Modern language models now have the capability to generate written text that approaches human-level writing, foreshadowing the eventual advent of Artificial Intelligence (AI) writing companions. However, while this new technology boasts increased productivity and support for writers, it also introduces the possibility of AI technology overstepping co-creative boundaries when it comes to control, autonomy and ownership over written work.
In the study, ‘From Tool to Companion: Storywriters Want AI Writers to Respect Their Personal Values and Writing Strategies,’ Oloff C. Biermann, Dr. Ning F. Ma, and Dr. Dongwook Yoon from UBC’s Department of Computer Science, explore the barriers to the adoption of human-AI co-writing, and the willingness of writers to collaborate with an AI companion; a co-writing agent which a human writer controls.
Most people are already familiar with AI-based writing tools that support grammar and style revisions, such as Grammarly or spellcheck in Microsoft Word. What differs with an AI companion is that it can directly translate the high-level ideas a writer has into fully developed text with minimal user input, with only a few examples. Large-scale models such as GPT-3 and Jurassic-1 are able to write natural language on a level which has led some to believe that AI technology has now passed the Turing test, creating endless potential for humans and AI to collaborate on writing tasks.
“In some interfaces, you can let [the AI companion] generate sentence by sentence, and then you decide where you’re going to take it. In other kinds of interfaces, you can just write a ‘seed’ idea in a sentence and it will realize the idea into a whole paragraph-long text just with a click,” said Yoon.
This new way of co-writing with an AI companion can bring in the question of psychological ownership over work, and writers’ desire for ownership was identified as a potential barrier to adoption. Biermann explained that writers often feel strongly about the work they produce and have a personal connection to their words, particularly in the case of storywriting.
“When you’re writing and are actively involved, it feels as though there’s ownership over it,” said Biermann. “But if someone else or some agent is doing the writing, then you can start to lose the sense that it’s your own work.”
Writing is a complex and dynamic task that requires the application of multiple skills being used simultaneously. The study found that some writers did not trust an AI companions’ capability to grasp the nuanced aspects of character development, general intuition and dialogue writing.
“Writing is not just writing. It involves a planning process, ideation, having the content in your mind as you write, and translating your ideas into words,” said Yoon. “Working with an AI writer is interesting because it’s really good at translating, but people don’t expect it to generate ideas by itself. Since it’s not a real human with real lived experience, it’s difficult to expect the AI to develop its own purpose in a creative endeavor without any human input.”
Additionally, the way that the AI companion is perceived as a support tool has a strong impact on its adoption by writers.
“Some [writers] just see it as software that can help make you a better writer,” said Biermann. “However, some go to the other extreme where [the AI companion] is seen like a collaborator that’s going too far and is overstepping the boundaries of what a co-author should be doing.”
The researchers found that those who write more as a hobby tend to appreciate greater control, as compared to writers who do so professionally. Professional writers tend to desire greater productivity and are willing to trade some level of control in exchange for the ability to focus more on high-level planning while an AI companion manages the translating.
“The value and the context of the writing matters,” said Yoon. “There are professional writers who are in the position to prioritize prolific writing over meaningful writing practices who see the value in productivity. However, it doesn’t make sense for a hobbyist writer, who finds joy in the writing process, to adopt an AI companion and let it do the writing for them.”
History has shown that technological advances can lead to human labor being substituted in some shape or form, however, this doesn’t mean advancing technology will make certain professions or human-led tasks obsolete.
“Technologies are developed to substitute human labor in some sense,” said Yoon. “For example, steam engines replaced human muscle to some extent, and same for computers supporting accountants. The definitions of these jobs have changed as technology develops.”
While there are endless possibilities for advancement and development in the world of AI technology, researchers want to remind the AI community not to let the people behind the technology, or those interacting with the technology, be forgotten.
“I think the whole community of academic researchers and application developers, within and across the domain of AI, needs to acknowledge the value of the users, rather than pushing themselves to develop better and better technology, without thinking of the ramifications of what it means by being human when we are using and interacting with technology,” said Yoon. “Those developing AI methods and AI applications should work to make the future more inclusive and better for humans, not just the business and the systems.”
The paper received an honorable mention award for best paper at the Designing Interactive Systems (DIS) 2022 conference. To read the full study, click here.
Written by Kelsea Franzke