Can AI Predict Individual Values and Attitudes? an Experimental Comparison

The growing impact of artificial intelligence (AI) on decision-making processes has sparked a lively debate in the social sciences, focusing on the integration of social and cultural competencies into machine behavior (Aragona et al., 2023). This discourse delves into the challenges and implications of enabling AI systems to comprehend and replicate the nuances of human values, ethics, and cultural contexts(Floridi & Cowls, 2022). Our study positions itself within this debate, exploring the capacity of AI to predict individual orientations regarding values and attitudes based exclusively on socio-demographic information.

Through an experiment that compares the responses provided by AI on attitude scales with data collected from questionnaires administered to real individuals, we aim to verify whether AI can identify with the reference social category and accurately reproduce its patterns of values and opinions. To accomplish this we compare two distinct datasets. The first dataset is composed of approximately one thousand responses generated by an AI language model, which was given initial social categories to simulate the responses of individuals belonging to specific socio-demographic groups. The second dataset consists of real responses collected through questionnaires administered to actual individuals. Matching between the two datasets is performed based on socio-demographic information, allowing a direct comparison between the AI's responses and those of humans.

Without predefined expected results, we find ourselves at an interpretative crossroads: if a significant alignment emerges between the AI's responses and those of human beings, this may indicate that AI possesses a predictive capacity that could—provocatively—open up the possibility of using it in sociological practice. Conversely, a misalignment between the datasets could be equally interesting, as such discrepancies might stem from intrinsic biases in the AI model or in the data collection methods, raising critical questions about the validity and reliability of AI technologies in complex human contexts.