Harvard, intriguingly entitled, “What Finnegans Wake Teaches Us about AI.”
Beguš defines a latent space within an AI model as “an inner geometry that establishes itself during the training process. This geometry represents abstract relations; it maps similar or related things together and keeps unrelated things further apart.” Latent spaces, if I get it right, are analogous to those conceptual realms within human minds where we connect the dots between related things, finding associations that spark creative thinking, a process that Joyce exploited in the extreme.
The goal of Beguš’s Finnegan’s Wake experiment was creating an architectural view of this latent space and, more specifically “peeking into the internal layers of how speech is produced.”
Mapping the processes of an established AI model, like ChatGPT or Anthropic’s Claude, would be far too complicated because those LLMs are trained on massive amounts of data and the internal structures are inestimably complex. Instead, the researchers used a strictly limited and controllable AI model called a GAN. With GANs, the researchers can start with a contained data set. Nina Beguš calls GANs “artificial babies” that start with a clean slate and learn human speech much in the way human infants do.
“We’ve been taken aback by the similarities, including the same kinds of errors,” Beguš said in her Harvard interview. “They start from zero, go through a ‘babbling stage,’ and finally produce a word or a ‘nonce word’: a term that doesn’t yet exist but is plausible, or is used only once.”
Beguš’s interest in nonce words is not coincidental. Finnegan’s Wake is rife with them. For example, Joyce invented the word “venissoon,” assumed to be a conflation of “venison,” “Vanessa,” and “very soon.” Shakespeare enjoyed spinning out nonces, as well. He is thought to have invented the words “alligator,” “gossip,” “zany,” and “obscene.”
Joyce’s wildly inventive use of language impressed the researchers as a fitting test for their “artificial baby” GAN. In explaining why, Beguš said, Finnegan’s Wake “foregrounds the limits of interpretability for speech that isn’t quite externalized.” In other words, I believe they liked the idea of using a rare form of language that would exist uniquely within the GAN’s latent space.
The researchers educated their baby, which they called “FinneGAN,” solely on the audio version of Joyce’s novel. They used a second GAN to act “like an ‘adult’ to the childlike GAN,” to help make sense of FinneGAN’s utterances.
FinneGAN’s universe was completely dependent on Joyce’s peculiar language. But according to Beguš, the “FinneGAN model pushed the novel even further into entropy,” meaning further into disorder and randomness than Joyce.
“It’s almost like speech play,” she said. “We wanted to show how one can intentionally deviate from the rules of ‘proper’ language to create novel or even illogical words.”
The science journalist Michael Pollan, whose new book, A World Appears, explores the mystery of consciousness, says scientists can learn a lot about that subject by studying how novelists invent characters and settings, thereby creating a signature form of reality to suit the purpose of their narrative. Beguš appears to recognize this ability in novelists to toy with consciousness and, by extension reality, as well.
“Literature can pick up yet-unarticulated realities,” she told Harvard. “Joyce was writing 100 years ago, and he was likely picking up on a new way of looking at reality, similar to the birth of quantum physics.” (FYI, Joyce invented the word “quark” for Finnegan’s Wake.)
As with most new advancements in AI, I am equal parts intrigued and terrified by the progress researchers are making in mapping out this cognitive territory in which creative inspiration is birthed. To artists and writers who struggle to put images and words to the inexpressible, this realm feels familiar, and also sacred. I’m resigned to the inevitability that generative AI apps will compose music, write fiction, and churn out visual art with superhuman proficiency. What happens when AI developers — and the AIs themselves — have the architectural blueprints they need to dial up the creative inspiration level to any degree they like?
Today, artists, poets, novelists, filmmakers, philosophers, scientists, and thinkers in all fields still hold out hope we’ll still be needed in the future to take those intuitive leaps that got us “alligator,” Cubism, and quantum physics, but it’s hard to deny that AI will soon be perfectly capable of giving us whatever “yet-unarticulated realities” we crave with a simple prompt. Scientific advancement like this is unavoidable, and it begs the question: What will be left for us?
Rick’s latest novel, Once a Man, was released in February, 2026. It’s about a teenager who discovers he’s part of a plan to shape humanity’s relationship with superintelligent AI.
You can explore Rick’s artwork on his website…