Is it possible for AI tools to match what professional human writers can do on their own? Just as importantly, are storytellers even open to human AI-co-writing?
That’s what UBC computer science researchers explored in a qualitative study unveiled at the Designing Interactive Systems 2022 Conference and published this week. Master’s student Oloff Biermann, postdoctoral fellow Ning Ma, and assistant professor Dr. Dongwook Yoon decided to find out what writers want from AI tools.
“This study is about values, especially human values when working with AI,” Dr. Yoon said. “It’s not about being replaced by AI. When we consider if AI can write better and more productively than humans, it’s scary to envision what that would look like. But if you look at other historical inventions, like the invention of the steam engine and how it changed the nature of physical labour, or how computers changed what it means to be an accountant in the 1980s, these jobs have kept going but in a different nature.”
So what do writers want from AI?
The researchers conducted a design workbook study with seven hobbyists and 13 professional writers. Both groups have somewhat different needs, and the study uncovered varying barriers to the adoption of human-AI co-writing.
“We learned about their personal values and motivation for writing,” said Biermann. “The reason why they write really influences the degree to which they want AI to assist them.”
“Hobbyists don't really want AI to take over their writing,” Oloff said. “They mostly want AI tools to act only like an assistant for them. But the professional writers who want to be productive in their writing welcome AI to do the writing for them in order to help them make more money.”
Hobbyist writers are less willing to give up their control to AI. They appreciate ownership over their storytelling, because it provides a feeling of integrity and enjoyment. Considering that writing has many different nuances and stages, the researchers wondered if hobbyists would be willing to hand over some parts of the process to AI in order to be more productive.
In terms of the writing process, the researchers considered AI support for three well-known writing stages identified by Flower and Hayes in A Cognitive Process Theory of Writing:
- Planning/ideation (what you are going to write)
- Translation (putting ideas into words)
- Revision (editing)
“We found that none of the writers wanted to give away the task of ideation because that means they’re not bringing a part of themselves as a writer into the process,” Dr. Yoon explained.
Using AI for some parts, not all parts, of storytelling
In the translation stage, AI is very capable of translating (writing) text from ideas, but hobbyist writers indicated they wanted to retain the “labour of love” involved in writing. Whether it’s experiencing a torturous writer’s block, or pulling their hair out at 2:00 a.m. while struggling to find the right words, they don’t want to miss this elemental part of the process.
Several professional writers in the study indicated they prefer to do the planning stage themselves, but would be happy to hand the translation stage over to AI as a more productive writing strategy. The researchers also discovered that the professional writers they interviewed want to retain editorial control with the final say on the end product.
“They [professional writers] actually assign a lot of emotional value to their vocation because they care about maintaining integrity,” Oloff said.
Adaptable program design is critical
“These results tell us that when designing AI tools for writers, it’s not just about designing for maximum productivity. It’s not as simple as that,” Oloff explained. “Designers need to ensure there is room for adaptation to the personal values that writers hold, not just productivity.”
Oloff admitted that before the study he thought writers would want productivity above all else. “The most surprising thing about the results is that, even in the case of some of the professional writers, they seem very resistant to AI-induced productivity across the board.”
“As indicated by this study, it tells us AI should ideally behave like a considerate human colleague, and similar to when a writer works with a ghost writer or a writing assistant,” Yoon added. “Those individuals adjust to the writer’s needs and honour their value as the primary author. So it’s almost like an anthropomorphic relationship, where the role of AI is adaptable and can personally adjust to the needs of the writer.”