Abstract

Extended Abstract
In the digital age, higher education institutions are increasingly charged with nurturing entrepreneurial behaviors in their students to drive economic and technological advancement. Yet, many graduates—especially in contexts such as Iran—face structural obstacles (e.g., limited funding, weak university–industry partnerships) that hinder their ability to convert innovative ideas into viable enterprises. Simultaneously, the rapid growth of artificial intelligence (AI) offers vast new prospects, but realizing these opportunities demands not only advanced digital literacy but also a strong sense of entrepreneurial self-efficacy: the conviction that one can effectively translate technical capabilities into concrete entrepreneurial actions. Despite existing research on each of these factors in isolation, little is known about how AI learning intentions and digital literacy interact to foster entrepreneurial behavior via enhanced self-efficacy. This study addresses that gap by posing the central question: How do intentions to learn AI and levels of digital literacy jointly shape postgraduate students’ entrepreneurial behaviors through the mediating influence of entrepreneurial self-efficacy? Insights from this inquiry aim to inform the development of integrated educational policies and programs that bolster university-based entrepreneurial ecosystems.
Theoretical Framework
Our investigation draws on three foundational theories:
1. Theory of Planned Behavior (Ajzen, 1991) posits that attitudes, subjective norms, and perceived behavioral control shape intentions, which in turn predict actual behaviors. We conceptualize AI learning intention as reflecting both favorable attitudes toward AI and perceived control over acquiring those skills.
2. Social Cognitive Theory (Bandura, 1986) highlights self-efficacy as the engine of motivation, effort, and perseverance. Here, entrepreneurial self-efficacy is expected to mediate the translation of technological competencies (AI and digital skills) into sustained entrepreneurial action.
3. Unified Theory of Acceptance and Use of Technology (Venkatesh et al., 2003) underscores performance expectancy and effort expectancy as key drivers of technology adoption. Digital literacy embodies users’ confidence and proficiency in applying digital tools, influencing their ease of integrating technology into entrepreneurship.
From an extensive literature review, three research streams emerged but remain insufficiently integrated:
• AI and Entrepreneurship: AI can uncover novel business opportunities and models (Mogonzova & Menchidi, 2024; Chen & Zhang, 2023), yet the psychological pathways—particularly self-efficacy—remain underexplored.
• Digital Literacy and Entrepreneurship: Digital competencies bolster perceived behavioral control and entrepreneurial intention (Arifin et al., 2023), but their joint effect with AI learning has not been examined.
• Self-Efficacy as Mediator: Self-efficacy mediates training outcomes on entrepreneurial behavior (Bouy & Duong, 2024), yet integrated models including both AI and digital literacy are scarce.
Hypotheses
• H1: AI learning intention has a positive direct effect on entrepreneurial behavior.
• H2: Digital literacy has a positive direct effect on entrepreneurial behavior.
• H3: Entrepreneurial self-efficacy has a positive direct effect on entrepreneurial behavior.
• H4: Entrepreneurial self-efficacy mediates the relationship between AI learning intention and entrepreneurial behavior.
• H5: Entrepreneurial self-efficacy mediates the relationship between digital literacy and entrepreneurial behavior.
Methodology
A descriptive–correlational, cross-sectional design was applied, utilizing Structural Equation Modeling (SEM) to test the proposed hypotheses.
• Population and Sampling: The universe consisted of 901 master’s and doctoral students in Electrical Engineering, Computer Science, and Engineering Technology at Urmia University (academic year 1403–1404/2024–2025). Based on Cochran’s formula, a minimum sample of 269 was required; anticipating non-responses, 350 questionnaires were distributed via stratified random sampling. Ultimately, 300 valid responses (85.7% response rate) were analyzed.
• Instruments: Four standardized, five-point Likert scales were employed:
1. AI Learning Intention Scale (Chai et al., 2024; 18 items, four subscales)
2. Digital Literacy Questionnaire (Martin, 2008; 20 items, three subscales)
3. Entrepreneurial Self-Efficacy Scale (DeNoble et al., 1999; 21 items, six subscales)
4. Entrepreneurial Behavior Questionnaire (Gyor et al., 2020; 7 items, two subscales)
• Validity and Reliability:
o Content Validity: Evaluated by ten experts via Lawshe’s Content Validity Ratio (CVR), yielding a mean CVR of 0.90 (minimum retained item CVR = 0.62).
o Construct Validity: Confirmatory Factor Analysis (CFA) in AMOS 26 produced fit indices within acceptable thresholds (CFI > 0.90, TLI > 0.90, RMSEA < 0.08).
o Reliability: Cronbach’s alpha coefficients ranged from 0.777 (Digital Literacy) to 0.849 (AI Learning Intention).
• Data Analysis: Normality (Kolmogorov–Smirnov p > 0.05) and absence of multicollinearity (VIF < 2.5) were confirmed using SPSS 26. Pearson’s correlation assessed bivariate associations. SEM in AMOS 26 tested both direct and indirect effects, with bootstrapping (2,000 resamples) generating bias-corrected 95% confidence intervals and Sobel tests determining mediation significance.
Results
• Demographics: Mean age was 27 years (SD = 2.8); 79.7% male, 20.3% female; 70.3% master’s and 29.7% doctoral students.
• Correlation Analysis: All variables correlated positively and significantly (p < 0.01). The strongest association was between AI learning intention and entrepreneurial behavior (r = 0.643), while the weakest was between AI learning intention and self-efficacy (r = 0.436).
• Direct Effects (SEM):
o H1: AI → Entrepreneurial Behavior: β = 0.507, C.R. = 7.673, p < 0.001
o H2: Digital Literacy → Entrepreneurial Behavior: β = 0.359, C.R. = 5.190, p < 0.001
o H3: Self-Efficacy → Entrepreneurial Behavior: β = 0.350, C.R. = 5.220, p < 0.001
• Mediation Effects (Bootstrapping):
o H4: AI → Self-Efficacy → Behavior: Indirect β = 0.105, Sobel z = 4.12, p < 0.001
o H5: Digital Literacy → Self-Efficacy → Behavior: Indirect β = 0.125, Sobel z = 5.45, p < 0.001
• Model Fit: χ²/df = 2.87; RMSEA = 0.070; GFI = 0.98; AGFI = 0.93; NFI = 0.904; CFI = 0.922, indicating a robust overall fit.
Discussion
The results align with TPB and UTAUT, affirming that favorable attitudes toward AI and strong digital competencies heighten entrepreneurial intentions and perceived control. Consistent with SCT, entrepreneurial self-efficacy emerged as a pivotal mediator: confidence in one’s entrepreneurial skills translates technical proficiencies into proactive venture creation. Mastery of AI and digital tools thus enhances perceived competence, fostering persistence, risk-taking, and resilience—cornerstones of entrepreneurial success.
Conclusions and Implications
This study substantiates that AI learning intentions and digital literacy exert both direct and indirect influences (via entrepreneurial self-efficacy) on entrepreneurial behavior among graduate students. To bolster university entrepreneurship ecosystems, it is recommended that higher-education curricula:
1. Integrate practical, project-based AI training and digital literacy modules.
2. Incorporate confidence-building interventions (e.g., entrepreneurship workshops, mentorship programs, simulation exercises) explicitly aimed at strengthening self-efficacy.
Such an integrated approach is essential for overcoming systemic barriers in Iran’s higher education context and for nurturing sustainable innovation and economic development.
Limitations and Future Directions
Key limitations include reliance on self-report data, cross-sectional design, and a single-institution engineering sample, which constrain causal inference and generalizability. Future research should employ longitudinal or experimental designs, diversify academic disciplines and institutions, and explore additional mediators (e.g., social support, institutional climate) and contextual moderators through mixed-methods approaches.