Decoding symmetric and asymmetric pathways in generative AI learning adoption: a multi-method study

Synthesis: This multi-method study of 842 students in Lesotho's low-resource higher education system integrates UTAUT3 and Keller's ARCS motivation model to explain GenAI adoption. Using PLS-SEM, fsQCA, and importance-performance mapping, the authors find that cognitive beliefs (performance expectancy, effort expectancy, social influence, hedonic motivation, habit) explain intention, while ARCS motivational dimensions are stronger determinants of actual use. Multiple equifinal configurational pathways lead to high adoption, with motivation, enjoyment, and habit as core conditions. The study advances a hybrid logic of adoption — cognitive beliefs enable acceptance, motivational experiences sustain engagement, and habit normalizes use — and highlights the contextual sensitivity of acceptance theory in resource-constrained settings.

Core Finding

GenAI adoption follows a hybrid logic: cognitive-instrumental beliefs enable acceptance and intention formation, while motivational-affective experiences sustain actual use, and habitual interaction normalizes continued engagement. Both symmetric (net-effects) and asymmetric (configurational) pathways are needed to explain adoption, and this logic holds differently in low-resource contexts than in the well-resourced settings where most adoption theory was developed.

Dual-Layer Model: Acceptance vs. Engagement

The integrated framework posits a two-layer explanation. UTAUT3 constructs — performance expectancy, effort expectancy, social influence, hedonic motivation, and habit — predict behavioral intention, capturing the cognitive-instrumental side. Keller's ARCS motivation model (Attention, Relevance, Confidence, Satisfaction) predicts actual use and sustained engagement. This addresses the limitation of dominant adoption theories that focus primarily on cognitive evaluations of usefulness and ease of use while neglecting motivational design factors and contextual constraints.

Configurational Pathways to Adoption

fsQCA revealed multiple equifinal configurations leading to high behavioral intention, with Motivation, enjoyment, and habit serving as core conditions across most pathways. This asymmetric (configurational) logic complements the net-effects PLS-SEM findings, showing that distinct combinations of conditions — not isolated predictors — jointly produce adoption. Some pathways functioned even with absent social influence, underscoring the equifinality of adoption routes.

Contextual Sensitivity in Low-Resource Settings

In Lesotho, structural constraints — unstable connectivity, limited digital infrastructure, and uneven digital literacy — create a markedly different adoption environment. Personal innovativeness and motivational moderation effects were weak, underscoring contextual sensitivity. IPMA identified habit as the most influential driver of intention but with only moderate performance, suggesting the largest practical gains come from routinizing GenAI use. The study argues the benefits of GenAI observed in well-resourced systems cannot be assumed to transfer directly to developing contexts.

Relevance to the knowledge base

This paper directly extends the knowledge base's coverage of technology acceptance and AI adoption in higher education. It complements organizational adoption studies (e.g., Exploring Organisational Drivers and Innovation Attributes of Artificial Intelligence Adoption in Higher Education) with a student-level, low-resource perspective, and it connects Motivation to sustained engagement with GenAI. Its hybrid acceptance–motivation–habit model informs AI education practice by arguing for motivation-centered design and context-responsive policy rather than uniform technology rollout. It is relevant to AI policy and to the knowledge base's growing set of adoption-pathway articles.

What this means for practice

• Instructors. Do not stop at usefulness arguments. Cognitive beliefs predicted intention, but motivational dimensions (attention, relevance, confidence, satisfaction) were the stronger determinants of actual use, so design GenAI activities that build confidence and visibly connect to academic achievement.
• Instructors. Routinize GenAI in a small number of recurring, authentic course tasks. Habit was the highest-importance antecedent of intention but scored only moderate performance on the importance-performance map, which is where the authors locate the largest practical gains.
• Administrators. Do not assume enthusiasm will diffuse on its own: personal innovativeness had only limited influence on adoption, so lecturers, academic support units, and peer networks must actively model and support GenAI practice.
• Administrators. Fix the basics before scaling. Unreliable connectivity, limited device access, and unclear institutional policy were the facilitating conditions that determine whether intention turns into sustained use in this resource-constrained context.
• Policymakers. Embed GenAI in curriculum standards and assessment guidelines so it becomes a legitimate learning partner rather than a contested tool, and pair access with guidance to avoid unregulated use.

Limitations

• The study surveyed 842 students at higher education institutions in a single country (Lesotho), recruited through non-probability, multi-site convenience sampling via institutional email lists, learning management systems, and student messaging groups, with non-users screened out at the start.
• The design is cross-sectional, which the authors state limits causal inference; structural paths are interpreted as theory-consistent associations and say nothing about how motivation and habit develop over time.
• All outcomes are self-reported measures of intention and use, leaving room for social desirability bias; no learning analytics or system logs were used to verify behavior.
• The single-country focus limits generalizability, and personal innovativeness and motivational moderation effects came out weak, so the authors call for comparison across African higher education systems before the configurational model is assumed to transfer.

Citation

Tian, X., Ayanwale, M. A., Molefi, R. R., Manchanda, P., Gençel, N., & Ogunjoun, B. O. (2026). Decoding symmetric and asymmetric pathways in generative AI learning adoption: a multi-method study. International Journal of Educational Technology in Higher Education.