How Does Self-Regulated Learning Resist AI Dependence? A Mediating Effect Study Based on College Students Who Frequently Use AIGC Tools

Abstract

This study examines the impact pathway and mediating mechanisms of self-regulated learning (SRL) on AI dependence among university students who frequently use Artificial Intelligence Generated Content (AIGC) tools. A questionnaire survey was conducted among 367 college students in Fujian Province in China. Structural equation modeling analysis revealed that: (1) SRL ability significantly negatively affects AI dependence (β = −.255) and further suppresses AI dependence through the mediating effect of reduced AIGC usage frequency, with the mediation effect accounting for 55.6%; (2) AIGC usage frequency positively exacerbates AI dependence (β = .656), with the highest dependence risk observed in course learning scenarios (r = .379); (3) Significant group differences exist—female students are more adept at utilizing AIGC for self-directed learning and exhibit lower dependence, while senior students effectively balance tool usage by enhancing their SRL ability. The study reveals the dual inhibitory mechanisms of SRL in technology dependence (direct and mediating effects), emphasizing the need for metacognitive training and differentiated tool usage guidelines tailored to various learning scenarios in educational practice. This provides a theoretical basis for mitigating the risks of AI dependence and ensuring the beneficial application of AIGC tools as an “empowering rather than substitutive” resource.

Keywords

self-regulated learning (SRL)Artificial Intelligence-Generated Content (AIGC)AI dependence mediating effect AIGC tool

Introduction

In recent years, the rapid development of Artificial Intelligence Generated Content (AIGC) technology has profoundly transformed the way humans learn and work (Y. Wang et al., 2023). Particularly in education, AIGC tools, such as ChatGPT, provide learners with powerful assistance, significantly enhancing learning efficiency. These tools can quickly generate high-quality learning resources, offer personalized learning support, and play a crucial role in various learning scenarios, including course learning, research assistance (Ilić et al., 2023; X. Yang et al., 2024). However, with the widespread application of AIGC tools, academia and educational practitioners have begun to focus on whether learners may become overly dependent on these tools, which could, in turn, impact their self-regulated learning abilities and the development of creative thinking (Çela et al., 2024; H. J. Wang et al., 2024). Excessive reliance on AIGC tools may lead to a loss of independent thinking and problem-solving skills among learners, potentially weakening their long-term autonomy in learning.

Self-Regulated Learning (SRL) is an active learning process that emphasizes learners’ ability to set learning goals, develop study plans, and monitor their learning progress. Studies have shown that self-regulated learning ability is a key factor in reducing technological dependence and promoting learning autonomy (Eckley et al., 2023; Mirmoadi & Satwika, 2022). However, in the context of widespread AIGC tool usage, the influence of self-regulated learning ability on learners’ behavior in using AIGC tools, and whether such usage behavior further affects their dependence on AI, remains an issue that requires in-depth exploration. In particular, the intensity of AIGC tool usage and the associated dependency risks may vary significantly across different learning scenarios, providing new perspectives for research.

This study aims to explore the impact pathways of self-regulated learning on AI dependence by constructing a theoretical model and conducting empirical research, while also analyzing the mediating role of AIGC usage in this relationship. Specifically, this study seeks to answer the following key questions: (1) How does self-regulated learning (SRL) exert a dual inhibitory mechanism on AI dependence, examining both the direct effect and the indirect mediating effect through learners’ AIGC usage frequency? (2) Does the relationship between AIGC usage and AI dependence vary across different learning scenarios (e.g., course learning support vs. research and academic support), and how does academic experience (grade level) moderate this relationship? (3) How do gender differences manifest in AIGC usage frequency and the resulting AI dependence in autonomous learning scenarios, and what implications do these findings have for targeted intervention strategies?

The theoretical significance of this study lies in enriching the theoretical framework of AIGC technology applications in educational contexts and revealing the potential mechanisms by which self-regulated learning influences technological dependence. From a practical perspective, this study provides scientific evidence for educators, helping them to guide learners in effectively utilizing AIGC tools while reducing the risks of technological dependence. By deeply exploring the relationship between AIGC tool usage and dependency, this study offers important insights for building an educational ecosystem that “empowers rather than replaces.”

Theoretical Review and Research Hypothesis

Theoretical Review

This study adopts Social Cognitive Theory (SCT) as its primary analytical framework and incorporates self-regulated learning (SRL) theory to explain the specific mechanisms at play. Social Cognitive Theory (Bandura, 1986) emphasizes that human functioning is the product of triadic reciprocal causality among personal factors (P), environmental factors (E), and behavior (B). This framework provides a solid theoretical foundation for analyzing the complex relationships among SRL, AIGC usage, and AI dependence. In this study, these core theoretical elements correspond to the following:

Personal Factors (P): Self-regulated learning ability (SRL).

Behavior (B): AIGC tool usage behavior (AIGC usage).

Environment/Outcome (E): AI dependence (as a behavioral outcome influenced by environmental factors such as tool affordances and comfort of use).

Self-Regulated Learning

Self-Regulated Learning (SRL) refers to the method by which learners accomplish their learning goals through autonomous planning, monitoring, and reflection (Schunk & Zimmerman, 2011). It is a core competency that empowers learners to effectively participate in and control their learning processes, encompassing various aspects such as cognition, motivation, and behavior. By setting clear goals, utilizing effective learning strategies, monitoring progress, and engaging in reflection, learners can complete learning tasks more effectively.

As a core personal factor (P) within the SCT framework, the level of SRL ability directly reflects learners’ capacity to monitor and regulate their own learning processes. SCT deeply influences SRL. According to the model of triadic reciprocity, individuals’ self-regulatory capabilities are shaped by the interaction among personal factors (e.g., self-efficacy, learning motivation), environmental conditions, and behavioral engagement (Zimmerman, 1989). Among these, self-efficacy and learning motivation are considered key drivers that motivate learners to engage actively in the learning process.

The SRL cyclical models proposed by Zimmerman (2000) and Pintrich (2000)—including phases such as planning, performance/monitoring/control, and reflection—provide a systematic framework for examining learners’ self-regulatory processes.

With the increasing prevalence of Artificial Intelligence-Generated Content (AIGC) and other innovative technologies, the practical context for self-regulated learning has undergone significant changes. Digital tools can provide abundant learning resources and personalized feedback, empowering learners to plan and monitor their learning more accurately (Junaštíková, 2024). However, these tools may also diminish learners’ autonomy, leading to dependency issues. Frequent use of AIGC tools, particularly among learners with weaker self-regulation abilities, can lead to superficial learning and insufficient deep thinking (H. J. Wang et al., 2024; J. Wu et al., 2023). Therefore, SRL, as the “personal factor (P)” within the SCT framework that enhances learner autonomy and mitigates dependency risks, serves as the theoretical starting point of this study.

AIGC Usage

AIGC tools themselves—such as their highly efficient content-generation capabilities and instant feedback mechanisms—constitute the environmental factors (E) that influence learners. In contrast, learners’ actual use of these tools (AIGC Usage) represents the core behavioral factor (B) within the SCT framework.

Due to their significant advantages in providing personalized resources for coursework, assisting data analysis and content generation in academic research, and enhancing learning efficiency and experience, AIGC technologies have rapidly transformed educational practices (Ilić et al., 2023; Onesi-Ozigagun et al., 2024). It is precisely the high convenience and immediacy of AIGC (environmental factor E) that prompt learners to use such tools across various learning contexts.

Learners’ frequency and patterns of AIGC usage constitute the core behavioral factor (B) in the SCT framework. When learners frequently outsource cognitive tasks to technology across multiple contexts, such behavior may lead to risks, including superficial thinking, reduced originality, and a lack of depth in academic work (Chai et al., 2024). The instant rewards offered by AIGC and its low cognitive demands may, within the triadic reciprocity of SCT, weaken the influence of personal factors (such as SRL) over behavior. Meanwhile, AIGC’s personalized intervention strategies based on learning-behavior data further demonstrate how environmental factors feed back into and shape behavioral factors (Y. Liu et al., 2024; Xie et al., 2025).

Therefore, AIGC usage (B) is the key link connecting personal factors (SRL) with the outcome (AI dependence), and the intensity and nature of this behavioral component directly reflect learners’ behavioral choices and levels of self-regulation in digital learning environments.

AI Dependence

AI Dependence is defined as an individual’s excessive reliance on AI tools, manifesting in both cognitive and behavioral aspects (Zhang et al., 2024). In educational and learning settings, such dependence is often accompanied by a decline in learning initiative, a weakening of self-regulation skills, and a deficiency in the depth of information internalization (Karran et al., 2024).

AI dependence (AID) represents the primary negative outcome (Outcome E) arising from an imbalance within the triadic reciprocity of SCT. The complex interaction of multiple factors shapes this imbalance. First, environmental factors—namely, the high convenience and instant feedback of AIGC—trigger behavioral factors such as frequent and unnecessary use. The habitual outsourcing of cognitive tasks to technology resulting from long-term overuse leads learners to develop dependent behavioral patterns (Bulawan et al., 2024; Dergaa et al., 2024). In addition, the broad applicability of AI tools (X. Li et al., 2023), the high intelligence yet low transparency of algorithms (Y. Wang et al., 2023), and users’ blind trust in AI outputs (Müller, 2022) constitute environmental and social factors that further reinforce this outcome.

From the perspective of SCT, the mechanism underlying the development of AI dependence can be described as follows: when learners’ personal factors (P)—such as low SRL or weak self-efficacy—are insufficient to counteract the behaviors (B) induced by environmental factors (E), particularly the extreme convenience of AIGC, AI dependence emerges as a negative consequence (E) (Huang et al., 2024). Moreover, the broader sociocultural and educational environments surrounding technology use also influence learners’ susceptibility to dependence (Fatimah et al., 2021).

The potential risks associated with AI dependence should not be overlooked, as it may weaken learners’ autonomy, erode critical thinking skills, and exacerbate issues of educational equity (Bulathwela et al., 2024; Nyaaba et al., 2024). Therefore, this study aims to explore how strengthening personal factors (SRL) within the SCT framework can help restore balance in the triadic interaction and mitigate the negative outcomes associated with AI dependence.

Research Hypotheses

Self-Regulated Learning and AI Dependence Tendency

Previous studies have shown that self-regulation not only plays a crucial role in improving learning outcomes but also serves as a protective factor in reducing addictive behaviors such as alcoholism and smoking (Ansell et al., 2012; Hamilton et al., 2013). Similarly, in the context of the widespread application of artificial intelligence (AI) tools, self-regulated learning ability may also be a key factor in mitigating AI Dependence tendencies.

Specifically, self-regulated learning optimizes learners’ cognitive resource allocation and motivation management, effectively reducing reliance on external tools. In complex task environments, individuals with high self-regulation skills are better able to balance the use of AI tools with independent thinking, thereby avoiding excessive dependence on AI (Seufert et al., 2024). Conversely, individuals with low self-regulation skills tend to rely on AI tools to simplify cognitive processes when faced with complex tasks, leading to a decline in learning initiative and an increase in cognitive inertia. Such dependency behaviors not only weaken learners’ autonomy but may also further reduce their self-efficacy (Ullah & Sreedevi, 2024).

Additionally, studies have indicated that frequent use of AI tools may erode learners’ self-control abilities, thereby exacerbating AI Dependence (Rodríguez-Ruiz et al., 2025). However, individuals with high self-regulation skills can counteract this effect by engaging in reflective and strategic cognitive activities, which help them maintain critical thinking when using AI tools and avoid blindly trusting AI-generated content (Choi et al., 2022). Therefore, in AI-dominated learning environments, self-regulated learning serves not only as a safeguard for independent learning but also as a crucial strategy for mitigating the negative effects of AI tools.

Based on the above theoretical analysis, this study proposes the following hypothesis:

H1: Among individuals who frequently use AI, self-regulated learning negatively affects AI Dependence.

AIGC Usage and AI Dependence

While the frequent use of AI tools offers convenience, it also raises concerns about excessive dependence. Specifically, AIGC tools can enhance learning motivation and improve knowledge acquisition, but when users have high trust in AI tools, their frequent usage may significantly increase tool dependence while weakening the willingness for independent learning (D. Liu & Cheng, 2024; Schreibelmayr et al., 2023). This dependency behavior is closely related to the “black-box effect” of AIGC tools, where users focus more on the immediate results provided by the tools while neglecting an in-depth understanding of the processes and methods involved (Y. Wang et al., 2023; S. F. Wang & Chen, 2024). Although AIGC tools significantly improve learning efficiency, they may also encourage users to rely excessively on them for complex tasks rather than leveraging their own abilities to complete the tasks, forming a paradox of technological dependence.

From the perspective of behavioral psychology, the instant generation function of AIGC tools lowers the user threshold and meets users’ needs for speedy problem-solving, thereby expanding tool awareness and psychological dependence (Lu et al., 2024; W. Yang et al., 2025). For example, Yao et al. (2025) analyzed designers’ appropriation of AIGC tools and found that perceived ease of use and perceived usefulness significantly influenced their willingness to use AIGC applications. Designers often combine tools such as ChatGPT (for research analysis) and Midjourney (for visual output) to enhance their work proficiency. However, this habitual behavior may reinforce reliance on tools, thereby smothering inventive and free-thinking capacities (H. J. Wang et al., 2024).

In summary, AIGC tools, due to their efficiency and personalized features, can meet users’ needs and upgrade work efficiency in a brief period. However, such highly dependent behavioral patterns may weaken users’ independence and learning abilities, potentially leading to long-term technological reliance. Particularly, frequent users of AIGC tools may gradually lose the ability to solve problems independently, relying instead on AI-generated answers or solutions, which can shape psychological reliance and behavioral inertia toward AI tools. This reliance not only affects learners’ autonomy but may also negatively impact their long-term learning ability and creative thinking.

Based on the above analysis, this study proposes the following research hypothesis:

H2: Among frequent AIGC users, AIGC usage positively influences AI dependence.

Self-Regulated Learning and AIGC Usage

Individuals with higher self-regulated learning capacities tend to effectively manage their learning processes, including goal setting, time allocation, and learning resource selection, subsequently decreasing over-reliance on external tools. Prasad and Sane (2024) suggested that students with strong self-regulation abilities are more likely to avoid blindly accepting generative AI content and instead critically evaluate the reliability and applicability of AI-generated outputs. This characteristic enables them to maintain independence and control over their learning processes, even in scenarios involving frequent AIGC usage, thereby significantly decreasing both the frequency of AIGC use and reliance on AI tools.

Furthermore, the frequent use of generative AI tools may negatively impact learning motivation and cognitive abilities, thereby weakening learners’ ability to think critically and innovate. In any case, this negative effect is less pronounced in learners with strong self-regulation abilities, as they excel in leveraging technology tools for time management and adjusting metacognitive strategies, thereby mitigating the path reliance effect induced by technology (Deng et al., 2021). Essentially, learners with higher self-regulation abilities can better balance the involvement of technological tools, preventing the scenario where technology dominates the learning process (Zheng et al., 2016). It should be noted that the measurement of AIGC use in this study primarily reflects usage frequency, rather than effective use or strategic use. According to resource management strategies and cognitive load regulation theory (Seufert et al., 2024), individuals with higher levels of self-regulated learning (SRL) tend to voluntarily reduce the frequency of using external, substitutive tools that may undermine autonomous cognitive processing, thereby maintaining control and autonomy in the learning process. Therefore, learners with higher SRL levels tend to exhibit stronger selectivity and restraint in their use of AIGC tools, reducing unnecessary or dependency-driven engagement. These studies collectively demonstrate that self-regulated learning not only enhances learners’ critical awareness of AIGC tools but also decreases the frequency of AI tool usage through independent, goal-oriented behavior, thereby playing a negative moderating role in the frequent use of AIGC tools.

Hence, this study proposes the following research hypothesis:

H3: Among frequent AI users, self-regulated learning negatively influences AIGC usage.

The Mediating Role of AIGC Usage

AIGC tools are not merely passive; they play a dual role in the learning process. On the one hand, they act as academic assistants, helping learners perform more effective tasks by providing instant feedback or simplifying complex tasks (Song et al., 2022). On the other hand, if they are misused, they can hurt learning, leading to excessive statements and hindering independent thinking (Pfost et al., 2023). This dual effect means that AIGC tools are a major element in understanding the relationship between self-regulated learning and AI dependence.

Particularly, for individuals with strong self-regulated learning abilities, AIGC tools function as “assistants.” They can critically assess AI-generated content and integrate it into their learning processes (Bo et al., 2024). Then again, for individuals with weaker self-regulated learning abilities, AIGC tools may become “substitutes,” where they specifically acknowledge and accept AI-generated outputs without engaging in deeper thinking (Fodouop Kouam & Muchowe, 2024). This difference is especially evident among frequent AIGC users: those with strong abilities use the tools to enhance learning efficiency. In contrast, those with weaker abilities are more likely to develop a reliance.

Based on the above analysis, this study proposes the following research hypothesis:

H4: Among frequent AI users, AIGC usage mediates the relationship between self-regulated learning and AI dependence.

To summarize, the hypothesis model of this study is illustrated in Figure 1.

Figure 1.

Hypothetical model.

Justification for Demographic Variables

In addition to the core hypotheses, this study aims to examine the heterogeneous effects of the proposed model across different user groups. Consistent with literature focusing on educational technology adoption and self-regulation, gender and grade level were included as key demographic variables for subsequent difference analysis.

Gender is selected based on extensive empirical evidence suggesting differences in Self-Regulated Learning (SRL) strategies and perceptions of technology risk. Prior studies indicate that females often demonstrate higher proficiency in metacognitive control and process orientation, which may lead to more strategic (instrumental) use of AIGC. At the same time, males are often more driven by efficiency and susceptible to shortcut-taking behaviors (Venkatesh & Morris, 2000). Examining gender disparities is crucial for developing targeted intervention strategies against AI dependence.

Grade level (Academic Experience) is included because it is a proxy for accumulated academic maturity and subject-specific knowledge. As students progress through their academic years (e.g., Freshman to Senior), their SRL abilities are expected to improve, and their needs shift from course-learning support to advanced research and academic support. Analyzing grade differences helps reveal how maturity influences the progression from mere “tool usage” to “tool empowerment.”

Research Methodology

Research Participants

This study was conducted in November 2024 at universities in Fujian Province. The participants were university students. To ensure the representativeness of the sample, participants were recruited from five different types of higher education institutions: a “Double First-Class” university, two provincially key universities, and two regular undergraduate institutions. The survey was distributed via the online questionnaire platform “Wenjuanxing,” where participants responded using their mobile phones or computers. A total of 730 questionnaires were collected. After excluding responses with an abnormally short reply time of less than 4 s, responses with identical answers for all items, and responses from participants who did not frequently use AI tools, 367 valid questionnaires remained. Among the participants, 140 were male, and 227 were female; 85 were first-year students, 123 were second-year students, 83 were third-year students, and 76 were fourth-year students. In total, 261 participants reported using AIGC tools every day, and 106 participants reported using them frequently but not daily.

Measurement Instruments

AIGC Usage Questionnaire

This study focused on college students’ use of AIGC tools in learning contexts. Based on the AIGC usage questionnaire designed by Li et al. (2024), we removed options unrelated to academic scenarios and integrated statements extracted from literature on the application of AIGC in different learning settings. After refinement and processing, we developed a questionnaire comprising 13 evaluation items. To assess the readability and suitability of the questionnaire items, we conducted a pilot survey with 106 university students before the formal data collection. Based on the pilot feedback, we revised the wording of several items and confirmed the usability of the questionnaire. Subsequently, to confirm the effectiveness of these factors, we conducted an exploratory factor analysis (EFA), extracted common factors using the main component analysis, and applied Varimax rotation to obtain factors for each item. In this process, we removed three items with a load below 0.500, resulting in one last set of 10 valid evaluation items (see Table 1).

Table 1.

Mean, Standard Deviation, and Factor Loadings of AIGC Usage Questionnaire.

AIGC Usage Questionnaire Items	Mean	SD	Course learning	Research and academic	Autonomous Learning
1. I use AIGC tools to solve difficult problems in my courses.	3.6	0.911	0.820	0.166	0.080
2. I use generative artificial intelligence to simulate exam questions or generate review materials.	3.46	0.993	0.889	0.049	0.072
3. I use generative artificial intelligence to search for information related to course content.	3.38	1.046	0.857	0.085	0.068
4. I use generative artificial intelligence to assist in completing course assignments.	3.65	0.980	0.715	0.220	0.082
5. I use generative artificial intelligence to assist in selecting research topics.	3.38	1.068	0.118	0.881	0.015
6. I use generative artificial intelligence to assist in writing and revising papers.	3.28	1.007	0.134	0.789	0.135
7. I use generative artificial intelligence to extract key information from literature or translate foreign materials.	3.46	1.085	0.158	0.736	0.144
8. I use generative artificial intelligence to create study plans or schedules	3.06	0.960	0.089	0.110	0.814
9. I use generative artificial intelligence to recommend additional learning resources (e.g., books, papers).	3.04	0.843	0.062	0.167	0.829
10. I use generative artificial intelligence to generate practice questions for autonomous learning.	3.19	0.825	0.088	0.022	0.855

Note. Factor loadings ≥ 0.50 are shown in bold.

Based on the characteristics of college students’ learning settings and the implicit meanings of factor items, we categorized these 10 items into three common factors: Course Learning Support, Research and Academic Assistance, and Autonomous Learning Management. Specifically, the Course Learning Support dimension focuses on students’ use of AIGC in classroom learning under teacher guidance, encompassing four key items: answering questions, completing assignments, retrieving information, and preparing for exams. The Research and Academic Assistance dimension refers to students’ use of AIGC tools in research activities, encompassing research topic selection, academic writing, and literature analysis (three items). The Autonomous Learning Management dimension relates to students’ independent learning outside the classroom and research contexts, including planning, resource recommendations, and problem generation (three items). These dimensions effectively reflect the different scenarios in which college students utilize AIGC tools.

The questionnaire employs a five-point Likert scale, ranging from 1 (“never use”) to 5 (“frequently use”), where higher scores indicate more frequent use of the AIGC tool. The mean, standard deviation, and factor loadings for each item are presented in Table 1.

To further verify the reliability of the questionnaire, we calculated Cronbach’s Alpha coefficient, which yielded an overall Alpha coefficient of .821. The Alpha coefficients for each dimension were: Course Learning Support (0.803), Research and Academic Assistance (0.812), and Autonomous Learning Management (0.795). These coefficients indicate good internal consistency.

Additionally, a confirmatory factor analysis (CFA) was conducted, and the results showed good model fit indices: χ²/df = 2.378, GFI = 0.961, CFI = 0.968, RMSEA = 0.062, confirming good structural validity. Hence, this questionnaire demonstrates strong reliability and validity, effectively distinguishing AIGC usage behaviors across different learning scenarios, making it suitable for measuring AIGC usage in this study.

Self-Regulated Learning Measurement Questionnaire

This study adopted the Self-Regulated Learning Questionnaire developed by Gaumer Erickson and Noonan (2018). The scale consists of 22 Likert-type items categorized into four dimensions: Planning, Monitoring, Control, and Reflection. Responses are rated on a five-point Likert scale, ranging from 1 (“completely disagree”) to 5 (“completely agree”). The specific questionnaire items are provided in Appendix 1.

In this study, the Cronbach’s Alpha coefficient for the scale was .935. The Alpha coefficients for the four dimensions were .877, .851, .858, and .835, respectively. The KMO coefficient was 0.944. Confirmatory factor analysis (CFA) results indicated good model fit indices: χ²/df = 1.376, GFI = 0.945, CFI = 0.984, RMSEA = 0.032.

College Students’ AI Dependence Questionnaire

The tendency for college students to be dependent on AI was measured using six items developed by Andreassen et al. (2012), which have been validated in previous studies (Hu et al., 2023; Lee-Won et al., 2015; Zhang et al., 2024). The questionnaire includes statements such as “I have tried to reduce my use of AI for tasks or assignments, but I have not been successful.” Responses are rated on a 5-point Likert scale, ranging from 1 (“completely disagree”) to 5 (“completely agree”). Higher scores indicate a stronger tendency toward AI dependence. The specific questionnaire items are provided in Appendix 2.

In this study, the Cronbach’s Alpha coefficient of the scale was .801. CFA results showed good model fit indices: χ²/df = 1.700, GFI = 0.986, CFI = 0.989, RMSEA = 0.044.

Data Analysis Process

After exporting the survey data from the online platform, statistical analysis was performed using SPSS 24.0 and AMOS 24.0. The analysis process included:(1) Conducting t-tests and ANOVA to examine differences in core variables across gender, grade level, and other demographic groups. (2) Performing correlation analysis to explore relationships among self-regulated learning, AIGC usage, and AI dependence. (3) Utilizing multiple regression analysis to further test the impact of self-regulated learning and AIGC usage on AI dependence. (4) Constructing a structural equation model (SEM) using AMOS 24.0 to verify the path relationships and mediating effects among self-regulated learning, AIGC usage, and AI dependence.

Through these multi-level statistical analyses, the study aims to comprehensively reveal the internal relationships and mechanisms among the key variables.

Results

Group Differences Analysis

According to the results of gender and grade-level differences analysis (Table 2), significant differences were observed in core variables:

Gender Differences: Female students scored significantly higher than male students in the Autonomous Learning dimension of AIGC usage (e.g., planning, resource recommendations; M_female = 3.172, M_male = 2.971, t = −2.622, p < .01). Conversely, AI dependence scores were significantly lower for females than males (M_female = 3.224, M_male = 3.385, t = 2.252, p < .05). This suggests that females are more inclined to use AIGC tools for self-regulated learning and exhibit a lower psychological dependence on AI.

Grade-Level Differences: Higher-grade students (fourth-year) demonstrated significantly stronger self-regulated learning abilities than lower-grade students (Fourth-year: M = 3.651, First-year: M = 3.328, F = 9.420, p < .001). Additionally, AI dependence decreased with increasing grade level (First-year: M = 3.406, Fourth-year: M = 3.160, F = 5.508, p < .001). Fourth-year students also reported significantly higher AIGC usage frequency in the Research and Academic Assistance dimension than first-year students (M = 3.645, SD = 1.05, M = 3.141, SD = 1.05, p < .001). This indicates that as self-regulated learning skills improve, students become less dependent on AI tools while utilizing them more effectively in academic research.

Table 2.

Mean Differences in Core Variables by Gender and Grade Level.

		Course learning			Research and academic			Autonomous learning			AI dependency			Self-regulated learning
Demographics factors		M ± SD	t	Post hoc multiple comparison (LSD)	M ± SD	t	Post hoc multiple comparison (LSD)	M ± SD	t	Post hoc multiple comparison (LSD)	M ± SD	t	Post hoc multiple comparison (LSD)	M ± SD	t	Post hoc multiple comparison (LSD)
Gender	Male (n = 140)	3.5143 ± 0.7591	−0.170		3.4048 ± 0.8527	0.525		2.9714 ± 0.6744	−2.622**		3.3845 ± 0.6823	2.252*		3.3885 ± 0.5734	−1.155
Gender	Female (n = 227)	3.5293 ± 0.8578	−0.170		3.3558 ± 0.8749	0.525		3.1718 ± 0.7670	−2.622**		3.2239 ± 0.6517	2.252*		3.4612 ± 0.5928	−1.155
Grade level	First-year①(n = 85)	3.6088 ± 0.8096	1.077		3.1412 ± 1.0818	4.792**	④>②④>①③>①	3.1451 ± 0.8333	0.797		3.4059 ± 0.7054	5.508***	①>③①>④②>③②>④	3.3283 ± 0.6541	9.42***	④>①④>②③>①③>②
	Second-year ② (n = 123)	3.5488 ± 0.8289			3.3415 ± 0.7747			3.1274 ± 0.5474			3.4051 ± 0.5322			3.2786 ± 0.4590
	Third-year ③(n = 83)	3.3904 ± 0.6744			3.4150 ± 0.4768			3.1004 ± 0.8260			3.0984 ± 0.7199			3.5716 ± 0.5526
	Fourth-year④(n = 76)	3.5329 ± 0.9525			3.6447 ± 0.9908			2.9825 ± 0.7961			3.1601 ± 0.7021			3.6510 ± 0.6306

p < .05, **p < .01, ***p < .001.

Correlation and Regression Analysis of Self-Regulated Learning, AIGC Use, and AI Dependence

Correlation Analysis

The correlation analysis (Table 3) shows:

Table 3.

Correlation Analysis of Self-Regulated Learning, AIGC Use, and AI Dependence.

Variables	M	SD	1	2	3	4	5
Self-regulated learning	3.4335	0.5857	1
AIGC usage (course learning)	3.5236	0.8205	−.196**	1
AIGC usage (research and academia)	3.3745	0.8656	−.219**	.248**	1
AIGC usage (autonomous learning)	3.0954	0.7386	−.194**	.104**	.216**	1
AI dependence	3.2852	0.6672	−.518**	.379**	.323**	.320**	1

p < .01.

Self-regulated learning is significantly and negatively correlated with AI dependence (r = −.518, p < .01), confirming H1, which suggests that the stronger the self-regulated learning ability, the lower the level of AI dependence.

AIGC use in different contexts (course learning, academic research, and autonomous learning) is significantly and positively correlated with AI dependence (r = 0.320–0.379, p < .01), supporting H2, which indicates that frequent use of AIGC tools exacerbates AI dependence. Among these, the correlation between course learning and AI dependence is relatively higher, and its mean score is also the highest, indicating that university students consider AI tools the most important resource for course learning.

Self-regulated learning is significantly negatively correlated with AIGC use in course learning (r = −.196), academic research (r = −.219), and autonomous learning (r = −.194), all at p < .01, suggesting that students with higher self-regulated learning abilities rely less on AIGC tools, supporting H3.

Regression Analysis

To further examine the relationship between self-regulated learning, AIGC use, and AI dependence, as well as the predictive power of each variable on AI dependence, a multiple regression analysis was conducted. As shown in Table 4:

Table 4.

Regression Analysis Results (AI Dependence as the Dependent Variable).

Predictor variables	βcoefficient	Standard error (SE)	t-Value	p-value	R ²	Adjusted R²
Self-regulated learning	−0.477	0.048	−9.879	0.000	.413	.407
AIGC usage (course learning)	0.210	0.035	4.148	0.000
AIGC usage (research and academia)	0.086	0.034	2.537	.0120
AIGC usage (autonomous learning)	0.146	0.038	5.452	0.000

Self-regulated learning has the strongest negative predictive effect on AI dependence (β = −.477, p < .001). AIGC usage in course learning (β = .210, p < .001), academic research (β = .086, p = .012 < .05), and autonomous learning (β = −.146, p < .001) significantly positively predict AI dependence.

Structural Equation Model Test

This study employed the structural equation modeling (SEM) method to test the hypothesized model. After modifications and adjustments, the model fit indices were as follows: χ²/df = 1.291, RMSEA = 0.028, and goodness-of-fit indices GFI = 0.902, CFI = 0.971, IFI = 0.969, and TLI = 0.969, all above 0.90, indicating good model fit and reliability (see Appendix 3). The coefficient test results (see Table 5 and Figure 2) confirm that hypotheses H1, H2, and H3 hold. Specifically, among students who frequently use AI tools, self-regulated learning negatively affects both AIGC use and AI dependence, while AIGC use positively affects AI dependence.

Table 5.

Structural Equation Model Path Coefficients.

Hypothesis	Path	Non-standardized coefficients	Standardization coefficient	S.E.	C.R.	p
H1	Self-Regulated Learning →AIGC Usage	−0.334	−0.486	0.068	−4.884	***
H2	AIGC Usage → AI Dependence	0.868	0.656	0.196	4.428	***
H3	Self-Regulated Learning →AI Dependence	−0.232	−0.255	0.079	−2.930	**

p < .01, ***p < .001.

Figure 2.

Structural equation model path test.

Mediation Effect Test

To examine the mediating effect of AIGC use in the relationship between self-regulated learning and AI dependence, this study adopted the Bootstrap method for validation. In Amos software, 2,000 resampling iterations were set with a 95% confidence level to estimate the confidence intervals of standardized correlation coefficients (Hayes, 2009). Amos provides two methods for confidence interval estimation: the Bias-Corrected Percentile Method and the Percentile Method. The estimation results of both methods are presented in Table 6.

Table 6.

Mediation Effect Test Results.

Effect types	Effect	Standard errors	Bootstrapping
			Bias-corrected		Percentile		Two-tailed significance
			Lower	Upper	Lower	Upper
Direct effect	−0.232	0.096	−0.389	−0.050	−0.391	−0.059	.0170 (*)
Indirect effect	−0.290	0.091	−0.514	−0.181	−0.464	−0.168	.001(***)
Total effect	−0.522	0.078	−0.693	−0.385	−0.685	−0.381	.001(***)

Note.*p < .05, ***p < .001. Bootstrap confidence intervals based on 5,000 resamples. BC = Bias-Corrected.

According to the Bias-Corrected Percentile Method, the confidence interval for the direct effect of self-regulated learning on AI dependence is (−0.389, −0.050), which does not include 0, indicating a significant direct effect. Additionally, the confidence interval for the indirect effect of self-regulated learning on AI dependence through AIGC use is (−0.514, −0.181), also excluding 0, confirming the partial mediating role of AIGC use. The Percentile Method yielded consistent results, further verifying the mediation effect.

Specifically, the direct effect of self-regulated learning on AI dependence is −0.232, while the mediating effect of AIGC use (indirect effect) is −0.290, resulting in a total effect of −0.522 (the sum of the direct and indirect effects). Among them, the direct effect accounts for 44.4% of the total effect, and the indirect effect accounts for 55.6%. These findings suggest that self-regulated learning not only directly reduces AI dependence but also indirectly mitigates it by decreasing the frequency of AIGC use, with the indirect effect playing a more substantial role.

In summary, the mediation effect analysis confirms that self-regulated learning can indirectly influence AI dependence through the use of AIGC, supporting Hypothesis H4. This finding further reveals the dual mechanism through which self-regulated learning reduces AI dependence: directly decreasing dependence tendencies and indirectly reducing AI dependence by lowering the frequency of AIGC use.

Discussion

This study reveals the dynamic mediating role of self-regulated learning (SRL) in the relationship between AIGC usage and AI dependence. Through an analysis of group differences, it further deepens the understanding of the heterogeneity of technological dependence. The results provide dual insights for educational practice in the era of intelligent technology. While taking advantage of the potential for AIGC tools, it is also important to be aware of the risks of AI dependence they may induce. The following discussion integrates the research conclusions with theoretical frameworks.

The Dual Inhibitory Mechanism of Self-Regulated Learning

This study found that self-regulated learning (SRL) inhibits AI dependence through both direct and indirect pathways, with a direct effect coefficient of β = −.255. In contrast, an indirect effect is generated through the reduction in AIGC usage frequency, accounting for 55.6% of the total mediating effect. This dual inhibitory mechanism can be explained from the perspectives of cognitive control theory and social cognitive theory.

From the perspective of cognitive control theory, learners with high SRL ability possess stronger cognitive resource management skills (Cole et al., 2017). Cognitive control, as a crucial function of the prefrontal cortex (Braver, 2012), enables learners to dynamically adjust their thinking and behavior in response to task demands. Research indicates that high-SRL learners can effectively avoid excessive consumption of cognitive control resources, thereby reducing passive dependence on AI tools (Shenhav et al., 2013). Specifically, they can proactively set boundaries for AI usage, preventing AI from replacing deep cognitive processes (Billman, 2024; D. Wu et al., 2024). In contrast, learners with low SRL ability often mismanage cognitive resources when facing complex tasks, leading to excessive reliance on AI-generated content.

Furthermore, social cognitive theory (SCT) posits that individual learning behavior is affected by the interaction of personal competence, behavioral strategies, and environmental factors (Bandura, 1986). In AI-enhanced learning environments, high-SRL learners exhibit greater control over their learning processes. This control is reflected in three aspects: first, they maintain learning autonomy by setting goals (e.g., prioritizing tasks) and adjusting strategies (e.g., balancing AI tool usage with independent thinking; Bandura, 2012); second, they effectively employ metacognitive strategies, ensuring that they remain the primary agents in the learning process while using AI tools (Baskara, 2024; Vorecol Editorial Team, 2024); third, they engage in reflective thinking to regulate the scope of AI’s role while improving their self-directed learning abilities (D. Wu et al., 2024).

More specifically, the difference in SRL ability leads to significant variations in how learners perceive AI tools. Low SRL learners tend to view AI as a tool to compensate for their learning abilities, which can help increase their confidence in their learning. On the other hand, high-content learners clearly identify AI’s role as a support tool, using it only when necessary to support their decisions, while maintaining autonomy in basic cognitive activities (Rodríguez-Ruiz et al., 2025). This difference eventually leads to a significant difference in the development of long-term learning capacity between the two groups of learners (Shi et al., 2025).

Furthermore, it should be emphasized that the measurement of AIGC use in this study mainly captures usage amount/usage frequency, rather than the’ quality of learners “strategic tool use. Consequently, the inhibitory mechanism of SRL observed in this study is primarily reflected in the reduction of unnecessary or substitutive tool use. It should not be interpreted as a decrease in learners” ability or willingness to use AIGC effectively. In other words, high-SRL learners display selective and disciplined use of AIGC tools rather than any decline in tool-use effectiveness.

In summary, self-regulated learning has a dual inhibitory effect on AI dependence, operating through mechanisms at both cognitive control and social cognitive levels. This effect is not only evident in immediate AI usage behaviors but also has significant implications for learners’ long-term skill development.

The Heterogeneous Effects of AIGC Usage Scenarios

This study confirms a positive correlation between AIGC usage frequency and AI dependence (β = .656), but the effect intensity is significantly influenced by task type.

In course learning support scenarios, AIGC becomes a major source of AI dependence risk (r = .379) due to its ability to meet high-frequency, rigid demands, especially in tasks such as assignment generation, answer retrieval, and knowledge summarization. This trend is consistent with the previous findings, which indicate that students with low SRL abilities tend to use AI tools for coursework to directly obtain answers rather than as an auxiliary resource, thereby weakening their critical thinking skills (Bulawan et al., 2024; Gerlich, 2025).

In contrast, in research and academic support scenarios (e.g., literature analysis), the “usage-dependence” relationship of AIGC varies by academic year. Senior students, despite having a higher usage frequency (senior year M = 3.645), exhibit lower dependence (M = 3.160), which is closely related to their improved SRL ability (M = 3.651). With the accumulation of academic experience and professional training, senior students develop greater advantages in task identification, critical evaluation, and the application of metacognitive strategies (Agudo-Saiz et al., 2021). They can integrate AI into the research process—such as cross-verifying AI-generated conclusions and strategically utilizing data analysis modules—rather than relying on AI for decision-making, thereby advancing from mere “tool usage” to “tool empowerment.”

Unlike previous studies (Ezeoguine & Eteng-Uket, 2024), this study identifies significant gender differences in AIGC usage in autonomous learning scenarios. Female students use AIGC tools more frequently in autonomous learning (M_female = 3.172 vs. M_male = 2.971) but exhibit lower AI dependence than males (M_female = 3.224 vs. M_male = 3.385). This phenomenon suggests that female students are more inclined to use AIGC tools as a resource integration and decision-support mechanism, emphasizing content selection and goal alignment—an approach reflecting “instrumental rationality.” This pattern aligns with research indicating that female learners generally demonstrate superior metacognitive strategies, such as planning and monitoring (Virtanen & Nevgi, 2010), which enables them to maintain cognitive control while leveraging the tool.

In contrast, male students are more susceptible to the “efficiency substitution” trap, relying on AI to generate direct answers while neglecting independent thinking and the internalization of knowledge. This finding is consistent with research indicating that male students are often more strongly driven by perceived usefulness (efficiency) in technology adoption (Venkatesh & Morris, 2000) and are reported to exhibit higher tendencies toward academic shortcut-taking behaviors (Hensley et al., 2013).

Theoretical and Practical Contributions

The “SRL→ AIGC usage→AI dependence” model constructed in this study by revealing the dynamic relationship between AIGC tool usage and dependence risk: when tool usage lacks the support of SRL abilities, perceived usefulness may turn into a dependence-inducing factor; however, at high SRL levels, tool usefulness transforms into an efficiency gain, safeguarding students’ autonomy.

Additionally, the study highlights the “double-edged sword” effect of AIGC tool usage: when used appropriately, these tools enhance technological literacy, but excessive reliance may weaken cognitive abilities. Future research should focus on differentiating between “tool capability” and “usage intensity” in AI dependence formation.

At the practical level, this study emphasizes the importance of maintaining a “human-centered” educational philosophy in the intelligent technology era. AIGC tools should be positioned as cognitive “scaffolding” rather than decision substitutes, providing empirical evidence for building an “empowering but not replacing” educational ecosystem. Specific recommendations include embedding metacognitive training modules (such as setting up “AI-Free Days”) into AI tool training programs to cultivate students’ deep thinking and innovation capabilities. Additionally, differentiated tool usage guidelines should be established for various tasks, such as academic research and coursework, and AI dependence should be incorporated into student evaluation systems through dynamic assessment mechanisms for timely intervention.

Moreover, this study warns that technological dependence may exacerbate educational inequality, as students with low SRL are prone to falling into a vicious cycle of “weakened cognitive ability → deepened technological dependence.” To address this, educational institutions should develop an inclusive technological framework that implements digital mentoring systems to provide customized metacognitive training for students with low Self-Regulated Learning (SRL) skills. Furthermore, assessment models should incorporate critical reflections on the “human-AI” collaboration process to prevent technological capital from becoming a new stratification criterion.

Limitations and Future Research

This study has the following limitations. First, the sample was drawn solely from universities in Fujian Province through convenience sampling, which may limit the external validity of the findings. Given the global penetration of AIGC tools, a single-region sample may not fully reflect differences in cultural context, technological accessibility, academic norms, or learning practices. Additionally, the data rely on self-reported measures that may be subject to social desirability bias. Moreover, the study does not differentiate between types of AIGC tools (e.g., writing assistants vs. data analysis tools), making it difficult to assess how specific functionalities contribute to AI dependence.

Future studies should include cross-regional, cross-cultural, and multidisciplinary samples to enhance the generalizability of the findings. Moreover, integrating self-report data with behavioral logs may improve measurement accuracy. It is also recommended to examine the differentiated effects of various AIGC functions and to clarify the roles of “tool capability” and “usage intensity” in shaping dependence risk. Based on specific learning tasks, more refined intervention strategies should be designed to strike a balance between technological efficiency and learner autonomy, thereby promoting educational equity and holistic student development.

Conclusion

This study empirically finds that self-regulated learning (SRL) is a key factor in mitigating AI dependence, exerting a dual inhibitory effect: directly reducing dependence tendencies (β = −.255) and indirectly minimizing unnecessary AIGC tool usage, with a mediating effect accounting for 55.6%.

Among different usage scenarios, course learning emerges as the primary risk source of dependence due to its high-frequency rigid demand (r = .379), while senior students, through critical evaluation and metacognitive strategies, achieve “tool empowerment” in research settings, effectively balancing tool efficiency with academic independence (senior SRL mean = 3.651, AI dependence mean = 3.160). Additionally, gender differences indicate that female students integrate AIGC tools into autonomous learning more effectively and exhibit lower dependence levels (M_female = 3.224 vs. M_male = 3.385).

Limitations include the study’s geographically restricted sample and the lack of differentiation among AIGC tool types. Future research should broaden cross-cultural and multidisciplinary sampling and investigate the distinct impacts of various AIGC functionalities on the formation of dependence.

From an educational practice perspective, this study advocates for a dynamic evaluation framework that incorporates AI dependence into competency assessments. By embedding metacognitive training and developing differentiated tool usage guidelines for different learning contexts, the sustainable development of “human-AI” collaboration can be promoted.

Footnotes

Appendix

ORCID iD

Bing Liu

Ethical Considerations

The authors declare that they have no relevant or material financial interests that relate to the research described in this paper.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the 2023 Fujian Social Science Project — General Project (No. FJ2023B022).

Declaration of Conflicting Interests

The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author.*

References

Agudo-Saiz

Salcines-Talledo

González-Fernández

(2021). Critical thinking in compulsory secondary education and upper secondary education: Perspectives of teachers and students from a secondary school in a province in northern Spain. Tendencias Peda-gógicas, 37, 121–133. https://doi.org/10.15366/tp2021.37.010

Andreassen

C. S.

Torsheim

Brunborg

G. S.

Pallesen

(2012). Development of a Facebook addiction scale. Psychological Reports, 110(2), 501–517. https://doi.org/10.2466/02.09.18.pr0.110.2.501-517

Ansell

E. B.

Tuit

Sinha

(2012). Effects of cumulative stress and impulsivity on smoking status. Human Psychopharmacology Clinical and Experimental, 27(2), 200–208. https://doi.org/10.1002/hup.1269

Bandura

(1986). Social foundations of thought and action: A social cognitive theory. Prentice Hall.

Bandura

(2012). On the functional properties of perceived self-efficacy revisited. Journal of Management, 38(1), 9–44. https://doi.org/10.1177/0149206311410606

Baskara

F. R.

(2024). Fostering metacognitive skills through AI-human dialogues: A new paradigm for language learning. Proceeding of International Conference on Language Pedagogy (ICOLP), 4(1), 140–151. https://doi.org/10.24036/icolp.v4i1.207

Billman

(2024, November). Impact analysis of the use of artificial intelligence with self-regulated learning theory on student academic performance in Indonesia [Conference session]. 2024 International Conference on Computer Engineering, Network, and Intelligent Multimedia (CENIM) (pp. 1–5). IEEE. https://10.1109/CENIM64038.2024.10882800.

J. G.

Qiao

Y. N.

Liu

Huang

(2024). Exploring the potential and challenges of AIGC technology in university programming courses. Journal of Computer Technology and Development, 34(6), 214–220. https://doi.org/10.20165/j.cnki.ISSN1673-629X.2024.0085

Braver

T. S.

(2012). The variable nature of cognitive control: A dual mechanisms framework. Trends in Cognitive Sciences, 16(2), 106–113. https://doi.org/10.1016/j.tics.2011.12.010

10.

Bulathwela

Pérez-Ortiz

Holloway

Cukurova

Shawe-Taylor

(2024). Artificial intelligence alone will not democratise education: On educational inequality, techno-solutionism and inclusive tools. Sustainability, 16(2), 781. https://doi.org/10.3390/su16020781

11.

Bulawan

A. A.

Tengco

R. L.

Barcenas

Primero

Convento

(2024). The use of generative AI in learning and its influence on students’ academic engagement in Noveleta Senior High School. International Journal of Advanced Multidisciplinary Research and Studies, 4(3), 461–469. https://doi.org/10.62225/2583049x.2024.4.3.2809

12.

Çela

Fonkam

M. M.

Potluri

R. M.

(2024). Risks of AI-assisted learning on student critical thinking: A case study of Albania. International Journal of Risk and Contingency Management, 12(1), 1–19. https://doi.org/10.4018/ijrcm.350185

13.

Chai

S. M.

Z. H.

Z. K.

(2024). Public attention and cognition of AIGC in education: A study based on online Q&A communities. Modern Educational Technology, 34(10), 42–52. https://doi.org/10.3969/j.issn.1009-8097.2024.10.005

14.

Choi

Cho

Y. M.

Ryu

H. K.

(2022). Structural relationship between mothers’ emotional expressiveness, young children’s self-regulation, and smart device overdependence tendency. Korean Journal of Child Studies, 43(2), 111–123. https://doi.org/10.5723/kjcs.2022.43.2.111

15.

Cole

M. W.

Braver

T. S.

Meiran

(2017). The task novelty paradox: Flexible control of inflexible neural pathways during rapid instructed task learning. Neuroscience and Biobehavioral Reviews, 81, 4–15. https://doi.org/10.1016/j.neubiorev.2017.02.009

16.

Deng

G. M.

X. F.

Zhu

Y. H.

(2021). Latent profile analysis and behavioral process mining of learners’ online self-regulated learning in blended learning environment. Journal of E-education Research, 42(1), 80–86. https://doi.org/10.13811/j.cnki.eer.2021.01.011

17.

Dergaa

Ben Saad

Glenn

J. M.

Amamou

Ben Aissa

Guelmami

Fekih-Romdhane

Chamari

(2024). From tools to threats: A reflection on the impact of artificial-intelligence chatbots on cognitive health. Frontiers in Psychology, 15, 1259845. https://doi.org/10.3389/fpsyg.2024.1259845

18.

Eckley

Allen

Millear

Rune

K. T.

(2023). COVID-19’s impact on learning processes in Australian university students. Social Psychology of Education, 26(1), 161–189. https://doi.org/10.1007/s11218-022-09739-x

19.

Ezeoguine

E. P.

Eteng-Uket

(2024). Artificial intelligence tools and higher education student’s engagement. Edukasiana: Jurnal Inovasi Pendidikan, 3(3), 300–312. https://doi.org/10.56916/ejip.v3i3.733

20.

Fatimah

Rajiani

S. I.

Abbas

E. W.

(2021). Cultural and individual characteristics in adopting computer-supported collaborative learning during covid-19 outbreak: Willingness or obligatory to accept technology? Management Science Letters, 11(2), 373–378. https://doi.org/10.5267/j.msl.2020.9.032

21.

Fodouop Kouam

A. W.

Muchowe

R. M

. (2024). Exploring graduate students’ perception and adoption of AI chatbots in Zimbabwe: Balancing pedagogical innovation and development of higher-order cognitive skills. SSRN 4748274. https://doi.org/10.37074/jalt.2024.7.1.12.

22.

Gaumer Erickson

A. S.

Noonan

P. M

. (2018). Self-regulation formative questionnaire. In Gaumer Erickson

A. S.

Noonan

P. M.

(Eds.), The skills that matter: Teaching interpersonal and intrapersonal competencies in any classroom (pp. 177–178). Corwin.

23.

Gerlich

(2025). AI tools in society: Impacts on cognitive offloading and the future of critical thinking. Societies, 15(1), 6. https://doi.org/10.3390/soc15010006

24.

Hamilton

K. R.

Ansell

E. B.

Reynolds

Potenza

M. N.

Sinha

(2013). Self-reported impulsivity, but not behavioral choice or response impulsivity, partially mediates the effect of stress on drinking behavior. Stress, 16(1), 3–15. https://doi.org/10.3109/10253890.2012.671397

25.

Hayes

A. F.

(2009). Beyond Baron and Kenny: Statistical Mediation Analysis in the New Millennium. Communication Monographs, 76(4), 408–420. https://doi.org/10.1080/03637750903310360

26.

Hensley

L. C.

Kirkpatrick

K. M.

Burgoon

J. M.

(2013). Relation of gender, course enrollment, and grades to distinct forms of academic dishonesty. Teaching in Higher Education, 18(8), 895–907.

27.

Huang

Lai

Wang

Zhao

Dai

Wang

(2024). AI technology panic—is AI dependence bad for mental health? A cross-lagged panel model and the mediating roles of motivations for AI use among adolescents. Psychology Research and Behavior Management, 17, 1087–1102. https://doi.org/10.2147/prbm.s440889

28.

Mao

Kim

K. J.

(2023). How social anxiety leads to problematic use of conversational AI: The roles of loneliness, rumination, and mind perception. Computers in Human Behavior, 145, 107760. https://doi.org/10.1016/j.chb.2023.107760

29.

Ilić

Mikić

Kopanja

Vesin

(2023). Intelligent techniques in e-learning: A literature review. Artificial Intelligence Review, 56(12), 14907–14953. https://doi.org/10.1007/s10462-023-10508-1

30.

Junaštíková

(2024). Self-regulation of learning in the context of modern technology: A review of empirical studies. Interactive Technology and Smart Education, 21(2), 270–291. https://doi.org/10.1108/itse-02-2023-0030

31.

Karran

A. J.

Charland

Martineau

J. T.

de Guinea

A. O.

Lesage

A. M.

Senecal

Leger

P. M.

(2024). Multi-stakeholder perspective on responsible artificial intelligence and acceptability in education. arXiv preprint arXiv:2402.15027. https://doi.org/10.48550/arXiv.2402.15027.

32.

Lee-Won

R. J.

Herzog

Park

S. G.

(2015). Hooked on Facebook: The role of social anxiety and need for social assurance in problematic use of Facebook. Cyberpsychology Behavior and Social Networking, 18(10), 567–574. https://doi.org/10.1089/cyber.2015.0002

33.

Fan

Cheng

(2023). AIGC in China: Current developments and future outlook. arXiv preprint arXiv:2308.08451. https://doi.org/10.1109/OJCS.2023.3300321.

34.

Jia

C. Y.

Zhai

(2024). Investigation of college students’ Generative Artificial Intelligence (GAI) usage status and its implication: Taking Zhejiang University as an example. Open Education Research, 30(1), 89–98. https://doi.org/10.13966/j.cnki.kfjyyj.2024.01.010

35.

Liu

Cheng

(2024). Cognitive risks of tool dependence — The myth of instrumentality and creativity of ChatGPT. Youth Journalist, 84(03), 17–22. https://link.oversea.cnki.net/doi/10.15997/j.cnki.qnjz.2024.03.010

36.

Liu

Ding

Peng

Zhang

(2024). Leveraging ChatGPT to optimize depression intervention through explainable deep learning. Frontiers in Psychiatry, 15, 1383648. https://doi.org/10.3389/fpsyt.2024.1383648

37.

Pan

(2024). A study on teachers’ willingness to use generative AI technology and its influencing factors: Based on an integrated model. Sustainability, 16(16), 7216. https://doi.org/10.3390/su16167216

38.

Mirmoadi

B. S.

Satwika

Y. W.

(2022). Hubungan Antara Literasi Digital Dengan self regulated learning Pada Mahasiswa. JDMP (Jurnal Dinamika Manajemen Pendidikan), 7(1), 8–23. https://doi.org/10.26740/jdmp.v7n1.p8-23

39.

Müller

(2022). Domesticating artificial intelligence. Moral Philosophy and Politics, 9(2), 219–237. https://doi.org/10.1515/mopp-2020-0054

40.

Nyaaba

Wright

Choi

G. L.

(2024). Generative AI and digital neocolonialism in global education: Towards an equitable framework. arXiv preprint arXiv:2406.02966. https://doi.org/10.48550/arXiv.2406.02966.

41.

Onesi-Ozigagun

Ololade

Y. J.

Eyo-Udo

N. L.

Ogundipe

D. O.

(2024). Revolutionizing education through AI: A comprehensive review of enhancing learning experiences. International Journal of Applied Research in Social Sciences, 6(4), 589–607. https://doi.org/10.51594/ijarss.v6i4.1011

42.

Pfost

Kuntner

Goppert

S. A.

Hübner

(2023). Self-regulated learning, learner characteristics and relations to webtool usage in higher education. Psychology Learning & Teaching, 22(1), 94–106. https://doi.org/10.1177/14757257221122267

43.

Pintrich

P. R.

(2000). The role of goal orientation in self-regulated learning. Academic Press.

44.

Prasad

Sane

(2024, March). A self-regulated learning framework using generative AI and its application in CS educational intervention design [Symposium]. Proceedings of the 55th ACM Technical Symposium on Computer Science Education V. 1 (pp. 1070–1076). https://doi.org/10.1145/3626252.363082.

45.

Rodríguez-Ruiz

Marín-López

Espejo-Siles

(2025). Is artificial intelligence use related to self-control, self-esteem and self-efficacy among university students? Education and Information Technologies, 30(2), 2507–2524. https://doi.org/10.1007/s10639-024-12906-6

46.

Schreibelmayr

Moradbakhti

Mara

(2023). First impressions of a financial AI assistant: Differences between high trust and low trust users. Frontiers in artificial intelligence, 6, 1241290. https://doi.org/10.3389/frai.2023.1241290

47.

Schunk

D. H.

Zimmerman

(Eds.) (2011). Handbook of self-regulation of learning and performance. Taylor & Francis. https://doi.org/10.4324/9781315697048.

48.

Seufert

Hamm

Vogt

Riemer

(2024). The interplay of cognitive load, learners’ resources and self-regulation. Educational Psychology Review, 36(2), 1–30. https://doi.org/10.1007/s10648-024-09890-1

49.

Shenhav

Botvinick

M. M.

Cohen

J. D.

(2013). The expected value of control: An integrative theory of anterior cingulate cortex function. Neuron, 79(2), 217–240. https://doi.org/10.1016/j.neuron.2013.07.007

50.

Shi

Liu

(2025). Exploring how AI literacy and self-regulated learning relate to student writing performance and well-being in generative AI-supported higher education. Behavioral Sciences, 15(5), 705. https://doi.org/10.3390/bs15050705

51.

Song

Soria Zurita

N. F.

Nolte

Singh

Cagan

McComb

(2022). When faced with increasing complexity: The effectiveness of artificial intelligence assistance for drone design. Journal of Mechanical Design, 144(2), 021701. https://doi.org/10.1115/1.4051871

52.

Ullah

A. I. M. V.

Sreedevi

(2024). Quantitative exploration of high AI usage, self-efficacy, learning performance, and social loafing among young adults in Bangalore. International Journal for Research in Applied Science and Engineering Technology, 12(5), 2051–2056. https://doi.org/10.22214/ijraset.2024.60655

53.

Venkatesh

Morris

M. G.

(2000). Why don’t men ever stop to ask for directions? Gender, social influence, and their role in technology acceptance and usage behavior. MIS Quarterly, 24, 115–139.

54.

Virtanen

Nevgi

(2010). Disciplinary and gender differences among higher education students in self-regulated learning strategies. Educational Psychologist, 30(3), 323–347.

55.

Vorecol Editorial Team. (2024, October 1). The impact of AI on learner autonomy in adaptive learning environments. Vorecol. https://vorecol.com/blogs/blog-the-impact-of-ai-on-learner-autonomy-in-adaptive-learning-environments-197750.

56.

Wang

H. J.

Wei

J. M.

Fei

J. X.

(2024). In-depth problem-solving ability: Conceptual characteristics, theoretical framework and cultivation path—Based on the perspective of AIGC technology empowerment. Chinese E-Education, 45(5), 97–104.

57.

Wang

S. F.

Chen

C. C.

(2024). Exploring designer trust in artificial intelligence-generated content: TAM/TPB model study. Applied Sciences, 14(16), 6902. https://doi.org/10.3390/app14166902

58.

Wang

Pan

Yan

Luan

T. H.

(2023). A survey on ChatGPT: AI–generated contents, challenges, and solutions. IEEE Open Journal of the Computer Society, 4, 280–302. https://doi.org/10.1109/ojcs.2023.3300321

59.

Zhang

Yue

X. G.

Dong

R. K.

(2024). Unlocking potential: Key factors shaping undergraduate self-directed learning in AI-enhanced educational environments. Systems, 12(9), 332. https://doi.org/10.3390/systems12090332

60.

Gan

Chen

Wan

Lin

(2023). Ai-generated content (AIGC): A survey. arXiv preprint arXiv:2304.06632. https://doi.org/10.48550/arXiv.2304.06632

61.

Xie

Chen

Ding

(2025). Can the use of artificial intelligence-generated content bridge the cancer knowledge gap? A longitudinal study with health self-efficacy as a mediator and educational level as a moderator. Cancer Control: Journal of the Moffitt Cancer Center, 32, 10732748251319487.

62.

Yang

Cui

Dai

Zhu

(2025). The impact of human-AIGC tools collaboration on the learning effect of college students: A key factor for future education? Kybernetes, 54, 7746–7763. https://doi.org/10.1108/k-03-2024-0613

63.

Yang

Ding

Chen

(2024). Factors affecting the use of artificial intelligence generated content by subject librarians: A qualitative study. Heliyon, 10(8), e29584. https://doi.org/10.1016/j.heliyon.2024.e29584

64.

Yao

Wang

Sun

(2025). Research on key factors influencing Chinese designers’ use of AIGC: An extension based on TAM and TRI. PLoS One, 20(2), e0314306. https://doi.org/10.1371/journal.pone.0314306

65.

Zhang

Zhao

Zhou

Kim

J. H.

(2024). Do you have AI dependency? The roles of academic self-efficacy, academic stress, and performance expectations on problematic AI usage behavior. International Journal of Educational Technology in Higher Education, 21(1), 34. https://doi.org/10.1186/s41239-024-00467-0

66.

Zheng

L. Q.

Chen

F. Y.

(2016). Empirical research on strategies for cultivating self-regulated learning abilities among college students. China Distance Education (Comprehensive Edition), 36(12), 30–36. https://link.oversea.cnki.net/doi/10.13541/j.cnki.chinade.20161216.003

67.

Zimmerman

B. J.

(1989). A social cognitive view of self-regulated academic learning. Journal of Education & Psychology, 81(3), 329–339. https://doi.org/10.1037/0022-0663.81.3.329

68.

Zimmerman

B. J.

(2000). Attaining self-regulation, A social cognitive perspective. Academic Press.