Integrating AI-generated content tools in higher education: a comparative analysis of interdisciplinary learning outcomes
Background and current status
The rapid development of artificial intelligence technologies has profoundly impacted various sectors, with education emerging as a significant arena for artificial intelligence generated content (AIGC) implementation1. Higher education institutions worldwide are increasingly integrating AIGC tools such as ChatGPT, DALL-E, and Midjourney into teaching methodologies, research practices, and administrative functions2. These technologies, powered by large language models (LLMs) and diffusion-based image generators, have demonstrated remarkable capabilities in content creation, problem-solving, and simulation that were previously unattainable through conventional educational technologies3.
The adoption rate of AIGC tools in higher education can be quantified using the Technology Acceptance Model (TAM):
$$\:A=\frac{PU\times\:PEOU}{R}$$
where \(\:A\) represents adoption rate, \(\:PU\) denotes perceived usefulness, \(\:PEOU\) indicates perceived ease of use, and \(\:R\) signifies institutional resistance4.
Research significance and objectives
Despite growing implementation, systematic evaluation of AIGC tools’ impact on interdisciplinary learning outcomes remains underdeveloped5. This research gap is particularly concerning as institutions invest significant resources in AI integration without comprehensive understanding of differential outcomes across disciplines. The educational efficacy of these tools varies substantially across disciplines due to varying knowledge structures, pedagogical approaches, and evaluation metrics6.
The relationship between AIGC implementation and learning outcomes can be expressed as:
$$\:LO=\alpha\:\times\:AIGC+\beta\:\times\:DP+\gamma\:\times\:\left(AIGC\times\:DP\right)+ϵ$$
where \(\:LO\) represents learning outcomes, \(\:AIGC\) denotes the level of AI-generated content tool implementation, \(\:DP\) represents discipline-specific pedagogical approaches, and \(\:ϵ\) signifies unaccounted variables7.
This study aims to conduct a comparative analysis of AIGC tool integration across five distinct academic disciplines: engineering, humanities, business, life sciences, and fine arts. By measuring pedagogical effectiveness through a multi-dimensional assessment model, we can derive the optimization function:
$$\:E=\sum\:_{i=1}^{n}{w}_{i}\times\:{O}_{i}\times\:{D}_{i}$$
where \(\:E\) represents educational effectiveness, \(\:{w}_{i}\) denotes the weight of outcome measure \(\:i\), \(\:{O}_{i}\) represents the measured outcome, and \(\:{D}_{i}\) indicates discipline-specific contextual factors8.
Structure and scope
This paper is organized as follows: Section II reviews relevant literature on AIGC applications in educational contexts; Section III details the methodological framework employed; Section IV presents the comparative analysis findings across disciplines; Section V discusses implications for educational policy and practice; and Section VI concludes with recommendations for future research directions and implementation strategies.
The significance of this research extends beyond academic interest, addressing pressing challenges faced by higher education institutions regarding AI integration. As AIGC tools become increasingly sophisticated and accessible, developing evidence-based approaches to their implementation becomes essential for maximizing educational benefits while mitigating potential drawbacks associated with automated content generation, academic integrity concerns, and differential technological access.
Types and characteristics of AI-Generated content tools
Text generation tools
Text generation tools, primarily based on large language models (LLMs), represent the most widely adopted AIGC technologies in higher education contexts1. These systems utilize transformer architectures with self-attention mechanisms to model sequential data, as represented by the equation:
$$\:Attention\left(Q,K,V\right)=softmax\left(\frac{Q{K}^{T}}{\sqrt{{d}_{k}}}\right)V$$
where Q, K, and V represent query, key, and value matrices respectively, and \(\:{d}_{k}\) denotes the dimension of the key vectors2. Current implementations such as GPT-4 and Claude demonstrate sophisticated capabilities in content generation, summarization, and text transformation that support diverse pedagogical applications3. Wang et al. conducted a comprehensive analysis of 78 higher education institutions, finding that text-based AIGC tools were integrated into 64% of humanities courses but only 37% of STEM-focused curricula4.
Image generation tools
Diffusion models have emerged as the dominant paradigm for image generation, operating through the iterative denoising process described by:
$$\:{x}_{t-1}=\frac{1}{\sqrt{{\alpha\:}_{t}}}\left({x}_{t}-\frac{1-{\alpha\:}_{t}}{\sqrt{1-{\overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\leftharpoonup}$}} {\alpha }{\alpha\:}}_{t}}}{\epsilon}_{\theta\:}\left({x}_{t},t\right)\right)+{\sigma\:}_{t}z$$
where \(\:{x}_{t}\) represents the image at noise level \(\:t\), \(\:{\alpha\:}_{t}\) and \(\:{\overset{\lower0.5em\hbox{$\smash{\scriptscriptstyle\leftharpoonup}$}} {\alpha }{\alpha\:}}_{t}\) are noise scheduling parameters, and \(\:{\epsilon}_{\theta\:}\) is the learned noise prediction function5. Platforms such as DALL-E 3, Midjourney, and Stable Diffusion have demonstrated significant utility in design courses, visual arts education, and architectural visualization6. The integration of image generation tools presents unique challenges regarding attribution and originality assessment, with Martinez identifying significant disparities in institutional policies governing their use7.
Audio generation tools
Contemporary audio generation systems employ neural vocoding techniques that transform spectrograms into waveforms through the following process:
$$\:p\left(x|c\right)=\prod\:_{t=1}^{T}p\left({x}_{t}|{x}_{1:t-1},c\right)$$
where \(\:x\) represents the audio waveform, \(\:c\) denotes the conditioning information, and \(\:p\left({x}_{t}|{x}_{1:t-1},c\right)\) is modeled using autoregressive neural networks8. Tools such as Bark, AudioLM, and MusicLM enable the creation of speech, music, and environmental sounds with applications in language learning, music composition, and audio production courses9. Research by Chen et al. indicates that audio AIGC tools improved student engagement by a factor of 1.8 compared to traditional instructional methods in language acquisition courses10.
Educational implications and limitations
The educational effectiveness of AIGC tools can be quantified using a composite index that incorporates learning outcomes, engagement metrics, and skill development:
$$\:{E}_{AIGC}={\omega\:}_{1}L+{\omega\:}_{2}G+{\omega\:}_{3}S-{\omega\:}_{4}B$$
where \(\:L\) represents learning outcome improvement, \(\:G\) denotes engagement gains, \(\:S\) indicates skill development, \(\:B\) represents bias or accuracy concerns, and \(\:{\omega\:}_{1-4}\) are contextual weighting factors11. Despite their potential, AIGC tools face significant limitations in higher education contexts, including accuracy concerns, potential reinforcement of biases, and challenges in developing critical evaluation skills among students12. The effective implementation of these technologies requires thoughtful pedagogical frameworks that emphasize critical evaluation, transparent attribution, and complementary application alongside traditional learning methodologies.
Technology integration models in higher education
The successful integration of emerging technologies, including AIGC tools, into higher education requires structured theoretical frameworks to guide implementation. Several established models offer valuable insights into this process, each with distinct approaches to technology adoption and pedagogical integration.
Technology integration models and their application to AIGC
The Technological Pedagogical Content Knowledge (TPACK) framework emphasizes the complex interplay between three knowledge domains: content, pedagogy, and technology11. The effectiveness of technology integration can be expressed through the following relationship:
$$\:{E}_{TPACK}=\alpha\:\left(CK\right)+\beta\:\left(PK\right)+\gamma\:\left(TK\right)+\delta\:\left(CK\times\:PK\times\:TK\right)$$
Where α, β, γ, and δ represent weighting factors for Content Knowledge (CK), Pedagogical Knowledge (PK), and Technological Knowledge (TK) respectively12.
The Substitution, Augmentation, Modification, Redefinition (SAMR) model provides a progression framework for technology adoption, moving from enhancement to transformation13. In the context of AIGC integration, this transition can be represented as:
$$\:{T}_{impact}=\sum\:_{i=1}^{4}{S}_{i}\times\:{W}_{i}$$
Where T represents technological impact, S represents each SAMR stage, and W represents the corresponding weight of pedagogical transformation14.
The Technology Acceptance Model (TAM) focuses on perceived usefulness (PU) and perceived ease of use (PEOU) as determinants of adoption15. The behavioral intention to use AIGC tools can be modeled as:
$$\:BI={\beta\:}_{1}\left(PU\right)+{\beta\:}_{2}\left(PEOU\right)+\epsilon$$
Where BI represents behavioral intention, and β1 and β2 are regression coefficients16.
The ADDIE model (Analysis, Design, Development, Implementation, Evaluation) provides a systematic approach to instructional design that can be adapted for AIGC integration17. The model’s effectiveness can be quantified through:
$$\:{E}_{ADDIE}=\frac{\sum\:_{i=1}^{5}{P}_{i}\times\:{w}_{i}}{5}$$
Where P represents performance in each ADDIE phase and w represents phase weight18.
Recent research indicates that these models must evolve to address the unique characteristics of AIGC tools, including their generative capabilities, ethical considerations, and potential for personalized learning experiences19. As shown in Table 1, these educational technology integration models differ in their core concepts and implications for AIGC integration. Digital transformation in higher education has accelerated the need for frameworks that specifically address AI-enhanced learning environments20.
The evolution of these models reflects the changing landscape of educational technology integration. While early models focused primarily on hardware and software adoption, contemporary frameworks increasingly address the socio-technical systems and pedagogical transformations enabled by advanced technologies. The rapid development of AIGC tools presents both challenges and opportunities for these theoretical frameworks, necessitating adaptations that account for AI’s unique capabilities and limitations.
Cross-disciplinary learning and AIGC technology correlation research
Cross-disciplinary learning has emerged as a critical educational paradigm for developing complex problem-solving capabilities in the 21st century knowledge economy1. The theoretical foundation for cross-disciplinary learning can be traced to constructivist learning theory, which emphasizes knowledge construction through authentic problem-solving experiences across multiple domains2. Recent educational frameworks have formalized this approach through the Integrated Cross-disciplinary Learning Index (ICLI), which can be expressed as:
$$\:ICLI=\sum\:_{i=1}^{n}{w}_{i}\left({D}_{i}\times\:{C}_{i}\right)$$
Where \(\:{D}_{i}\) represents domain knowledge integration, \(\:{C}_{i}\) indicates cognitive transfer across disciplines, and \(\:{w}_{i}\) denotes the weighted importance of each disciplinary component3.
The integration of AIGC technologies into cross-disciplinary learning environments introduces new dynamics in knowledge co-construction and representation capabilities4. Studies indicate that AIGC tools can enhance cross-boundary thinking by providing computational support for domain knowledge synthesis through the Cross-domain Knowledge Synthesis Factor (CKSF):
$$\:CKSF=\frac{KI\times\:CT}{BD}$$
Where KI represents knowledge integration, CT denotes creative transformation, and BD indicates boundary-crossing difficulty5.
Contemporary research on AIGC applications in cross-disciplinary education demonstrates promising results across various institutional contexts6. As shown in Table 2, various cross-disciplinary learning models can be supported by different AIGC strategies to enhance educational outcomes. A meta-analysis of 27 studies revealed that AIGC-enhanced cross-disciplinary projects improved student learning outcomes by an average effect size of 0.68, particularly in complex problem-solving scenarios7. This improvement can be quantified through the Interdisciplinary Learning Enhancement Rate (ILER):
$$\:ILER=\frac{{P}_{post}-{P}_{pre}}{{P}_{max}-{P}_{pre}}\times\:100\text{\%}$$
Where \(\:{P}_{post}\) and \(\:{P}_{pre}\) represent post-intervention and pre-intervention performance, respectively, and \(\:{P}_{max}\) indicates maximum possible performance8.
Despite promising developments, significant research gaps persist in understanding the optimal integration mechanisms for AIGC tools in cross-disciplinary learning9. Current studies predominantly focus on short-term interventions rather than longitudinal impacts on knowledge transfer and disciplinary identity development10. Additionally, research methodologies require greater standardization to facilitate meaningful cross-study comparisons and generalizable design principles for AIGC-enhanced cross-disciplinary learning environments.
link
