Empowerment or Deskilling: A Study on the Knowledge Gap Concerning the Effects of Large Language Models Dependence

Large Language Models (LLMs) are increasingly mediating individuals’ cognitive learning and capability development processes, sparking ongoing debates over whether they contribute to empowerment or deskilling. Despite the growing interest, prior research remains inconclusive and has largely overlooked how dependence on LLMs impacts differ across social groups, especially from the perspective of the knowledge gap hypothesis. Based on the frameworks of the knowledge gap and digital divide, this study examines how LLMs dependence influences different dimensions of human capabilities, and how socioeconomic status (SES)—especially educational level and LLMs understanding—moderates these effects. An online survey (N = 1,180) was conducted. The results show that the impact of LLMs dependence is both dual and context-sensitive. Specifically, it is negatively associated with cognitive ability, skill ability, and social ability, but positively associated with creative ability. In addition, higher levels of education and a better understanding of LLMs appear to moderate these effects by mitigating the risk of skill decline. These results indicate that human capabilities are shaped not only by technological factors but also by personal agency and forms of social capital. This study highlights the importance of promoting autonomous learning and cognitive resilience, offering insights from a communication perspective on how individuals and societies can respond to the evolving challenges presented by LLMs and related technologies.