Validation of the Short General Attitudes Towards Artificial Intelligence Scale: The Short GAAIS-10

Abstract

With Artificial Intelligence becoming ever more widespread, it is important to have instruments that measure people’s attitudes toward Artificial Intelligence efficiently. Providing such a tool was the aim of the current report. Its authors report the validation of a shortened and further purified version of General Attitudes toward Artificial Intelligence Scale (GAAIS) consisting of ten items (Short GAAIS-10) based on empirical survey data from UK-based participants, collected online (total N = 1406). The multi-phase study design was based on Factor Analysis, Item Response Theory, correlation, and multivariate multiple regression. Phase 1 was a rigorous selection phase based on three independent prior samples of approximately 300 participants each (collected in 2021 and 2022). Confirmatory Factor Analysis (CFA) and Polytomous Rasch Analysis (PTA) alongside semantic judgements were used to select 10 items. In phase 2, a new representative UK sample (N = 500) was drawn in 2024 to further validate the Short GAAIS-10. CFA and PTA of the new data revealed good psychometric properties of the Short GAAIS-10, which is bidimensional with two subscales (Positive, Negative). The Short GAAIS-10 showed predictive validity against the Technology Readiness Index based on correlation analysis. Rated comfortableness with seven types of AI applications was positively predicted by the Short GAAIS-10 subscales based on multivariate multiple regression. The Short GAAIS-10 is a valid and streamlined instrument with which to measure General Attitudes toward Artificial Intelligence. The ambivalent nature of public AI attitudes is discussed as an important observation for a general AI attitude scale to capture. The contribution of this research is the creation of a short version of a valid and efficient psychometric instrument with which to measure general attitudes toward Artificial Intelligence.