Type: article
Abstract: Biases are central to human decision-making, and their role grows more relevant as Generative AI enters judgment-intensive domains. This pilot study examines the halo effect in human participants and large language models (LLMs), testing whether AI reproduces or diverges from human judgment. We used a fictional, abstract persona – a Star Wars commander – to isolate impression formation from the accountability pressures of judging real individuals, such as job candidates. Assigning the character to a fleet (Imperial vs Rebel) served as a proxy for perception (villain vs hero), complemented by open-ended responses. No clear halo effect emerged in the binary measure, yet human and LLM decisions differed significantly, and qualitative findings revealed subtle bias patterns not captured quantitatively. LLMs appeared less prone to halo effects but reasoned differently, suggesting AI’s susceptibility may hinge on the evaluation’s stakes and realism – a boundary condition future HR research should test.
Source: AIS Electronic Library (AISeL)
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7222992562
