How You Collaborate With AI on Creative Work Makes All the Difference

Did you use AI at your job today? Maybe it helped you work faster. But too much AI may reduce both the quality of your work and your satisfaction with the job, according to a study published Nov. 10 in the PNAS Nexus science journal. 

The team behind the study knew from prior research that generative AI was a mixed bag, capable of dramatically boosting productivity, but also prone to throttling creative diversity and worker satisfaction.

“The question we wanted to answer was whether these trade-offs are inevitable,” Kartik Hosanagar, a University of Pennsylvania professor of technology, digital business and marketing, and one of the study’s authors, told CNET via email. “Our research suggests they aren’t.”

More from CNET

AI is changing work quickly, with dizzying contradictions. Tech companies like Microsoft and Meta are spending millions on integrating AI into their business for greater efficiencies and output, while getting rid of human workers through mass buyouts and layoffs. Workers who use AI report some advantages, like saving 40 to 60 minutes a day at work, while suffering consequences like loss of confidence, loss of skills and a tougher labor market. 

Teachers have also sounded the alarm, saying AI is essentially ending homework. Researchers say that if we don’t pay attention to how we collaborate with AI, we could start losing critical thinking along the way. 

“Think about GPS navigation. It’s incredibly useful, but many of us have lost the ability to navigate even familiar neighborhoods without it,” Hosanagar said. “Something similar could happen with AI and cognitive skills.”

The study by Hosanagar’s team focused on the different ways we collaborate with AI on creative tasks. They found that the way we use AI — not just whether or not we use it — has a big effect on the final product and on our perception of the work.

“The future of work isn’t just about how capable AI becomes,” Hosanagar said, “but how we design the collaboration between humans and AI so as to support a broader set of goals.” 

Be careful what you automate with AI

The researchers tasked 285 students, split into three groups, with writing assignments over two days. On the first day, no students used AI for writing. On the second day, each group used a different process for writing with AI. 

Each of the groups used a separate “collaboration design” when working with the AI on the writing assignment. The collaboration designs represent different ways people interact with AI on the job. All three approaches delivered similar productivity benefits, but differed significantly in quality, satisfaction and creative diversity, according to Hosanagar.

In the first group, Human Confirmation, the AI planned, outlined, wrote and edited the assignment, with the humans just approving the work during each step. In the second, Human Creativity, humans did the outlining and planning, while AI did the writing and editing. In the third, Copilot, humans and AI collaborated during the entire process. 

The researchers found that using AI in creative writing did not improve the quality of the work in any of the groups, but it did speed up the task. 

People in the Human Confirmation produced the lowest quality and least diverse work, and participants reported feeling unsatisfied with it. But the other two groups, Human Creativity and Copilot, produced better work at the same speed, with participants reporting a more positive experience. 

The study showed that the best outcomes happen when humans are actively involved in creative tasks, especially generating ideas and creating outlines, while AI assists with execution, Hosanagar said. “The goal shouldn’t be to eliminate human effort, but to direct that effort toward the parts of the process where human judgment and creativity matter most.”

It would be good to be wary, too, of the unintended consequences of people playing a smaller role in their daily work as they collaborate with AI. 

“Our research also suggests that people derive satisfaction from having a meaningful creative role, not simply from producing a good final output,” Hosanagar said. “So, we should be thoughtful about what we automate.”

Writer

Julian is a contributor and former staff writer at CNET. See full bio