Determinants of the Adoption of Generative Artificial Intelligence in Academic Research: A Structural Equation Model from Colombian Universities

Abstract

This study examined the determinants of both the intention to use and the self-reported use of generative artificial intelligence (GAI) in academic research by means of a partial least squares structural equation model (PLS-SEM), collecting data from 167 research-active faculty members at three higher education institutions in southwestern Colombia. The findings show that perceived ease of use is the strongest predictor of usage intention, followed by social influence and perceived benefits, whereas social influence emerges as the primary predictor of perceived benefits. Contrary to the conventional reading that associates resource deficits with non-adoption, individual resource barriers are positively associated with perceived benefits, supporting a compensatory logic: the greater the constraints of time and knowledge, the higher the value attributed to a tool capable of mitigating them. Concerns about AI output quality exhibit an asymmetric pattern, as they are unrelated to perceived benefits yet are associated with lower usage intention, suggesting that these concerns operate as a behavioral brake without eroding the recognized utility. The main contribution lies in identifying this dissociation between the recognition of utility and the willingness to act accordingly. It is concluded that GAI is valued as a useful tool for compensating research-related deficits, yet it coexists with concerns about the quality of its outputs that must be addressed by institutions through training, communities of practice, and clear guidelines on verification and responsible use.