The purpose of this study was to conceptualize university students’ behaviors in critically evaluating and using AI-generated information in learning contexts, and to develop and validate a reliable and valid scale to measure these behaviors. Genera...
The purpose of this study was to conceptualize university students’ behaviors in critically evaluating and using AI-generated information in learning contexts, and to develop and validate a reliable and valid scale to measure these behaviors. Generative AI has become an important learning tool in university students’ information seeking, academic task performance, and knowledge construction. However, AI-generated information may involve limitations such as hallucinations, bias, and opacity of sources. Accordingly, it is necessary to measure students’ behaviors in not merely accepting AI-generated information, but in checking its accuracy and reliability, reconstructing it in accordance with their learning context, reflecting on and regulating the overall process of use, and complying with academic ethics.
To this end, this study operationally defined university students’ critical evaluation behaviors toward AI-generated information and developed preliminary items based on a review of prior research. Subsequently, a two-round Delphi survey was conducted with five experts in related fields to examine content validity, through which four components and preliminary items were finalized. A pilot survey was then conducted with 125 university students to examine item response distributions, reliability, item comprehensibility, and exploratory factor structure. In the main survey, data from 516 university students were analyzed using confirmatory factor analysis. The final items were determined by comprehensively considering standardized factor loadings, modification indices, inter-factor correlations, model fit, and the theoretical validity of item content.
The results of the study were as follows. First, the scale developed in this study consisted of four factors: “Information Checking,” “Critical Knowledge Construction,” “Information Use Monitoring,” and “Ethical Compliance.” Second, the confirmatory factor analysis of the main survey data supported a final four-factor, 17-item model after excluding Items 15, 8, and 1. In addition, the inter-factor correlations derived from the confirmatory factor analysis suggested that the subfactors were theoretically distinguishable, despite relatively high correlations among some factors. Third, the internal consistency reliability of the final scale was found to be acceptable for both the overall scale and its subfactors. Fourth, the overall mean score of the final scale was significantly and positively correlated with overall AI literacy, the critical evaluation and ethical use domains of AI literacy, and the logical and critical thinking domain of university students’ core competencies. These findings provide evidence of convergent validity by demonstrating that the scale is associated with theoretically relevant external criterion variables in the expected direction. At the same time, the correlations suggest that although the scale is related to existing constructs such as AI literacy and logical and critical thinking, it may not be completely redundant with them.
This study is significant in that it developed an instrument for measuring the process of evaluating AI-generated information based on learners’ actual behaviors in university learning environments where the use of generative AI has become increasingly common. In particular, beyond existing approaches centered on AI literacy, critical thinking, and ethical awareness, this study proposed a behavior-based scale that encompasses Information Checking, Critical Knowledge Construction, Information Use Monitoring, and Ethical Compliance to AI-generated information. Therefore, while the scale is related to AI literacy and general critical thinking competencies, it can be used as a distinct measurement tool to specifically diagnose how university students evaluate and use AI-generated information in learning contexts. The scale may serve as a foundation for diagnosing university students’ behaviors in using AI-generated information, designing educational programs that foster critical thinking and responsible AI use in the era of generative AI, and measuring changes in students’ critical evaluation behaviors before and after participation in such programs.