Creative Work with AI: The Human Role Alters Outcomes

Key Takeaways

1A study of 285 students shows that AI accelerates writing without improving its quality

2Quality, diversity, and satisfaction depend on the role given to humans in the process

3Delegating all the work to AI degrades these criteria, while maintaining control over ideation and structure improves them

Why it matters : Designing human-AI collaboration helps avoid a loss of creativity and satisfaction, unlike total automation.

An experiment published in PNAS Nexus concludes that AI accelerates writing without improving its quality. The results vary mainly depending on the level of human involvement: delegating most tasks to the machine degrades quality, diversity, and satisfaction, while maintaining control over ideas and structure yields better outcomes.

Keeping Control Improves Experience and Deliverables

Participants who retained a central role in the design and execution of the text received better feedback. The Human Creativity and Copilot configurations produced work deemed superior at a comparable speed and were associated with a more positive experience. In contrast, the Human Confirmation scheme, where the AI drives almost everything and the human merely validates, resulted in the lowest quality and least diversity, accompanied by a sense of dissatisfaction. Overall, AI primarily served as an accelerator without enhancing quality, and productivity gains remained similar across approaches. Participants also reported deriving satisfaction from a significant creative role, beyond just the final outcome.

Three Modes of Collaboration Tested on 285 Students

285 students participated in two days of writing. None used AI on the first day. On the second day, each group followed a distinct collaboration design with AI. In Human Confirmation, the machine planned, sketched, wrote, and edited, with the human only approving at each stage. In Human Creativity, participants handled planning and sketching while delegating writing and editing to the AI. In the Copilot mode, humans and AI worked together throughout the process.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

The General Framework: Productivity, but Risks for Diversity and Skills

Previous work by the same team had already indicated that generative AI can boost productivity while compressing creative diversity and satisfaction. In practice, users report a gain of 40 to 60 minutes per day, but also mention a loss of confidence, erosion of skills, and a tougher job market. Meanwhile, companies like Microsoft and Meta are investing significant amounts to integrate AI, in a context described as marked by waves of layoffs. In education, teachers are warning about the de facto end of homework, while researchers caution that poorly framed collaboration with AI could undermine critical thinking.

Recommendations from the Team: Design Collaboration, Don’t Automate Everything

According to Kartik Hosanagar, the future of work will hinge on how human-machine cooperation is designed. The best results emerge when humans generate ideas and plans, and AI mainly assists in execution. The goal is not to eliminate human effort but to focus it where judgment and creativity matter. He warns against the unintended effects of reducing individual roles and calls for careful selection of what should be automated. He compares AI to GPS, useful but potentially detrimental to skills when overly relied upon. Published in PNAS Nexus, the study precisely measured the impact of collaboration modalities, confirming that the way AI is used affects outcomes. The authors argue that compromises once seen as inevitable are not, and they remind us of Hosanagar's identity and role as a co-author.