OBJECTIVE

With the rapid development and widespread adoption of generative artificial intelligence (GenAI) technologies, their unique characteristics-such as conversational capabilities, creative intelligence, and continuous evolution-have posed challenges for traditional technology acceptance models (TAMs) in adequately explaining user adoption intentions. To better understand the key factors influencing users' acceptance of GenAI, this study extends the AIDUA model by incorporating system compatibility, technology transparency, and human-computer interaction perception. These variables are introduced to systematically explore the determinants of users' intention to adopt GenAI. Furthermore, the study examines the varying mechanisms of influence across different user groups and application scenarios, providing theoretical insights and practical guidance for optimizing and promoting GenAI technologies.

METHODS

During the data collection phase, this study employed a survey method to measure behavioral intentions and other key variables within the proposed framework. The survey design included demographic information about the respondents as well as detailed information related to their use of GenAI. In the data processing and analysis phase, a Structural Equation Modeling (SEM) approach was utilized to systematically examine the path relationships among the variables. Additionally, to compare the differences in variable relationships across different subgroups, a multi-group structural equation modeling(MGSEM) analysis was conducted.

RESULTS

(1) Effects on Key Expectations: Social influence significantly enhances performance expectancy (β = 0.109, < 0.05) but negatively impacts effort expectancy (β = -0.135, < 0.01). Hedonic motivation notably mitigates effort expectancy (β = -0.460, < 0.001), yet shows no significant effect on performance expectancy (β = 0.396, = 0.76). The newly extended variables-technological transparency (β = 0.428, < 0.001), system compatibility (β = 0.394, < 0.001), and human-computer interaction perception (β = 0.326, < 0.001)-demonstrate positive influences on performance expectancy while generally mitigating effort expectancy. (2) Emotional Mechanisms: Performance expectancy significantly mitigates negative emotions (β = -0.446, < 0.01), while effort expectancy significantly increases negative emotions (β = 0.493, < 0.001). Negative emotions exert a significant negative influence on usage intention (β = -0.256, < 0.001). (3) The MGSEM analysis revealed significant heterogeneity in the extended AIDUA model paths across different user segments. Specifically, systematic variations were observed across demographic characteristics (gender, age, and educational level), occupational backgrounds, and usage patterns (task types and AI tool preferences). These findings underscore the heterogeneous nature of generative AI acceptance mechanisms across diverse user populations and usage contexts.

DISCUSSION

This study reveals several key findings within the extended AIDUA model. Our results indicate that technological transparency emerges as the strongest predictor of performance expectancy, alongside system compatibility and human-computer interaction perception, significantly enhancing users' perceived system performance. Regarding effort expectancy, hedonic motivation and technological transparency demonstrate the most prominent effects, implying that system design should emphasize user experience enjoyability and transparency. Notably, the lack of significant influence of hedonic motivation on performance expectancy, contradicting our initial hypothesis. Furthermore, the MGSEM analysis reveals significant heterogeneity in acceptance mechanisms across user groups, providing crucial implications for the differentiated design of GenAI systems tailored to diverse user needs.