Vanessa Chang is an independent curator; a program manager at Leonardo, the International Society for the Arts, Sciences and Technology; and a lecturer in visual and critical studies at California College of the Arts.
In July, the artificial intelligence research company OpenAI released its latest language generator, GPT-3, to great fanfare. GPT-3, or Generative Pre-trained Transformer 3, uses deep learning software and neural networks to create text that is legible to humans. Across the internet, people rushed to create poems, plays and other literary forms with it, feeding it everything from Shakespeare to Dr. Seuss. GPT-3 gorged on literary data.
As is usually the case with new developments in artificial intelligence, pundits and the public rushed to assess the language model’s creative sophistication. Despite its impressive linguistic fluency, the program ultimately came up short: It is unable to truly comprehend its own prose, so it veers into incoherence the longer it writes.
A long road remains until AI might pose an existential threat to creative writers. But it has already begun to serve as a creative writing tool. And like all writing tools, it challenges the notion that the skull marks the border of the human mind.
Writing systems of all kinds have reified the intangible; collecting ideas into archives, they are the building blocks of external memories that can turn thinking into an interchange between mind and database. As prosthetic memories, the earliest forms of writing can be understood as ancestors of the information retrieval and analysis that characterize contemporary machine learning. AI offers new ways of working with databases — new ways of thinking and creating.
By extending humans’ cognitive capacities, writing helped to sustain profound cultural transformations. AI may yet do the same. But as the uneven legacies of literacy suggest, the stories we tell with our writing tools are just as critical to cultural change as those tools themselves.
The Database And The Extended Mind
In 2017, the artist and self-styled “gonzo data scientist” Ross Goodwin drove a Cadillac from New York to New Orleans with a computer loaded with AI software that was trained on three literary corpora including science fiction, poetry and “bleak” literature, as well as Foursquare location data. He had coded the algorithms and curated an archive of hundreds of books to contour the AI’s linguistic matrix and aesthetic sensibility. And with Google’s support, he rigged the car with a microphone, clock, GPS unit and a rooftop camera.
As Goodwin piloted the car south, the system of neural networks synthesized information from these sensory inputs and generated writing in a poetic idiom evocative of Jack Kerouac’s spare stream-of-consciousness prose. Along the way, data from the sensors inspired sometimes lyrical, sometimes surreal sentences of sights and sounds from the road.
One algorithmic recognition of a Foursquare location produced “Eagles Nest Diner: a American restaurant in Goldsborough or the Marine Station, a place of fish seemed to be a man who has been assembled for three days.” The AI’s words were printed on long rolls of receipt paper, alongside timestamps of when they were generated, ultimately producing “1 the Road,” which Goodwin claims is “the longest novel in the English language.”
“1 the Road,” at once text and performance, highlights the complexities around assigning authorship when artificial intelligence is involved in the creative process. The final product was written by algorithms that Goodwin composed. The AI achieved its prose style — from word choice to syntax — from dynamic interactions with its datasets, mediated by Goodwin’s code, hardware choices and driving. Performing a 21st-century version of the buddy road trip, Ross and his AI enacted a vision of creative writing as both collaborative and distributed. This continuing interplay of author, algorithm, environment and database — a memory far larger than the human mind can easily access — belies the complex system that brought it to life.
“Where does the mind stop and the rest of the world begin?” This question opened “The Extended Mind,” a 1998 paper by the philosophers of mind Andy Clark and David Chalmers, perhaps the most well-known and influential articulation of the extended mind thesis. Their view, which has become increasingly popular since the dawn of the millennium, rejects the traditional understanding that the mind emerges only from the physical processes of the brain and stops at skull and skin. Rather, the extended mind thesis frames cognition as an interplay of these processes and the tools we use to complete cognitive tasks.
Such tools, like the pencil and paper a mathematician uses to solve equations, can become so seamlessly integrated into our thinking that they functionally bring about our cognitive abilities as much as our brains do. Through such tools, our minds extend into our worlds — and our worlds into our minds. Cognition, therefore, emerges from an ecology of brain, body and world.
By embodying information outside of the human mind, writing systems — from handwriting to some contemporary AIs — install new elements into this dynamic ecology. Granthika, a new storytelling software start-up spearheaded by the novelist Vikram Chandra, puts this understanding of extended cognition into practice. Granthika is an intelligent system and integrated writing environment imagined as a writer’s assistant and bookkeeper — an external brain. Among other tools in development, the software helps fiction writers build and track complex worlds and timelines. It takes care of the grunt work so writers can focus on narrative elements like themes, plot and character.
According to Chandra, Granthika merges text and semantics; writing in the program generates knowledge that is then reflexively integrated into the text. By continually adding to this knowledge base, Granthika builds an ontology of the novel’s universe. It can use classical first-order reasoning to construct a behind-the-scenes ontology of time; it can forge connections between people, places, things and events that are meaningful, unlike other databases. A partner in imaginative continuity, the software aims to free writers for the ephemeral work of craft.
By delegating the work of worldbuilding to a computational intelligence, Granthika serves as a cognitive extender for creative writers. It incarnates a symbiotic relationship between the creative writer and the intelligent database. “1 the Road,” similarly, distributed the labor of writing across human and machine actors, although Goodwin flipped the script and positioned his AI as the thinking and sensing center of the process.
Against traditional notions of the lone author, these projects conceive of creative writing as the product of humans coupled with technologies. But rather than a radical disruption of writing history, these machine intelligences enact the next logical step in the evolution of writing as a mnemonic technology.
Memory And Literacy
In an era where keyboards and touchscreens mediate much of our communication, earlier forms like handwriting emanate an almost natural aura. As organic as handwriting may seem, it too is a technology that wrought profound transformations on communication, cognition and culture.
Before the invention of writing, the spoken word lived in the evanescent spaces between people, evaporating as soon as they were spoken. Writing captures fugitive speech in flight and reifies it in a visible and enduring form. Where once a conversation had to take place between two living speakers, in writing, words can exist beyond the presence and lifetime of the scribe. For these reasons, Plato famously condemned writing in “Phaedrus,” saying it would erode the human capacity to remember.
From grocery lists to encyclopedias, writing extends the human mind by offloading cognitive processes of information storage and retrieval. Writing is a technology that allows us to outsource individual and collective memory. By sustaining the creation of informational archives that can be referenced, literacy made possible new forms of interaction with language — new techniques of information storage afforded the structured accumulation of knowledge. Literacy, therefore, is a central pillar in the the systematic study of logic, philosophy and science in general — the knowledge infrastructures that ultimately yielded AI.
Beyond these general aspects, scholars of literacy and orality — that is, cultures with no knowledge whatsoever of writing — disagree about how writing transforms modes of thought as well as modes of communication. Some, notably Walter Ong and others in the 1960s and 70s, made generalized claims about how oral cultures use particular linguistic and mnemonic strategies for cultural transmission. Referring to famous literary remnants of oral texts such as the Bible and the Odyssey, in his influential book “Orality and Literacy” Ong related the syntax of language to the structure of thought. In so doing, he excavated language itself as a site for mnemonic practices.
In the decades since, this view has been challenged for its oversimplification of “primitive” oral cultures and literacy’s totalizing effects. Even as they have affirmed the impact of writing and reading on thinking, critics noted that these effects were too complex and localized to assert specific cultural characteristics. Due to wide variations in the nature and use of writing systems, scholars describe a universe of multiple literacies, attuned to the particular cultures from whence they emerge.
Technologies Of Storytelling
The writer Ted Chiang explores the relationship between multiple literacies, memory and subjectivity in his short speculative-fiction story, “The Truth of Fact, the Truth of Feeling.” The story tells two parallel tales — one in the not-so-distant past, and one in the near future. Both are about writing and recall technologies, and how they are entwined with knowing and feeling.
In the first narrative, an older journalist relates his experience of researching and using a new technology called Remem. Remem is a form of artificial memory that allows users to index and search their lifelogs — video recordings of their daily lives. The second narrative is about Jijingi, a young man from the Tiv oral culture in Nigeria, who is taught to read and write by a missionary. As his literacy develops through his studied consumption of colonial and religious texts, Jijingi struggles to reconcile his oral culture with the written word.
Throughout the story, readers are privy to both the journalist and Jijingi’s mounting curiosity, confusion and anguish as these new literacies begin to destabilize the certitude of their own beliefs. Both narratives orbit around the notion of truth; as the story’s title suggests, truth has both factual and emotional dimensions. Memory is never simple recall, nor is its value measured by accuracy.
From the beginning, Chiang constructs clear analogs between the distinct literacies of writing and new media. The story opens with the journalist reflecting on the nature of his daughter’s digital literacy. Rather than using a pen or keyboard, “Nicole subvocalizes, her retinal projector displays the words in her field of vision, and she makes revisions using a combination of gestures and eye movements. For all practical purposes, she can write.”
But without the assistive software, the journalist notes, Nicole would have trouble spelling out the words on her own. “English becomes a bit like a second language to her,” the journalist reports, “one that she can speak fluently but only barely write.” Although Nicole can produce and decipher English, their lingua franca, for the narrator the distinct writing technologies they use signify a much larger gulf between them.
The journalist’s investigation of Remem interlaces with his reflections on his troubled relationship with his daughter. In his own flawed memory, he had worked to repair their poor relationship and remains a good parent, unfairly held at a distance. Despite his suspicions, the journalist begins to use Remem to explore his own memories, a task that requires him to reach out to his network to use their own lifelogs to plug the gaps in his own Luddite database. As the technology knits this archive together, he discovers the shattering truth that he’s been lying to himself about his parenting.
Jijingi’s tale is similarly fraught. His cultural clash comes to a head when he turns to written European records to resolve a disagreement in the tribal courts, pitting oral history against written archive. He returns to his clan with evidence refuting their claims, but his elder shrugs and asks, “Have you studied paper so much that you’ve forgotten what it is to be Tiv?”
Jijingi soon comes to understand that the factual truth encoded in writing is precise, but ultimately not in service to his community. Jijingi reflects that the practice of writing had begun to shape his very thoughts in the form of the European culture it contained; in so doing, it had also begun to distance him from the priorities and beliefs of his own.
These truths — the factual and the emotional — form a study in contrasts that grounds the story’s cultural expressions of literacy and memory. At the story’s end, the journalist reveals that he had constructed Jijingi’s narrative based loosely on historical elements, imbuing it with an affective richness he had imagined and brought into being through storytelling, in order to underscore the limits of artificial memory.
The journalist characterizes both himself and Jijingi as “cognitive cyborgs,” their minds extended by writing and recall technologies old and new. New technologies, and the new literacies they beget, leave both loss and gain in the wake of transformation. The rendition of Tiv oral storytelling affirms the vibrant presence of the human body — “he told you the story with his whole body, and you understood it the same way.” Although writing enabled Jijingi to examine the past at a distance, it was bereft of this embodied plenitude and its living connection to his culture.
“The Truth of Fact, the Truth of Feeling” explores how technologies contour the stories we tell — to ourselves and to each other. At the same time, it emphasizes how the stories we tell also shape the technologies we use to tell them. In the journalist’s telling, “people are made of stories. Our memories are not the impartial accumulation of every second we’ve lived; they’re the narrative that we assembled out of selected moments.” Chiang ultimately forecasts the limits of these synthetic forms of recall, and the necessity of creative and cultural practices in enlivening these cognitive extenders.
By recording information outside of the human mind, writing systems spawned new ways of sorting and retrieving knowledge. AI and the advent of big data has amplified and automated that reading capacity far beyond the scale of human perception.
While Remem remains a speculative technology, its augmentation of archival search and indexing evokes those operations in machine learning. As external and increasingly intelligent forms of memory, AI can deepen the effects writing systems had in earlier times on the physical borders of the human mind.
Yet, as “The Truth of Fact, the Truth of Feeling” affirms, writing and reading machines extend thinking beyond the material edges of these tools. As part of a cognitive ecology distributed across brain, body and environment, these technologies are also embedded within social and cultural worlds. New tools breed new literacies, which can engender nascent forms of knowing, feeling and telling.
Early writing systems scaffolded the emergence of new modes of creativity and communication. So too might AI. The stories — and truths — we tell with and about AI can help to tune and integrate these technologies in service of human expression. And AI might also one day be a partner in writing stories as yet unimagined.