Recent research suggests that using artificial intelligence (AI) to challenge, rather than simply complete, work tasks can lead to better outcomes for knowledge workers. Experts caution that overreliance on generative AI risks diminishing critical thinking skills and job performance, a phenomenon known as "cognitive offloading," where individuals outsource too much mental effort to AI tools.

Monideepa Tarafdar, a professor at the University of Massachusetts (Amherst), highlights the importance of creating intentional "friction" in human-AI interactions to foster deeper analysis and innovation. She argues that instead of asking AI to produce straightforward answers or content, users should prompt AI systems to critique or find flaws in their ideas. This approach can counteract the “echo chamber” effect often seen with AI, where models generate responses shaped by previous user inputs, potentially reinforcing existing biases and blind spots.

In practical examples from her research, a marketing professional tasked AI with creating virtual customer profiles that disliked the product being marketed. This negative feedback led to unexpected product improvement ideas previously unconsidered by the design team. Similarly, an attorney asked AI to identify obscure legal loopholes and explore potential exploitations. The AI’s responses included realistic, unforeseen scenarios that highlighted vulnerabilities in workplace behavior.

Tarafdar’s findings are supported by broader studies on human-AI collaboration. One study on loan evaluations demonstrated that combined human and AI efforts outperformed work conducted by either humans or AI independently. Notably, collaboration involving disagreement between human and AI partners yielded the most effective results. Another investigation in creative writing showed that authors producing drafts interactively with AI as a sounding board created superior work compared to those using AI to ghostwrite content alone.

To maximize the benefits of AI in professional settings, Tarafdar recommends that organizational leaders encourage employees to incorporate “friction-generating queries” into their workflows. Drawing on personal knowledge and expertise to question AI outputs prompts the system to generate more nuanced and diverse perspectives. Although this process can feel as though the AI is resisting, it ultimately enhances the quality and reliability of the results.

These insights emphasize that thoughtful engagement with AI, rather than passive reliance, is critical to maintaining cognitive skills and achieving high-quality outcomes in work projects.