Attitudinal Profiles of Artificial Intelligence Among Students

Background and Purpose: Artificial intelligence (AI) is increasingly integrated into higher education, yet students’ attitudes toward AI remain complex and not fully understood. Prior research often treats attitudes unidimensionally, focusing either on perceived benefits or concerns. This study conceptualizes AI attitudes as multidimensional, examining how students simultaneously evaluate benefits and risks and how these evaluations cluster into distinct attitudinal profiles.

Methods: Data were drawn from a multi-university survey of social science students (N ≈ 382). Two composite variables were constructed to represent perceived benefits (e.g., usefulness, productivity, societal contributions) and perceived concerns (e.g., bias, ethical misuse, privacy, misinformation). Median splits were applied to each composite to generate a four-category typology: Optimistic (high benefits, low concerns), Ambivalent (high benefits, high concerns), Concerned/Resistant (low benefits, high concerns), and Disengaged (low benefits, low concerns). Multinomial logistic regression analyses were conducted to examine factors associated with profile, using a Concerned/Resistant group as the reference category.

Results: Students were distributed across four distinct AI attitude profiles: Optimistic (27.78%), Ambivalent (14.90%), Concerned/Resistant (32.83%), and Disengaged (24.49%), with Concerned/Resistant representing the largest group. Multinomial logistic regression indicated overall model significance, LR χ²(15) = 111.76, p = .0001, pseudo R² = .108 (N = 382). Frequency of AI use was the most consistent and statistically significant predictor of profile membership. Higher AI use was associated with a greater likelihood of belonging to the Optimistic (β = 2.62, SE = 0.35, p < .001), Ambivalent (β = 2.71, SE = 0.40, p < .001), and Disengaged (β = 1.18, SE = 0.34, p = .001) groups relative to the Concerned/Resistant group. In contrast, self-reported AI knowledge did not significantly differentiate profiles, and exposure to AI-related training and institutional guidelines was also not significant. Mental health indicators were largely unrelated to profile membership, although higher depressive symptoms were associated with a lower likelihood of being Disengaged relative to Concerned/Resistant (β = -0.11, p = .009).

Conclusions and Implications: Findings suggest that students’ attitudes toward AI are not unidimensional but reflect distinct configurations of perceived benefits and concerns. The presence of sizable Optimistic, Ambivalent, Concerned/Resistant, and Disengaged groups suggests that students are actively negotiating both the opportunities and risks associated with AI, rather than uniformly accepting or rejecting it. Importantly, frequency of AI use emerged as a more consistent predictor of attitudinal profiles than either self-reported knowledge or exposure to formal training and institutional guidelines. This pattern suggests that students’ orientations toward AI may be shaped more through direct, experience-based engagement than through structured educational inputs. The lack of association between institutional exposure and attitudes raises questions about the current role of higher education in guiding students’ ethical and critical engagement with AI. For social science education, these findings point to the need for approaches that move beyond awareness-based instruction and toward integrating applied, context-specific discussions of AI use, including opportunities for reflection on both its benefits and risks within professional practice.