Artificial intelligence self-efficacy: Scale development and validation

Abstract

With the development of artificial intelligence (AI) applications, it has become critical for scholars, educators and practitioners to understand an individual’s perceived self-efficacy regarding the use of AI technologies/products. Understanding users’ subsequent behaviors toward the advancement of AI technology is also critical. Despite the growing focus on AI, a suitable scale for measuring AI self-efficacy (AISE) has yet to be developed. Current scales for measuring AISE (i.e., technology self-efficacy scales) are considered inapplicable because they neglect to evaluate perceptions of specific AI characteristics (e.g., AI-based configuration or anthropomorphic design). Given the limitations of existing self-evaluation and diagnostic instruments, the aim of this research is to investigate the construct of AISE, and develop and validate an AISE scale (AISES) for measuring an individual’s perceived self-efficacy in regard to the use of AI technologies/products, in accordance with established exploratory and confirmatory scale development procedures. Specifically, a literature review is employed to generate initial items. An exploratory factor analysis is then performed for item purification purposes. At this stage, potential elements of AISE are extracted. Subsequently, factor extraction and confirmatory factor analysis are used to verify the construct structure of AISE. An analysis of 314 responses indicates that the AISE construct contains four factors: assistance, anthropomorphic interaction, comfort with AI, and technological skills. The scale is comprised of 22 items, and is found to have good fit, reliability, convergent validity, discriminant validity, content validity, and criterion-related validity. Moreover, nomological validity is built by the positive correlation between the AISE construct and motivated learning behaviors. This paper is the pioneer in developing and validating a scale to measure AISE. The findings extend existing knowledge of AISE and can help scholars further develop AISE theories. Our findings will also help educators and practitioners assess individuals’ AISE and explore related behaviors.