GenAI Boosts Productivity but Divides Users by AI Interaction Skill

A recent randomized controlled study indicates that generative AI (GenAI) can enhance task performance among early-career knowledge workers, though the benefits are not uniformly distributed. The research, available on arXiv (2605.18143), involved participants self-studying a technical subject with either conventional resources or support from large-language models (LLMs). While, on average, access to GenAI improved performance, these enhancements were not correlated with GPA or existing knowledge. Instead, AI Interaction Competence (AIC)—the skill to extract, filter, and validate model outputs—emerged as a crucial factor. Participants with high AIC experienced significant improvements, whereas those with low AIC faced minimal or negative returns. An intervention using conceptual maps helped reduce performance variability, emphasizing the need for standardized workflows to bridge the gap. The results underscore the significance of human-AI collaboration in education and knowledge-based tasks.