An experimental study of various ways of working with AI on a creative task finds that preserving human involvement in early creative decisions can help organizations balance efficiency, quality, and satisfaction. AI is fast and efficient, but research suggests that using generative AI can reduce the diversity of creative outputs and sap worker morale. Kartik Hosanagar and Daehwan Ahn asked 285 students in an elective course on emerging technologies to write a 1,000-word story. The next class period, the same students used ChatGPT 3.5 to write a story, using one of three approaches. In the "Human Confirmation" condition, AI did all the work—coming up with ideas, outlining, drafting, and editing—and the humans merely approved the outputs. In the "Human Creativity" condition, humans generated ideas, selected a story concept, outlined the story, and then AI drafted and edited the story. In the "Copilot" condition, participants could use AI support at any stage of the creative process.
The stories were written 36.2% more quickly with AI help. When read by human readers, stories written with the Human Confirmation approach were rated as lower quality than human-only stories, but both Human Creativity and Copilot stories were rated as higher quality than human-only stories. Weaker writers showed larger gains in story quality between Day 1 and Day 2. The Human Creativity and Copilot groups also reported higher levels of satisfaction than the Human Confirmation group and the human-alone condition.
In post-study reflections, participants wrote appreciatively of the efficiency of writing with AI, the benefits of the tool for idea generation, and for its ability to add editorial polish. However, they also noted the AI stories lacked emotional depth, originality, and nuance, and were unable to maintain a consistent tone and plot. Participants in Human Confirmation expressed less pride in the final story than participants in Human Creativity and Copilot combined. According to the authors, collaboration design affects trade-offs among productivity, creative diversity, and human satisfaction in AI-assisted work.
