Large Language Model-Based Artificial Intelligence as a Relational Leader? Effects of Leadership Style and Stress on Artificial Intelligence Leadership

Abstract

This study provides an initial investigation into the objective performance, subjective perceptions, and neural correlates of relational leadership behaviors demonstrated by large language model (LLM)-powered artificial intelligence (AI) leaders in the workplace. A 2 × 2 × 2 mixed-design experiment was conducted, with leadership style (relational versus non-relational) and supervisor type (human versus AI) as between-subjects variables, and stress level (high versus low) as a within-subjects variable. Eighty-two university students performed simulated automotive parts inspections under either human or AI supervision. The results showed that LLM-based AI leaders could effectively simulate relational leadership behaviors, enhancing workplace well-being, self-perceptions, and leadership evaluations. Most subjective perceptions were similar under human and AI supervision, with a few measures favoring human leaders. While high stress negatively affected subordinates’ subjective perceptions, relational leadership helped buffer these effects. A functional near-infrared spectroscopy (fNIRS) scanner was used to assess prefrontal cortex activity, and results revealed that activation in brain areas associated with task switching and emotional regulation was linked to AI leadership, relational leadership behavior, and stress levels. These findings suggest that AI-driven relational leadership offers benefits in organizational settings, but careful design is needed to mitigate stress-related challenges. The study provides initial empirical evidence supporting the potential of LLMs as relational AI leaders while emphasizing the importance of deliberate design to promote workplace well-being.