Researchers at the HJ Institute of Technology and Management (HJITM) have published a study empirically analyzing which characteristics of generative AI heighten office workers’ “job replacement anxiety” in Technological Forecasting and Social Change (TFSC, SSCI), a leading international journal in the fields of technology management and forecasting. The study was conducted jointly with Professor Do-Hyung Park of the Graduate School of Business IT at Kookmin University, who served as corresponding author.

The researchers surveyed 346 office workers across various industries in Korea who use ChatGPT in their work and used partial least squares structural equation modeling (PLS-SEM) to analyze how four characteristics of generative AI — memorability, self-learning, anthropomorphism and personalization — relate to job replacement anxiety. They also examined how skepticism toward new technology changes these relationships.

The analysis showed that anthropomorphism (β=0.185), the degree to which AI speaks and responds like a person, and personalization (β=0.184), its delivery of responses tailored to individual users, both significantly increased job replacement anxiety. By contrast, the ability to remember conversation content and the self-learning ability to improve its own performance showed no significant effect in the final model, which accounted for other factors. This means replacement anxiety depends less on AI’s ability to remember and learn than on how humanlike it appears and how much it adapts to the individual.

Technology skepticism was itself the strongest driver of anxiety (β=0.333). The moderating effects are particularly notable. Higher skepticism strengthened the tendency for self-learning to lead to anxiety (β=0.153), but the direction was reversed for anthropomorphism (β=−0.238). Among people with low or moderate skepticism, anxiety rose as AI became more humanlike, whereas among highly skeptical people, anxiety tended to decrease somewhat.

Among the demographic variables, only gender was significant. Female respondents reported greater job replacement anxiety than male respondents, while differences by age, job rank, tenure and frequency of use were not significant. The final model explained 39.2% of the variance in job replacement anxiety.

The study is significant in empirically demonstrating that it is not the “performance” of generative AI but its “humanlikeness” and “personalization” that connect directly to workers’ perceptions of threat. It also suggests that functional training alone is not sufficient when companies adopt generative AI, and that communication and retraining approaches should be tailored to employees’ levels of technology skepticism and characteristics.

HJITM continues to conduct research published in international journals on the acceptance and use of generative AI, organizational performance and digital transformation.