Make it new: The challenge for authors in a ChatGPT world

As with graphic design software, AI can be a tool to overcome expectations, not provide readymade solutions. (Image by Tara Winstead via Pexels)

When advertising and design agencies first adopted personal computers as creativity tools, there were some who were less than pleased. Designers like Neville Brody felt that this would lead to a flattening and homogenisation of work. There would be less customisation and more blandness because of the ease of creating designs that would pass muster.

Design software did lead to a sea of uninteresting layouts. And yet, many talented designers used it to create work that was revolutionary. Brody himself pushed boundaries by combining organic and inorganic forms, breaking up typefaces, and layering patterns and colours in unusual ways. His layouts for The Face and Arena magazines in the 1980s are still inspirational.

FF Typeface Six and Seven, by Neville Brody. (Photo by Andrew Nash via Wikimedia Commons 2.0)

The introduction of ChatGPT is likely to have the same effect on fiction, in positive and negative ways. For now, though, generative AI chatbots are still a work in progress. Look at the eccentric conversations between New York Times technology writer Kevin Roose and Bing’s AI-powered version, for example, as well as the embarrassingly incorrect responses of Google Bard.

These are teething troubles, and there is a case to be made that Silicon Valley companies have been too hasty in latching on to what they see as the Next Big Thing. Chances are that erratic behaviour will be sorted out in time, but not before several more unexpected results.

Reams have been written about how generative AI will affect education, content marketing, coding, and more. Meanwhile, enterprising authors are already employing it to help them with works in progress. Jennifer Lepp, who writes “paranormal cosy mystery” stories, uses Sudowrite, a tool built on the GPT-3 model, to churn out plots and descriptions. Others are using ChatGPT’s answers to inspire their work in various ways.

The effects are already being felt. Earlier this month, the science fiction magazine Clarkesworld had to shut down their submissions because they were inundated by AI-driven stories. Unfortunately, the editor tweeted, most of these were from those outside the SF/F community, driven by “side hustle” experts making claims of easy money with ChatGPT.

For author and law professor Stephen L. Carter, ChatGPT “gets the elements of fiction but isn’t sure how to arrange them”. This leads to plot-driven, page-turning formulas, Carter concludes, with weak endings and a decidedly inconsistent tone of voice.

ChatGPT will likely evolve to become a collaborative tool for writers, who will use specific prompts for their needs. The problem, however, lies in the way it is designed in the first place.

Ask ChatGPT how it works, and you will be told that it is trained on a large amount of text data to learn language patterns and structures. It uses this to come up with responses to user inputs. Where does the text data come from? The answer: websites, Wikipedia, open domain books, and social media.

In other words, this language model draws from a pool of already existing content. It generates text based on statistical patterns, not intuitive understanding. ChatGPT is incapable of being original.

In a recent piece, speculative fiction writer Ted Chiang uses the analogy of photocopier art to make the distinction clear. “Starting with a blurry copy of unoriginal work isn’t a good way to create original work,” he explains. It’s in the very process of writing, he says, that writers discover original ideas.

The young Ezra Pound would have agreed. “Make it new,” he famously proclaimed, which became the rallying cry for Modernism in the early decades of the twentieth century. The effects of mass industrialisation and the fallout of World War I led to work that was experimental, individual and broke with conventional structure and form.

Many of these works remain profoundly influential. James Joyce’s Ulysses and Virginia Woolf’s Mrs Dalloway are the usual suspects, but there are many others over the years, from Djuna Barnes’s Nightwood to Robert Musil’s The Man Without Qualities to Jean Rhys’s Good Morning, Midnight. These novels are deeply subjective, sometimes fragmentary, and structurally innovative. They take the material of the past and combine it in fresh new ways that artificial intelligence can’t.

Also read: Notion's generative AI model that can refine writing skills available to public

When we start to drown in a tidal wave of work based on ChatGPT’s derivative directives, the novels that stand out will be those that transcend them. As with graphic design software, AI can be a tool to overcome expectations, not provide readymade solutions.

Add to that rampant climate change, inequality and conflict, and one can only repeat the words of Samuel Beckett. “To find a form that accommodates the mess, that is the task of the artist now.”