Artificial Intelligence in Higher Education: Predictive Analysis of Attitudes and Dependency Among Ecuadorian University Students

Abstract

This study examines the relationship between attitudes toward artificial intelligence (AI) and AI dependency among Ecuadorian university students. A cross-sectional design was used, applying two validated instruments: the Artificial Intelligence Dependence Scale (DAI) and the General Attitudes Toward Artificial Intelligence Scale (GAAIS), with a sample of 540 students. Structural equation modeling (SEM) assessed how both positive and negative attitudes predict dependency levels. Results indicate a moderate level of AI dependency and an ambivalent attitudinal profile. Both attitudinal dimensions significantly predicted dependency, suggesting dual-use behaviors shaped by perceived utility and ethical concerns. Urban students reported higher dependency and greater sensitivity to AI-related risks, highlighting digital inequalities. Although the SEM model showed adequate comparative fit (CFI = 0.976; TLI = 0.973), residual indicators (RMSEA = 0.075) suggest further refinement is needed. This study contributes to underexplored Latin American contexts and emphasizes the need for equity-driven digital literacy strategies in higher education. Findings support pedagogical frameworks promoting critical thinking, ethical reasoning, and responsible AI use. The study aligns with Sustainable Development Goals 4 (Quality Education) and 10 (Reduced Inequalities), reinforcing the importance of inclusive, learner-centered approaches to AI integration.

1. Introduction

Artificial intelligence (AI) has become one of the most transformative technologies of the 21st century, reshaping productive, social, and educational sectors [1]. From its origins in formal logic and computation [2] to current generative models, AI systems now emulate complex cognitive functions such as autonomous learning and decision-making [3]. In higher education, AI facilitates personalized instruction, adaptive feedback, and automated assessment through platforms like intelligent tutors and virtual assistants [4,5]. When properly integrated, these tools promote equity and inclusion [6], fostering the emergence of data-informed, learner-centered pedagogies [7].

To illustrate the conceptual landscape of the study, Figure 1 presents a semantic network generated using VOSviewer version 1.6.20, based on Scopus-indexed publications from 2020 to 2025. The map highlights the strongest co-occurrences among terms such as AI tools, attitudes, dependence, critical thinking, and higher education, which form tightly linked clusters. These relationships reveal a high degree of conceptual alignment and validate the focus of this study on attitudinal and behavioral dimensions of AI use in academic contexts.

While the global literature increasingly explores these themes, the network analysis also revealed a geographic gap: empirical studies focusing on Latin America, particularly Ecuador, remain scarce. This supports the relevance of investigating how students in digitally emerging regions engage with AI tools and the implications for equity-driven educational policies.

However, this integration raises critical concerns. The widespread use of tools like ChatGPT has improved student productivity in academic writing and organization [8,9], but may also lead to over-reliance and weakening reflection and independent reasoning [10,11]. Excessive dependence on AI can inhibit higher-order thinking skills [12,13] and compromise academic integrity [14,15]. Students express ambivalence: while many value AI’s efficiency, others raise ethical concerns, particularly about originality and self-reliance [16,17,18].

Theoretical models such as the Technology Acceptance Model (TAM) and the Unified Theory of Acceptance and Use of Technology (UTAUT) have been widely used to explain users’ attitudes and behavioral intentions toward digital technologies [19,20]. These frameworks emphasize the role of perceived usefulness, ease of use, and social influence in shaping acceptance patterns, particularly in relation to perceived benefits, ethical concerns, and behavioral dependence on AI systems in academic settings [21,22].

Attitudes toward AI significantly shape its adoption and use [23], yet few studies explore these dynamics in Latin American contexts, where digital divides persist [24]. While digital tools may improve learning outcomes, their uncritical use may foster superficial engagement [25,26,27] and reduce cognitive autonomy [28,29]. This is especially problematic in under-resourced regions, where access and training are uneven [30,31].

Although prior research has examined AI in global settings, there is limited empirical evidence from Ecuador, a country undergoing rapid but unequal technological adoption [32,33,34]. Understanding how students in such contexts relate to AI is crucial to developing ethical, sustainable, and inclusive educational policies [35,36]. The present study addresses this gap by analyzing both the degree of AI dependency and attitudinal orientations among Ecuadorian university students, using validated psychometric instruments.

Anchored in frameworks of digital equity and Education for Sustainable Development (ESD), this study contributes to the literature by exploring how cognitive, ethical, and demographic variables shape the academic use of AI. Findings are discussed in relation to Sustainable Development Goals, particularly SDG 4 (Quality Education) and SDG 10 (Reduced Inequalities), and support the design of learner-centered, context-sensitive strategies for responsible AI integration.

This study aims to examine the relationship between Ecuadorian university students’ attitudes toward AI and their perceived dependence on AI tools in academic contexts. Two validated instruments were used: the Artificial Intelligence Dependence Scale (DAI) and the General Attitudes Toward Artificial Intelligence Scale (GAAIS). These scales were selected due to their strong psychometric properties and prior validation in Latin American university populations, which ensures cultural relevance and measurement reliability in this context. A more detailed explanation of their structure and psychometric validation is provided in the Section 2.3.

The focus lies in identifying predictive and sociodemographic patterns that may inform pedagogical and policy strategies for responsible AI integration. Therefore, the following objectives are proposed: (a) to assess the degree of AI dependence among university students; (b) to identify students’ general attitudes, both positive and negative, toward the educational use of AI; (c) to examine differences in dependence and attitudes based on gender, age, academic major, and area of residence; and (d) to model the predictive relationship between attitudes and AI dependence, considering sociodemographic moderators.

From these objectives, it is hypothesized that the frequent use of AI tools is associated with higher levels of perceived dependence (H₁); students in technological or applied science majors report more positive attitudes toward AI than those in traditional fields (H₂); significant differences exist in AI dependence and attitudes based on gender, age, academic major, and residential area (H₃); and both positive and negative attitudes significantly predict AI dependence, and this relationship is moderated by variables such as academic level and housing area (H₄).

2. Materials and Methods

2.1. Research Design

This study employed quantitative, non-experimental, cross-sectional, and correlational-predictive design [37]. It aimed to examine the relationships between students’ attitudes toward AI and their perceived dependence on AI tools in academic settings. The approach allowed for the identification of attitudinal patterns and predictive variables relevant to understanding AI adoption in higher education environments, especially in emerging digital contexts such as Ecuador, also aligned with ethical AI integration and Education for Sustainable Development (ESD), contributing to SDG 4 (Quality Education) and SDG 10 (Reduced Inequalities).

2.2. Participants

The present study included a sample of 540 Ecuadorian university students, composed of 38.57% men and 61.42% women, whose ages ranged from 17 to 35 years (M = 20.68; SD = 3.02). The ethnic composition revealed that 93.88% identified as mestizo, while the remaining 6.12% were distributed between Indigenous, Afro-Ecuadorian, and white identities. A significant proportion of the participants resided in urban areas (67.40%). Regarding the predominant marital status, it was single, representing 92.77% of the sample. Regarding their mode of study, 99.81% stated that they were enrolled in face-to-face mode.

Participants were selected through purposive sampling with institutional mediation, targeting students enrolled in diverse academic programs across multiple universities, a purposive sampling strategy was implemented using formal institutional channels to recruit active university students from diverse academic programs and regions. This approach allowed for the inclusion of participants relevant to the research objectives while ensuring ethical access and demographic variability.

These inclusion criteria were considered: (a) voluntary participation; (b) educational level according to age to guarantee understanding of the instruments. Exclusion criteria included: (a) presence of intellectual disability; (b) effects of substances or drugs that affect consciousness; (c) lack of command of the Spanish language.

2.3. Instruments

Two validated scales were used for data collection:

General Attitudes Towards Artificial Intelligence Scale (GAAIS): Developed by Schepman and Rodway [38] Keele University, Staffordshire, UK. This scale aims to assess general attitudes towards AI in various contexts, including education. The GAAIS consists of 20 items, grouped into two dimensions: positive and negative attitudes towards AI. The measurement is conducted using a 5-point Likert scale, where 0 represents “Strongly disagree” and 4 “Strongly agree”. Regarding its psychometric properties, the GAAIS has shown a high internal consistency, with a Cronbach’s alpha coefficient of 0.85 for the dimension of positive attitudes and 0.82 for the dimension of negative attitudes [39].

Artificial Intelligence Dependency Scale (AID): Developed by Morales-Garcia et al. [40], Universidad Nacional Mayor de San Marcos, Lima, Peru. The purpose is to measure the degree of dependence of students on AI tools in the educational context. It consists of 5 items, measured on a 5-point Likert scale (1 = Strongly disagree, 5 = Strongly agree). With respect to their psychometric properties, they are particularly good, with a Cronbach’s alpha value of 0.87, which constitutes an exceptionally good robustness of the instrument.

Both instruments are considered suitable for assessing psychological constructs essential to understanding AI use within sustainable and ethically aware learning environments, also the scales were reviewed linguistically and conceptually to ensure cultural relevance to the Ecuadorian academic context.

2.4. Procedure

Data collection was conducted online through Google Forms (Google LLC, Mountain View, CA, USA). Participants were first presented with a detailed informed consent form outlining the study objectives, voluntary participation, and guarantees of anonymity and confidentiality. The survey link was distributed via institutional academic channels, including email lists and internal communication platforms. Upon completion, responses were securely stored in Google Sheets (Google LLC, Mountain View, CA, USA), facilitating structured data analysis.

The study received ethical approval from the institutional review board, and all procedures adhered to the ethical principles of the Declaration of Helsinki. Cultural sensitivity and inclusivity were prioritized throughout the data collection process, reflecting international standards for research in underrepresented contexts.

2.5. Data Analysis

The statistical analysis was structured in four interrelated methodological blocks, with the aim of robustly examining the relationships between dependence on AI and student attitudes in the context of higher education. In the first stage, a preliminary descriptive analysis of the variables was conducted, calculating measures of central tendency such as the arithmetic mean (M), dispersion through the standard deviation (SD) and distribution through the coefficients of asymmetry (g₁) and kurtosis (g₂). Univariate normality was assessed by accepting values of g₁ and g₂ within the range ±1.5 [41,42]. And multivariate normality through Mardia’s test [43], considering that when there is no significance (p > 0.05) in multivariate g₁ and g₂, indicates the fulfillment of this assumption.

Subsequently, the general fit model was examined by means of an analysis of the internal consistency of the instruments using McDonald’s omega coefficient (ω), a more appropriate indicator than Cronbach’s alpha on scales with multidimensional structures [44,45]. Additionally, a confirmatory factor analysis (CFA) using Structural Equation Modeling (SEM) was conducted to evaluate the relationships between the latent variables—positive attitudes, negative attitudes, and AI dependence. The SEM was estimated using the Diagonal Weighted Least Squares (DWLS) method, appropriate for ordinal data that do not meet the assumption of multivariate normality, and the variables are ordinal [46,47,48]. For the fit of the model, the indices Chi square (χ²), Chi square normed (χ²/df), Comparative Fit Index (CFI), Tucker–Lewis Index (TLI), Standardized Mean Square Residual (SRMR), and Mean Square Error of Approximation (RMSEA) and the factor loads (λ) of the items were considered. The fit model is considered appropriate when the value of χ² is p > 0.05 or χ²/df is less than 4; CFI and TLI is greater than 0.95; and SRMR and RMSEA are less than 0.06 (although tolerances of 0.08 are acceptable) [49,50,51,52]. Finally, the factor loads (λ) of the items were considered adequate if they exceed the value of λ > 0.40 [53].

In a third stage, the scores obtained in the scales were compared according to sociodemographic variables (sex, area of residence) and academic variables (general average), for dichotomous variables the Wilcoxon rank sum test (also known as the Mann–Whitney U) test was applied [54], and for comparisons between more than two groups, the Kruskal–Wallis test was used [55]. The level of statistical significance was established at p < 0.05 [49].

Finally, a robust regression model with standard errors corrected for heteroskedasticity was estimated, to identify the significant predictors of dependence on AI [56]. In the first model, the global score of the attitude scale (GAAIS) was introduced; the second model incorporated the subdimensions of positive and negative attitudes; and the third model additionally included sociodemographic variables such as area of residence and academic level. In all cases, standardized coefficients (β), standard errors (SE) and t-values were reported, considering those with p < 0.05 significant [57]. All predictors were treated as continuous variables, using total or subscale scores as independent variables.

The statistical analyses used in the studies were performed using the programming language R, version 4.3.1 [58] (R Foundation for Statistical Computing, Vienna, Austria; available at https://www.r-project.org) and the packages used were lavaan, semTools, MBESS, and dplyr.

2.6. Use of Generative Artificial Intelligence (GenAI)

No generative AI tools were used in the research design, data collection, analysis, or interpretation. Limited AI-assisted tools were employed only for grammar correction and improving cohesion, without influencing scientific reasoning or argumentation.

3. Results

3.1. Descriptive Analysis

Table 1. Preliminary analysis of the instruments.

Factors/Dimensions	Descriptive Statistics				Correlation Matrix
	M	SD	g1	g2	1	2	3	4
GAAIS	67.72	13.93	−824	1.87	-
Positive attitudes	33.56	7.32	−647	1.23	0.877 **	-
Negative attitudes	34.15	7.70	−596	0.936	0.896 **	0.592 **	-
	Mardia		4375.376 **	70,437 *
DAI	12.04	5.42	−222	−128	0.297 **	0.268 **	0.278 **	-
	Mardia		105,476 **	21,063 *

Note. ** (p < 0.01); * (p < 0.05); M = mean, SD = standard deviation; g₁: Asymmetry; g₂: Kurtosis; GAAIS: Scale of General Attitudes towards Artificial Intelligence; DAI: Artificial Intelligence Dependency Scale.

shows the descriptive analysis of the General Attitudes Toward Artificial Intelligence Scale (GAAIS) and the Artificial Intelligence Dependency Scale (DAI), including means (M), standard deviations (SD), skewness (g₁), and kurtosis (g₂). The GAAIS yielded an overall mean of 67.72 (SD = 13.93), with subdimensions of positive (M = 33.56, SD = 7.32) and negative attitudes (M = 34.15, SD = 7.70). The DAI presented a total mean of 12.04 (SD = 5.42). All variables exhibited negative skewness and moderate kurtosis. Mardia’s multivariate tests revealed significant deviations from normality (p < 0.01), guiding the decision to apply robust, non-parametric and SEM methods adjusted for non-normality.

3.2. Internal Consistency and Measurement Model

Table 2. Internal Consistency Analysis of Instruments.

Instruments	ω	(Total)
GAAIS	0.93	[0.92–0.94]
Positive attitudes	0.87	[0.84–0.89]
Negative attitudes	0.92	[0.90–0.93]
DAI	0.85	[0.83–0.87]

Note. ω: McDonald’s omega coefficient. GAAIS: Scale of General Attitudes towards Artificial Intelligence; DAI: Artificial Intelligence Dependency Scale.

shows the analysis of the internal consistency of the instruments used, expressed through McDonald’s omega coefficient (ω) and their respective 95% confidence intervals. GAAIS is highly dependable, with a total omega coefficient of 0.93 [0.92–0.94]. Its subdimensions also show robust internal consistency, with ω values of 0.87 [0.84–0.89] for positive attitudes and 0.92 [0.90–0.93] for negative attitudes. On the other hand, the DAI shows adequate reliability with an omega coefficient of 0.85 [0.83–0.87]. These results indicate that both instruments have a high internal coherence in the sample analyzed, which supports their use for the measurement of attitudes and dependence towards AI.

Figure 2 displays the structural equation model (SEM) assessing the relationships between students’ attitudes toward AI (positive and negative) and their perceived dependency. The three latent constructs presented significant standardized factor loadings, mostly above 0.60, indicating good structural validity.

The model exhibited a chi-square value of χ²(129) = 1974.06, p < 0.001. Although significant (as expected with large samples), the Comparative Fit Index (CFI = 0.976) and Tucker–Lewis Index (TLI = 0.973) indicate excellent incremental fit. The RMSEA value of 0.075 (90% CI [0.649, 0.842]) and SRMR of 0.072 fall within acceptable thresholds, suggesting that the model achieves a satisfactory approximation of the observed data despite minor residual deviations.

Taken together, these indicators support the overall adequacy of the model for representing the attitudinal and behavioral constructs under investigation.

3.3. Group Comparisons

Table 3. Comparison of scales according to sociodemographic and academic variables.

Independent Variable	Scale	W	χ2	gl	p
Sex	DAI	32,105	--	--	0.168
	Positive attitudes	31,800	--	--	0.122
	Negative attitudes	37,215	--	--	0.127
	GAAIS	34,523	--	--	0.998
Housing Area	DAI	28,258	--	--	0.026
	Positive attitudes	32,253	--	--	0.897
	Negative attitudes	37,208	--	--	0.0023
	GAAIS	3304	--	--	0.054
Grade point average	DAI	--	4.05	2	0.132
	Positive attitudes	--	2.47	2	0.291
	Negative attitudes	--	0.56	2	0.755
	GAAIS	--	1.51	2	0.470

Note. W: Wilcoxon rank sum test; χ²: Kruskal–Wallis test; GAAIS: Scale of General Attitudes towards Artificial Intelligence; DAI: Artificial Intelligence Dependency Scale.

reports the results of non-parametric comparisons across sociodemographic and academic variables. No significant differences were observed by sex or academic performance (p > 0.05). However, living area significantly affected DAI scores (p = 0.026) and negative attitudes (p = 0.0023), with urban students reporting higher values. This may reflect contextual or access-related variables affecting perceptions and use of AI.

3.4. Robust Regression Models

Table 4. Robust Regression Models for AI Dependency Prediction.

Model	Variable	β	Standard Error	t
Model	(Intercept)	2.008	1.906	1.832
	GAAIS	0.151	0.016	9.549
Model 2	(Intercept)	2.002	1.100	1.819
	Positive attitudes	0.155	0.043	3.573
	Negative attitudes	0.148	0.041	3.584
Model 3	(Intercept)	0.384	1.218	0.315
	Positive attitudes	0.127	0.044	2.904
	Negative attitudes	0.176	0.042	4.236
	Housing Area	0.603	0.456	1.322
	Average	0.986	0.633	1.558

Note: β: Coefficient; GAAIS: Scale of General Attitudes towards Artificial Intelligence; Vivienda_1 Zone: Urban; Promedio_2: Remarkable; Promedio_3: Enough.

summarizes the three robust regression models predicting AI dependence. Model 1 confirms a significant positive association between general attitudes and dependence (β = 0.151, t = 9.549), indicating that more favorable global perceptions of AI correspond to higher self-reported reliance. Model 2 distinguishes positive (β = 0.155, t = 3.573) and negative (β = 0.148, t = 3.584) attitudes as independent and significant predictors, suggesting that even critical perceptions may coexist with behavioral dependence, likely due to instrumental use despite ethical reservations.

Model 3 incorporates sociodemographic moderators. Although attitudes remain significant (especially negative attitudes, β = 0.176, t = 4.236), residential area and academic performance do not show statistically significant predictive effects. These findings confirm the dual attitudinal pathway: both enthusiasm and skepticism toward AI contribute to its academic use, reflecting the complex psychosocial dynamics of technology adoption in higher education.

4. Discussion

This study provides empirical evidence on the relationship between Ecuadorian university students and AI-based technologies in academic contexts. The findings indicate a moderate level of dependence, suggesting a phase of progressive but incomplete consolidation. This aligns with prior research highlighting a gradual incorporation of AI in higher education, where ethical concerns and perceived utility interactively shape acceptance [4,5].

Attitudes toward AI show marked duality, with comparable levels of positive and negative perceptions. This ambivalence is consistent with theoretical models from technology adoption, such as the Technology Acceptance Model (TAM) and the Unified Theory of Acceptance and Use of Technology (UTAUT), which explain user behavior as a balance between perceived usefulness, perceived ease of use, and contextual concerns [21,22]. From a psychological perspective, it also aligns with attitude theory, where simultaneous favorable and unfavorable views may reflect cognitive dissonance or instrumental adaptation in pressured academic environments [10,31].

These findings reinforce the notion that AI integration must go beyond technological deployment to include pedagogical frameworks that promote metacognitive reflection and student agency [7,30]. Particularly relevant is the observation that both positive and negative attitudes significantly predict AI dependence. This paradox reflects the functionalization of AI: it is viewed as indispensable regardless of critical views, given the academic pressures to adapt to digital environments [59,60].

Therefore, instructional designs must include scaffolding and formative assessments that foster ethical reasoning and reduce superficial learning risks. At the sociodemographic level, no significant differences were found by gender, contrary to studies noting digital gaps [61]. This may indicate progress in digital equity at the university level in Ecuador. However, significant differences due to residence area were observed: urban students reported greater dependence and more negative attitudes. This may stem from increased exposure and digital saturation, generating both intensive use and heightened ethical concerns [31].

These patterns point to the need for targeted institutional strategies that address digital divides through infrastructure investment and localized digital training programs. In contrast, academic performance showed no significant relationship with attitudes or dependence, highlighting the influence of contextual and cultural factors. This suggests that curricula should embed digital ethics, critical thinking, and self-regulation skills as transversal competencies [62,63].

Moreover, students’ concerns about plagiarism and loss of creativity emphasize a pressing pedagogical challenge: ensuring academic integrity in AI-rich environments. Authentic assessment models that privilege process over product are essential in this regard [15,35]. Taken together, these results support the view that AI integration in education must be pedagogically intentional, ethically grounded, and culturally responsive.

This is especially critical in Latin America, where digital transformation intersects with structural inequalities. In this context, the study aligns with the goals of Education for Sustainable Development (ESD), which emphasizes responsible innovation, social justice, and the development of digital and cognitive competencies for sustainability. Moreover, the findings contribute to the advancement of Sustainable Development Goals (SDGs), notably SDG 4 (Quality Education) and SDG 10 (Reduced Inequalities), by highlighting the sociotechnical dynamics that condition educational access and equity [64].

From an applied perspective, the study identifies priority actions for higher education institutions. First, it is essential to implement data protection and algorithmic transparency policies, particularly regarding AI systems that process student data [31]. Second, sustained teacher training on AI integration is required to ensure meaningful, ethical use in diverse instructional contexts [59]. Third, equity-focused policies must be adopted to guarantee fair access to technological resources, especially in underserved areas.

In relation to the proposed hypotheses, the findings confirm H₁, as frequent use of AI tools was positively associated with higher levels of perceived dependence. H₂ was partially confirmed: although students in technological or applied science majors tended to show more positive attitudes toward AI, these differences were not statistically significant in all comparisons. H₃ was also partially confirmed, with significant differences in AI dependence and negative attitudes according to the area of residence, but no differences by gender, age, or academic performance. Finally, H₄ was confirmed, as both positive and negative attitudes significantly predicted AI dependence, with these relationships persisting even when sociodemographic variables were included as moderators.

Finally, future research should expand on these findings through longitudinal and mixed methods approaches, which would allow deeper analysis of attitudinal shifts over time. It is also necessary to explore the perspectives of educators and institutional leadership, as these are key actors in scaling sustainable AI use across educational systems. The broader educational challenge is not simply to integrate AI, but to do so in ways that support inclusion, justice, and intellectual autonomy pillars of a truly sustainable digital education paradigm.

5. Conclusions

This study evidences a nuanced relationship between Ecuadorian university students and AI technologies, marked by moderate dependence and ambivalent attitudes. Both favorable and critical perceptions significantly predict AI reliance, indicating instrumental use beyond personal preference.

Sociodemographic comparisons revealed contextual disparities, particularly in urban settings, where higher dependence and ethical concerns were reported. These findings highlight the need for equity-oriented strategies that address digital exposure, access, and ethical guidance. For higher education systems, the results underscore the importance of pedagogical models that foster critical thinking, ethical awareness, and student autonomy, supported by teacher training and responsible institutional governance of AI tools.

Given the specific socio-cultural, infrastructural, and policy conditions of Ecuador’s higher education system, these results should be interpreted with caution when considering their applicability to other regions. While certain attitudinal and behavioral patterns toward AI may be comparable in contexts with similar digital divides, differences in technological infrastructure, cultural perceptions, and institutional governance may limit direct extrapolation of the findings.

Beyond offering empirical insight into an underexplored context, this research supports the design of inclusive, sustainable educational strategies tailored to the digital transformation of learning environments. Future studies should adopt longitudinal and institutional approaches to better understand the long-term impacts and readiness for ethical AI integration.