Using artificial intelligence tools at work can speed up tasks, but excessive reliance might reduce both work quality and job satisfaction, as reported in a study published Nov. 10 in the PNAS Nexus science journal. Researchers examined whether trade-offs between productivity and creative diversity are inevitable when utilizing generative AI.
The Impact of Collaboration Design
Kartik Hosanagar, a University of Pennsylvania professor and one of the study’s authors, told CNET via email that their research suggests these trade-offs are not inevitable. The study investigated the rapid changes brought by AI in the workplace, where companies integrate technology for efficiency while employees report saving time alongside concerns about losing skills and confidence. Teachers have also warned that AI could diminish critical thinking and cognitive skills, comparing the shift to how GPS navigation has reduced human wayfinding abilities in familiar neighborhoods.
To understand these dynamics, the research team assigned writing tasks to 285 students divided into three groups over two days, with no AI used on the first day. On the second day, each group applied a different collaboration design to interact with the AI during the writing process. While all three methods provided similar productivity boosts, they produced markedly different results in quality, worker satisfaction, and creative diversity, according to Hosanagar.
Finding the Right Balance for Human Effort
In the first approach, named Human Confirmation, the AI handled planning, outlining, writing, and editing while humans merely approved each step. The second approach, Human Creativity, had humans perform the planning and outlining while AI executed the writing and editing. The third approach, Copilot, involved continuous collaboration between humans and AI throughout the entire process.
The study found that AI assistance did not improve writing quality across any group, though it accelerated the task. Participants in the Human Confirmation group created the lowest quality, least diverse work and expressed dissatisfaction, whereas the Human Creativity and Copilot groups achieved better work at the same speed with more positive feedback. Hosanagar stated that optimal results occur when humans remain actively engaged in idea generation and outlining while AI supports execution, emphasizing that human effort should be directed toward judgment and creativity rather than eliminated.
