📄 social science

Multimodal Generative AI Chatbots for Enhancing Scientific Research Skills: A Self-Determination Theory-Based Mediation Model

This study demonstrates that multimodal generative AI chatbots, particularly those using video summaries, significantly enhance scientific research skills among Omani teacher qualification students by reducing cognitive load and fostering autonomy, competence, and relatedness through Self-Determination Theory-based mechanisms.

Original authors: Sami AlMazrouii

Published 2026-10-20

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Sami AlMazrouii

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the modern classroom, the tools for learning are shifting rapidly. Artificial intelligence has moved from a futuristic concept to a practical assistant, capable of generating text, images, and even video summaries of complex topics. But simply having access to smart technology does not guarantee that students will learn better. The real question for educators is how these tools interact with the human mind. Two major ideas help explain this interaction. First, the brain has a limited amount of mental energy available for learning at any one time; if a lesson is too cluttered or confusing, that energy is wasted on figuring out the format rather than understanding the content. Second, people learn best when they feel a sense of control over their learning, when they feel capable of mastering the material, and when they feel supported by their environment. These feelings are not just nice to have; they are the fuel that drives deep understanding.

A recent study conducted in Oman sought to test how these principles work when students use a new kind of AI helper. The researchers focused on a group of teacher-training students who were learning the difficult subject of scientific research methods. This is a field that requires students to design studies, analyze data, and write reports, tasks that demand high levels of concentration and critical thinking. The team wanted to see if an AI chatbot that could summarize information in different ways would help students master these skills. They compared three specific formats: video summaries that combined pictures and narration, static infographics that organized data visually, and audio summaries that relied on sound alone. The goal was to determine which format not only made the information easier to digest but also helped students feel more confident and in control, ultimately leading to better performance on research tasks.

The study involved 308 students who were divided into three groups. Each group used a generative AI chatbot designed to help them with their research methods course, but the way the information was presented differed for each group. One group received video summaries, another received infographics, and the third received audio summaries. The researchers measured how well the students performed on scientific research skills, how much mental effort they felt they had to expend, and how satisfied they felt regarding their sense of control, their competence, and their connection to the learning process. The results were clear and distinct. The students who used the video summaries outperformed the other two groups significantly. They scored the highest on the research skills assessment, and they also reported the highest levels of satisfaction with the quality of the information and the greatest reduction in mental effort.

The video group did not just get better grades; they also felt more capable and more in charge of their learning journey. The study found that the quality of the information provided by the AI and the reduction in mental clutter were key factors. When the AI presented information clearly and reduced the need for students to struggle with confusing formats, the students felt more autonomous and more competent. These feelings, in turn, directly improved their ability to perform complex research tasks. The researchers discovered that the video format was particularly effective because it could show the step-by-step nature of research procedures, combining visual cues with spoken explanations. This helped students follow the logical flow of a study without getting lost. In contrast, while the infographic group performed better than the audio group, the advantage of the infographic over the audio was mostly direct, meaning it helped performance without necessarily boosting the students' psychological feelings of competence and control to the same degree as the video did.

The audio summaries, while flexible and accessible, proved to be the least effective of the three formats for this specific type of learning. Because spoken information is temporary and lacks a persistent visual structure, students found it harder to integrate complex methodological concepts. The study suggests that for tasks requiring the understanding of sequences and relationships, such as designing a research study, a format that allows students to see and hear the process simultaneously is superior. The findings indicate that the success of AI in education is not just about the technology itself, but about how that technology is designed to support the human mind. High-quality summaries that reduce mental strain and foster a sense of capability are essential. The video format succeeded because it aligned with how the brain processes complex, sequential information, allowing students to focus their energy on thinking rather than on decoding the lesson.

This research provides a concrete answer to how AI can be best used in higher education. It is not enough to simply provide students with AI-generated content; the format of that content matters deeply. The study shows that when AI tools are designed to present information in a way that respects the limits of human attention and supports the student's sense of agency, they can significantly enhance learning outcomes. The video summaries acted as a powerful scaffold, helping students navigate the complexities of scientific research with greater confidence and skill. The results suggest that educators and designers should prioritize formats that combine visual and auditory elements to support the most demanding cognitive tasks. By doing so, they can create learning environments where technology does not just deliver information, but actively supports the development of critical thinking and research skills.

Technical Summary: Multimodal Generative AI Chatbots for Enhancing Scientific Research Skills

Problem Statement

Despite the rapid integration of Artificial Intelligence (AI) in education, existing literature remains fragmented regarding the application of Generative AI (GenAI) to higher-order scientific research skills. Most studies focus on general engagement, language learning, or academic writing, often treating chatbots, video summarization, and infographics as independent variables rather than components of an integrated framework. Furthermore, there is a scarcity of empirical research examining these technologies within postgraduate teacher education contexts in the Arab region, specifically Oman. Critical gaps include a lack of comparative analysis between different GenAI-generated summary modalities (video, infographic, audio) and insufficient understanding of the psychological mechanisms—specifically within Self-Determination Theory (SDT)—that mediate the relationship between AI-supported learning environments and the development of scientific research competencies.

Methodology

The study employed a quantitative, posttest-only quasi-experimental design involving 308 students enrolled in a Research Methods course within the Educational Qualification Diploma Program at the Arab Open University in the Sultanate of Oman.

Participants: The sample consisted of 308 Omani nationals (78.9% female, 85.3% aged 20–29), all holding bachelor's degrees. Participants were assigned to three experimental groups: Video Summaries ( $n=98$ ), Infographic Summaries ( $n=104$ ), and Audio Summaries ( $n=106$ ).
Intervention: All groups utilized a Generative AI-enhanced chatbot designed to support research-related learning activities. The instructional content and objectives were identical across groups; the independent variable was the modality of the AI-generated summary provided to the students (video, infographic, or audio).
Variables:
- Independent Variables: Perceived Information and Summary Quality, Cognitive Load Reduction.
- Mediating Variables: Basic Psychological Needs (Autonomy, Competence, Relatedness) derived from Self-Determination Theory (SDT).
- Dependent Variable: Scientific Research Skills (measured via a 16-item researcher-developed instrument).
Data Analysis: The study utilized Partial Least Squares Structural Equation Modeling (PLS-SEM) with 10,000 bootstrap resamples to test the mediation model. Group differences were analyzed using one-way ANOVA (or Welch's ANOVA where variance homogeneity was violated) with post-hoc comparisons. Measurement validity was confirmed through reliability checks (Cronbach's alpha, Composite Reliability), convergent validity (AVE), and discriminant validity (HTMT).

Key Results

1. Modality Effects

Significant differences were observed across all study outcomes based on the summary modality.

Scientific Research Skills: The Video condition yielded the highest scores ( $M=0.86$ ), followed by Infographic ( $M=0.71$ ) and Audio ( $M=0.59$ ). All pairwise comparisons were statistically significant, with the video condition showing the largest effect size ( $\omega^2 = .164$ ).
Perceptual Outcomes: The video condition consistently outperformed the audio condition in perceived information quality, cognitive load reduction, and satisfaction of psychological needs. The infographic condition generally performed similarly to the audio condition, except in perceived information quality where it outperformed audio.

2. Structural Model and Direct Effects

The structural model explained 52.8% of the variance in scientific research skills.

Information and Summary Quality (ISQ) significantly predicted Autonomy ( $\beta=.337$ ), Competence ( $\beta=.464$ ), and Relatedness ( $\beta=.335$ ).
Cognitive Load Reduction (CLR) also significantly predicted all three psychological needs, though with smaller coefficients ( $\beta$ range: .140–.202).
Psychological Needs: Autonomy, Competence, and Relatedness all significantly predicted Scientific Research Skills. Competence showed the strongest association ( $\beta=.316$ ).
Direct Paths: While ISQ and CLR had significant direct effects on skills in the initial model, the direct effect of ISQ became non-significant when controlling for the experimental condition, suggesting its influence is largely mediated.

3. Mediation Analysis

All six specific indirect effects were significant, supporting a complementary partial mediation model.

ISQ Path: The indirect effect of ISQ on skills through Competence was the strongest ( $\beta=.147$ ), followed by Autonomy ( $\beta=.091$ ) and Relatedness ( $\beta=.061$ ).
CLR Path: Indirect effects through the three needs were smaller but significant ( $\beta$ range: .036–.055).
Experimental Mediation: When the experimental manipulation was included, the Video condition improved performance both directly ( $\beta=.275$ ) and indirectly through the SDT mechanism ( $\beta=.200$ ). In contrast, the Infographic advantage over Audio was primarily direct, with no significant indirect effect through the psychological needs chain.

Key Contributions and Significance

Theoretical Contributions

Extension of Self-Determination Theory (SDT): The study extends SDT into the domain of GenAI-supported scientific research education, demonstrating that autonomy, competence, and relatedness are critical mediators linking instructional design to higher-order research skills.
Integration of Cognitive and Motivational Mechanisms: Unlike studies that treat cognitive and motivational factors as competing explanations, this model integrates Cognitive Load Theory and SDT. It posits that high-quality, cognitively manageable GenAI content fosters psychological need satisfaction, which in turn drives skill acquisition.
Multimodal Comparative Evidence: The paper provides rare empirical evidence comparing GenAI-generated video, infographic, and audio summaries simultaneously. It challenges the assumption of equivalent learning outcomes across modalities, highlighting the specific affordances of video for sequential research processes.

Practical Implications

Instructional Design: Educators and designers should prioritize the quality and structure of AI-generated content over mere technological availability. Video summaries are particularly effective for demonstrating sequential research procedures, while infographics may better serve conceptual relationships.
Cognitive Scaffolding vs. Offloading: The findings suggest that effective GenAI use involves "productive cognitive load management"—reducing extraneous processing while preserving the intellectual effort required for critical evaluation and synthesis.
Psychological Support: GenAI interventions should be designed to explicitly support learner autonomy (choice/pace), competence (feedback/mastery), and relatedness (responsive interaction), as these are the mechanisms through which AI translates into improved research skills.

Limitations and Modesty of Claims

The author acknowledges that the study is context-specific to the Sultanate of Oman and a single educational program, limiting broad generalizability. The research is cross-sectional, and while it identifies mediation, it does not establish long-term causal changes in psychological needs. Furthermore, the study distinguishes between beneficial cognitive scaffolding and potential "cognitive offloading," noting that reduced cognitive effort does not automatically equate to deeper reasoning if the AI replaces rather than supports critical thinking. The paper concludes that GenAI is most valuable when it scaffolds rather than replaces thinking, and its effectiveness is contingent on the strategic alignment of modality with the specific cognitive demands of the research skill being taught.