Niko A. Grupen, a doctoral candidate in the CS department advised by Bart Selman and Daniel Lee, won a student competition, and as a result his work is included in a new book—The Future of Text (2020)—edited by Frode Alexander Hegland.
Grupen's contribution "From Author to Editor: Our Place Alongside the New Life Cycle of Text" is among the concluding entries in this multi-author collection. In a section entitled "The New Life Cycle of Text," Grupen writes:
Historically, text has followed a pattern of conception, dissemination, and transmission that mimics, roughly, the biological life cycle. Consider, for example, the work of Thomas Pynchon, whose nuanced appraisals of American culture (in V. and The Crying Lot of 49) garnered critical acclaim in the 1960’s before fertilizing the imagination of author David Foster Wallace, amongst others, decades later. Pynchon’s own ideas were influenced by earlier pieces, including Jack Kerouac’s On the Road and Saul Bellow’s The Adventures of Augie March. The same process can be found in text of all sorts, from academic publications—Newton’s “standing on the shoulders of giants” quote sums it up well for the scientific community—to viral online blog posts.
Like many of the media we humans use to describe our sensory experiences, text’s life cycle is transforming radically. Gone are the days of mortal text, in which passages pass transiently through time, with only a select few surviving to inspire the next generation of writers. In the Digital Age, everything sticks. Every piece of textual content produced on the Internet—or salvaged from the Old World of printed text— has been granted permanence and, in turn, the right to consume disk space on a server for all of eternity. Consequently, today’s networked society has enabled generations of authors, researchers, and armchair experts to contribute simultaneously to the largest stockpile of knowledge in human history.
The natural question to ask is: who benefits most from eternal text? And no-one can fault you for asking that question. But the correct question to ask is not who, but what benefits most from eternal text? Surprisingly, or unsurprisingly depending on who you ask, the principal beneficiary of this information treasure trove is not the next generation of human writers, but rather, the next generation of artificial intelligence. AI researchers have spent an extraordinary amount of time, effort, and money scraping every last token of text from the Web with the goal of training massive neural network models to convincingly mimic the statistical properties of human language. The craziest part is, they have largely succeeded!
We have already witnessed demonstrations in which an AI scribe seamlessly generates paragraphs of coherent text, given only a broad theme and a prompt to elaborate upon. Such feats are not limited to written text in essay form—competing demonstrations have shown that a single model (OpenAI’s GPT-3, to be specific)
can generate poetry, screenplays, song lyrics, recipes, financial statements, Excel spreadsheets, LaTeX formatting, and even HTML/JSX code for website layouts. Though there are still more questions about this technology than there are answers, we are not far from a world in which AI moves out of the research lab and into the newsroom, the design studio, and the engineering scrum. In fact, it seems that at least one contributing author has already embraced this future, allowing an AI model of the sort I have described to write portions of text for this book.
Continue reading Grupen's chapter and the rest of The Future of Text