Using Large Language Models to Understand Leadership Perception and Expectation

Using multimodal approaches for measuring propositions in natural language (the Fill-Mask Association Test, FMAT) and ChatGPT 4.0, based on the Five-Factor Model of personality traits, we examined how leaders are perceived and expected in natural language. Leaders/employers (vs. workers/employees) are perceived as less agreeable and more extraverted, yet expected to be more conscientious and emotionally stable. Compared to leadership perceptions, leadership expectations are relatively higher for agreeableness, conscientiousness, and emotional stability, but lower for extraversion and equal for intellect/openness. Our findings provide new insights into implicit leadership theories (leadership prototypes) in natural language, with meaningful discrepancies between the perception and expectation of leadership.