Extending Cognitive Load Theory: The CLAM Framework for Biometric, Adaptive, and Ethical Learning

Abstract

Cognitive Load Theory (CLT) and the Cognitive Theory of Multimedia Learning (CTML) have long served as foundational frameworks in instructional design. However, their applicability to contemporary, technologically mediated learning environments remains under-theorized. This review critically examines CLT and CTML, focusing on their assumptions, empirical contributions, and current limitations in addressing the complexities of dynamic, AI-enhanced educational settings. The discussion is further enriched through engagement with complementary perspectives, including self-regulated learning, dual process theory, and connectivism. These frameworks illuminate conceptual convergences but also expose theoretical tensions, particularly regarding unresolved constructs such as germane cognitive load and the methodological challenges associated with real-time cognitive load measurement. In response to these gaps, this paper proposes the Cognitive Load-Aware Modulation (CLAM) strategy—a conceptual model designed to extend cognitive load principles in adaptive, ethically responsive learning environments. Synthesizing insights from cognitive psychology, educational technology, and affective computing, CLAM supports the design of personalized, data-driven instructional systems attuned to learners’ cognitive and emotional states. The model emerges not merely as a theoretical contribution, but as a future-oriented framework rooted in the critical synthesis of the reviewed literature. Its practical applications for real-world educational settings are outlined, and its empirical validation constitutes the next phase of our ongoing research project.

1. Introduction

In the context of accelerated technological transformation in education, understanding how learners process and manage information is more vital than ever. The proliferation of digital, hybrid, and AI-enhanced learning environments has dramatically expanded access to content and interactivity. Yet, this progress also introduces new cognitive demands that challenge conventional models of instructional design. Cognitive Load Theory (CLT) has emerged as a central framework for addressing these demands, offering a structured approach to optimizing instruction by aligning content complexity with the limited capacity of working memory and the mechanisms of long-term knowledge acquisition (Sweller et al., 2011a). Complementing this, Mayer’s Cognitive Theory of Multimedia Learning (CTML) provides guidance on how learners process verbal and visual information simultaneously, emphasizing principles such as segmenting, signaling, and modality (Mayer, 2005, 2024).

Despite their widespread application, both CLT and CTML were developed in educational settings that differ substantially from the dynamic, multimodal ecosystems of contemporary learning. These theories were conceptualized in relatively static environments and often presuppose uniform learner characteristics and linear instructional progression. Their current use in technology-enhanced contexts, particularly in AI-driven systems capable of personalization and real-time responsiveness, raises critical questions about their theoretical adequacy and empirical support.

Although enthusiasm for CLT’s relevance in digital education continues to grow, there remains a notable absence of integrative reviews that evaluate how its principles are operationalized across diverse digital modalities—such as mobile learning applications, immersive virtual environments, and responsive tutoring systems. Much of the existing literature is fragmented, limited to isolated instructional techniques or small-scale experimental studies, thereby restricting its applicability to real-world, large-scale educational contexts (Kalyuga, 2021; Wang & Lajoie, 2023).

This article is designed as a conceptual review, aiming to critically synthesize the literature on Cognitive Load Theory (CLT), the Cognitive Theory of Multimedia Learning (CTML), and related frameworks in the context of adaptive and AI-enhanced educational environments. Conceptual or integrative reviews offer valuable contributions by identifying theoretical gaps and constructing new frameworks grounded in existing research (Torraco, 2005; Snyder, 2019). In this light, the proposed CLAM (Cognitive Load-Aware Modulation) model is introduced not as an empirically tested tool, but as a theory-driven framework developed in response to well-documented limitations of current cognitive models in dynamically evolving learning contexts.

Throughout this paper, while the term “adaptive” is used to maintain terminological clarity, we also consider closely related concepts such as responsive, personalized, and modulated instruction—each emphasizing distinct dimensions of instructional flexibility within cognitively informed learning environments.

Furthermore, emerging evidence questions the universal efficacy of multimedia learning strategies. Some studies suggest that under conditions of high cognitive complexity, minimalist text-based formats may yield superior outcomes, particularly for novice learners with limited prior knowledge (Kalyuga et al., 2000; Sweller et al., 2019). At the same time, affective and motivational variables—such as anxiety, interest, and goal orientation—remain largely peripheral in most CLT-informed instructional design, despite consistent findings linking emotional states with cognitive efficiency and learning performance (Evans et al., 2023; Kosel et al., 2024; Plass & Kalyuga, 2019).

Another underexplored area concerns the capacity of adaptive systems to meaningfully apply CLT principles in real time. While artificial intelligence holds promise for tracking learners’ cognitive states and modifying instruction accordingly, many such systems are conceptual or experimental, lacking robust validation in authentic educational settings (Nasri, 2025).

The present review aims to bridge conceptual and practical gaps by critically synthesizing contemporary literature in cognitive load theory and multimedia learning—particularly in the context of AI-enhanced and biometric-supported personalized learning environments. While CLT retains substantial theoretical value, its translation into modern digital learning remains uneven and insufficiently responsive to real-world complexity (Skulmowski & Xu, 2021; Mayer, 2024). There is a pressing need for theoretical refinement that accounts for the dynamic, personalized, and affectively rich nature of contemporary educational settings (Çeken & Taşkın, 2022). In response to these limitations, we introduce the Cognitive Load-Aware Modulation (CLAM) strategy, a conceptual model designed to extend traditional cognitive frameworks into ethically responsive, real-time instructional contexts. CLAM is not based on empirical data but is grounded in interdisciplinary theoretical integration and guided by urgent educational challenges.

The structure of the paper is as follows. Section 2 outlines the review scope and methodology. Section 3, Section 4, Section 5 and Section 6 examine key instructional principles and theoretical tensions. Section 7 and Section 8 present the CLAM framework and its five core components. Section 9 and Section 10 critically assess its implications, limitations, and future directions for research and applications. Finally, Section 11 concludes the review by summarizing the model’s contributions and outlining directions for empirical validation.

2. The Current Review and Empirical Integration

Although the present review is conceptual in orientation, we consider it important to maintain transparency in how the relevant literature was identified and selected. As noted by Snyder (2019), integrative reviews are particularly well suited for capturing emerging insights across diverse domains—an approach we deemed appropriate given the interdisciplinary nature of the topic. To this end, we focused on peer-reviewed articles published between 2020 and 2025, retrieved exclusively from Scopus-indexed journals to ensure academic quality and consistency. While the review focused primarily on recent sources, a limited number of earlier studies were retained when they offered foundational insights or introduced concepts that remain central to the field. For example, M. Chen et al. (2007) provides one of the earliest models of learner profiling in web-based systems and is referenced here due to its enduring relevance. Selection was guided by thematic relevance to four intersecting areas: cognitive load theory (CLT), multimedia learning (CTML), biometric indicators of cognitive processing (e.g., EEG, HRV, pupillometry), and adaptive instructional technologies. Studies that did not engage substantively with these areas or lacked theoretical or empirical soundness were excluded. This strategy enabled us to construct a coherent and analytically grounded foundation while preserving the flexibility that conceptual synthesis requires.

The structure of this review aligns with the conceptual review methodology described by Snyder (2019). This approach is particularly appropriate for areas undergoing rapid theoretical development or requiring interdisciplinary synthesis. Instead of exhaustively cataloging empirical studies, the aim here is to identify conceptual patterns, theoretical tensions, and emerging needs within the literature, and to use these to inform the development of a forward-looking framework. In this case, the CLAM model is proposed not as a summary of past research findings, but as a theoretically grounded response to current limitations in cognitive load theory and adaptive instructional design.

The current review synthesizes key theoretical and empirical developments related to Cognitive Load Theory (CLT), with a focus on its integration into instructional design and multimedia learning. Building upon foundational principles, the review evaluates the extent to which CLT accommodates the demands of contemporary educational contexts—especially those shaped by digital transformation, learner heterogeneity, and adaptive technologies.

At the theoretical level, the review revisits core components of CLT, including the distinction between intrinsic, extraneous, and germane cognitive load (Sweller et al., 2011b). Emphasis is placed on the design principles derived from this taxonomy, such as the minimization of extraneous load and the structured promotion of germane load. These are explored in conjunction with complementary frameworks, notably Mayer’s Cognitive Theory of Multimedia Learning (CTML), which extends CLT’s application to multimodal instructional environments. Principles such as segmentation, coherence, and signaling are examined for their cognitive efficiency, particularly when implemented in digital and hybrid learning contexts.

Empirical findings further substantiate CLT’s instructional relevance. Interventions such as worked examples and scaffolding have demonstrated positive outcomes in supporting novice learners, especially in domains such as mathematics (O. Chen et al., 2023). Meta-analytic evidence confirms that spatial contiguity and signaling reduce extraneous load and enhance learning performance (Schroeder & Cenkci, 2018). In addition, segmented instructional sequences have been found to improve retention and engagement, particularly among learners with limited prior knowledge (Spanjers et al., 2011).

The review also incorporates findings from emerging fields, including biometric monitoring and artificial intelligence-based instructional platforms. Studies employing EEG and eye-tracking technologies provide insight into real-time cognitive states, allowing for more responsive instructional adjustments (V. Liu et al., 2025). These technologies highlight the need for models such as CLT to evolve, incorporating real-time cognitive feedback mechanisms that maintain efficiency without overwhelming the learner (Baradari et al., 2025).

Furthermore, the review acknowledges the importance of learner diversity in cognitive processing. Research on neurodiverse populations, including individuals with ADHD and dyslexia, illustrates the need for differentiated approaches to managing cognitive load (Bannert & Mengelkamp, 2024). The expertise reversal effect reinforces this point, demonstrating that instructional techniques must be tailored to the learner’s existing knowledge base to avoid redundant or counterproductive cognitive demands (Kalyuga, 2007, 2009b).

Theoretical boundaries of CLT are also critically examined. Sweller (2010) has questioned the conceptual independence of germane load, proposing its integration within intrinsic load. In contrast, Bjork and Bjork (2020) advocate for the inclusion of “desirable difficulties” in learning, suggesting that certain levels of challenge may benefit long-term retention, despite their initial cognitive cost. These perspectives introduce valuable nuance to the design of cognitively effective instruction.

Finally, the review calls for enhanced methodological precision in cognitive load measurement. The incorporation of biometric data (Juliano et al., 2022) and principles from a self-regulated learning theory (Song et al., 2023) opens new avenues for developing adaptive, data-informed instructional designs. These approaches support the broader aim of aligning CLT with dynamic, learner-centered educational paradigms.

Taken together, these theoretical tensions and empirical insights point to the need for instructional frameworks that move beyond static principles and respond dynamically to learners’ cognitive states (Nasri, 2025). The limitations of current models—particularly in addressing real-time adaptation, individual cognitive profiles, and ethically informed personalization—underscore the importance of developing new approaches (Sajja et al., 2023). In response to these challenges, this review proposes the Cognitive Load-Aware Modulation (CLAM) strategy, introduced in the following section, as a conceptual advancement that builds upon and extends the foundational insights of CLT.

3. Applying Cognitive Load Principles in Practice: Instructional Design Strategies for Contemporary Learning

Although CLT and CTML provide theoretical foundations, their value is realized in practical implementation. This section discusses how empirically validated principles derived from CLT and CTML can guide the development of instructional materials that are both cognitively efficient and pedagogically sound. Emphasis is placed on strategies that regulate cognitive demands, facilitate learner engagement, and promote sustainable comprehension across diverse instructional settings. (Lopez, 2024).

3.1. The Multimedia Principle: Integrating Visual and Verbal Modalities

Research indicates that learners benefit more from materials that combine spoken explanations with meaningful visuals than from text-based instruction alone (Mayer & Moreno, 2003). This principle, grounded in dual-channel processing theory, acknowledges the brain’s capacity to simultaneously handle visual and auditory inputs. By distributing cognitive load across modalities, multimedia instruction supports schema development and reduces overload (Schroeder & Cenkci, 2018).

3.2. The Coherence Principle: Eliminating Irrelevant Content

Extraneous information—such as decorative graphics, background audio, or stylistically complex animations—can distract learners and interfere with processing. The coherence principle advocates for the removal of non-essential elements that do not contribute to learning objectives (Mayer, 2009). Simplifying visual and auditory input enables learners to focus on core content, enhancing clarity and retention (Clark & Mayer, 2016).

3.3. The Modality Principle: Leveraging Speech to Support Visual Learning

When information is presented visually—through images, diagrams, or animations—complementing it with spoken narration rather than written text improves processing efficiency (Mayer & Moreno, 2003). This principle prevents overloading the visual channel and allows learners to allocate cognitive resources more effectively, particularly when engaging with complex material (Y. Liu et al., 2023). Research has shown that combining narration with visuals reduces cognitive load and supports comprehension (Schroeder & Cenkci, 2018). By minimizing the need to read dense text alongside visual input, this approach helps learners concentrate on understanding core concepts rather than decoding written information.

3.4. The Redundancy Principle: Avoiding Unnecessary Repetition

Duplicating identical content across modalities—such as presenting the same text on screen while reading it aloud—can split attention and hinder comprehension. The redundancy principle suggests that learners benefit more when each modality contributes distinct, complementary information (Kalyuga et al., 2004). When used judiciously, this approach reduces unnecessary processing and supports integration.

3.5. The Segmenting Principle: Structuring Information for Gradual Processing

Breaking down complex information into smaller, self-contained segments allows learners to process content incrementally. This reduces cognitive overload by aligning instructional pacing with the learner’s capacity to assimilate new information (Spanjers et al., 2011). Educators can implement this by delivering lessons in short, structured segments, with pauses for reflection. Segmenting is particularly effective in digital platforms that support modular learning experiences (D. Liu, 2024). Online platforms apply this by using micro-lessons rather than lengthy lectures, improving engagement and retention.

3.6. The Signaling Principle: Emphasizing Instructionally Relevant Content

Guiding attention to key elements through visual or auditory cues—such as color highlights, arrows, bold text, or spoken emphasis—can significantly reduce extraneous load and improve comprehension (Van Gog et al., 2014). By making instructional priorities more salient, signaling enables learners to allocate their limited cognitive resources more effectively. This is particularly beneficial in multimedia environments, where complex visual and auditory information often competes for attention. Research has shown that structured cues not only support initial processing but also facilitate long-term retention by reinforcing the organization of knowledge in memory (Mayer, 2009).

3.7. The Worked Example Principle: Demonstrating Expert Reasoning

Novice learners often struggle with integrating new information while solving unfamiliar problems. Worked examples offer a model of expert performance, allowing learners to observe step-by-step procedures before attempting independent application (O. Chen et al., 2023). Their effectiveness in reducing cognitive demand has been demonstrated in various domains, including mathematics, science, and academic writing (Schwonke et al., 2009). Similarly, in writing instruction, analyzing annotated essays helps students internalize structure before composing their own. This method fosters deeper learning and improves problem-solving skills (Wexler, 2019).

4. Integrating Cognitive Theory of Multimedia Learning and Cognitive Load Theory: Foundations for Optimized Instructional Design

As education evolves, multimedia learning has become central to effective instructional design. Richard Mayer’s Cognitive Theory of Multimedia Learning (CTML) expands on Cognitive Load Theory (CLT) by explaining how learners process, retain, and apply multimedia-based instructional materials. Both theories acknowledge the limits of working memory and emphasize strategies that enhance cognitive efficiency (Mayer, 2014).

CTML is based on three core principles: dual-channel processing, which explains how verbal and visual information are processed through separate but complementary cognitive pathways; limited capacity, which highlights the restrictions of working memory and the potential risks of cognitive overload; and active processing, which underscores the importance of meaningful engagement in constructing and integrating knowledge (Mayer, 2005).

Understanding how multimedia learning interacts with cognitive load is essential for designing effective instructional materials. Research shows that well-structured multimedia—such as videos, animations, narrated presentations, and interactive simulations—reduces extraneous load, directs attention to key concepts, and fosters schema construction. By integrating CTML with CLT, educators can develop strategies that promote deeper cognitive engagement, improve knowledge retention, and enhance learning across digital and traditional educational settings (Sweller, 2020).

4.1. Dual-Channel Processing: The Role of Visual and Auditory Input

The dual-channel processing principle suggests that the brain processes information through two separate channels: the visual-pictorial channel, responsible for handling images, diagrams, animations, and graphs, and the auditory-verbal channel, which processes spoken words and sounds (Mayer, 2005). Instructional materials that rely too heavily on one modality, such as dense text without visuals, can overwhelm cognitive resources, reducing comprehension and retention. Multimedia instruction should balance both channels to optimize learning and prevent cognitive overload. Research confirms that combining spoken narration with visuals—rather than relying solely on text—reduces extraneous cognitive load and improves understanding (Schroeder & Cenkci, 2018). For example, narrated videos can replace text-heavy presentations, allowing learners to absorb spoken explanations while engaging with visuals. Similarly, using diagrams or animations alongside verbal instruction promotes deeper cognitive processing, leading to more effective learning outcomes.

4.2. Limited Capacity: Managing Working Memory

Working memory has a limited capacity, meaning students can only process a finite amount of new information at a time (Cowan, 2012). Presenting excessive information at once leads to cognitive overload, reducing retention and comprehension. To prevent this, instructional materials should be concise, well-paced, and structured in alignment with learners’ cognitive limits. Studies suggest that shorter instructional videos (5–10 min) are more effective than long lectures, as they enhance engagement and retention (Spanjers et al., 2011). Breaking lessons into logically sequenced sections also improves cognitive processing and minimizes mental fatigue.

4.3. Active Processing: The Key to Meaningful Learning

For learning to be effective, students must actively engage with new information—organizing, connecting, and integrating it with what they already know. Simply receiving information is not enough; instructional material should promote active involvement. Research shows that techniques like guided questions, pause prompts, and interactive quizzes improve retention by encouraging deeper cognitive engagement (Van Gog et al., 2014). Additionally, summarizing content in one’s own words strengthens understanding and facilitates schema development.

5. Optimizing Learning: Cognitive Load Theory and Multimedia Instructional Design

Instructional design grounded in cognitive science aims not only to transmit information but to facilitate meaningful learning, long-term retention, and transfer across contexts. Cognitive Load Theory (CLT) provides a framework for regulating mental effort (Sweller et al., 2011a). When combined with CTML, which addresses how the mind processes multimodal input, instructional design becomes more cognitively aligned (Plass & Kalyuga, 2019; Schroeder & Cenkci, 2018). CTML contributes principles such as dual-channel processing, coherence, and signaling, which are especially relevant in digital environments involving multimedia content, interactive simulations (Plass & Kalyuga, 2019), and AI-driven systems (Y. Liu et al., 2023). These technologies present both opportunities and challenges. While they offer personalization and real-time feedback, they can also increase cognitive load if poorly designed. Future research should continue examining how such innovations affect cognitive load and learning outcomes, especially in dynamic, data-rich environments (Gkintoni et al., 2025).

Applying cognitive load-informed strategies has been shown to enhance learners’ engagement and understanding while reducing unnecessary processing (Clark & Mayer, 2016). Well-designed instructional materials promote accessibility and learning efficiency, making them valuable across educational levels and settings. The following subsections explore how CLT and CTML intersect in multimedia learning and offer guidance for managing task complexity in alignment with learners’ cognitive capacities.

5.1. How CLT and CTML Work Together in Multimedia Learning

Cognitive Load Theory (CLT) and the Cognitive Theory of Multimedia Learning (CTML) offer complementary perspectives on instructional design. CLT manages the learner’s mental effort by regulating intrinsic, extraneous, and germane load (Sweller et al., 2011b). CTML structures multimedia input to align with cognitive functioning (Mayer, 2024). Rather than reiterating multimedia principles, this section refers to the foundational applications discussed in Section 3.

5.2. Balancing Task Complexity and Cognitive Load in Learning Environments

Effective instruction entails more than simply reducing content difficulty. Rather, it involves carefully calibrating task complexity in accordance with learners’ cognitive capacity. Within the framework of CLT, the goal is to present material that challenges learners appropriately without overwhelming their working memory (Sweller et al., 2011b; Kalyuga, 2009a). This calibration depends not only on the inherent difficulty of the content but also on the degree of element interactivity and the cognitive effort required for integration (Kalyuga, 2007).

Learners’ prior knowledge and experience significantly influence how task complexity is perceived. Novices, who lack established schemas, tend to process each element individually, increasing the likelihood of overload. In contrast, experienced learners can draw on well-formed mental structures that support more efficient processing (Kalyuga, 2009b). These differences necessitate differentiated instructional approaches. Beginners typically benefit from guided instruction and scaffolded problem-solving, whereas more advanced learners may thrive under inquiry-based or open-ended conditions (Kirschner et al., 2006).

As outlined in Section 3, techniques such as segmentation, signaling, and the use of worked examples can help manage cognitive demands by pacing information flow and reducing extraneous processing. To further enhance cognitive efficiency, educators can incorporate retrieval-based tasks, generative activities, and opportunities for reflection (Garnett, 2020; Roelle et al., 2020). These strategies foster deeper engagement while maintaining the intellectual integrity of the task. Ultimately, by aligning task design with learners’ evolving cognitive resources, instruction can remain both rigorous and sustainable.

6. Reevaluating Cognitive Load Theory: Challenges, Limitations, and the Need for a More Adaptive Approach

Although Cognitive Load Theory (CLT) has significantly shaped contemporary instructional design, its limitations have increasingly become the subject of critical discussion (Kirschner et al., 2006; Sweller et al., 2011a). Core issues include the classification of cognitive load types, the methodological challenges of accurate measurement, and the theory’s adaptability to the complexities of real-world learning environments. Some scholars have questioned whether the tripartite categorization of intrinsic, extraneous, and germane load fully captures the dynamic and context-dependent nature of learning (Plass & Kalyuga, 2019; Sweller, 2010). In particular, the notion that cognitive load should always be minimized is contested. Research on “desirable difficulties” suggests that engaging with challenging material—when appropriately scaffolded—can enhance long-term retention and promote deeper learning (Bjork & Bjork, 2020). Overemphasizing load reduction may inadvertently limit opportunities for cognitive growth and resilience, particularly in higher education or advanced skill acquisition (Kapur, 2008).

Another enduring challenge relates to the measurement of cognitive load. Despite widespread use, self-report scales such as the NASA-TLX and Paas’ mental effort rating scale (Paas, 1992) offer limited granularity and are prone to subjective biases (de Jong, 2010). Advances in physiological and behavioral monitoring—such as eye-tracking, EEG, pupillometry, and heart rate variability—offer more objective means of capturing cognitive effort in real time (Juliano et al., 2022; Y. Liu et al., 2023). However, such methods remain largely confined to laboratory contexts and are not yet broadly integrated into educational practice. The potential for AI-enhanced monitoring systems to incorporate biometric data and adapt instruction accordingly represents a promising avenue for future research (Aleven et al., 2017; D’Mello & Graesser, 2015).

Further theoretical ambiguity exists around the distinction between intrinsic and germane load (Sweller et al., 2011b). Sweller (2010) has argued that germane load may be more appropriately understood as a subset of intrinsic load, as both relate to processing task-relevant information. This reconceptualization challenges the instructional goal of “increasing germane load,” suggesting instead that meaningful learning results from optimizing overall cognitive efficiency (Kalyuga, 2011). Additionally, the assumption that reducing extraneous load benefits all learners equally is being reevaluated. Empirical evidence shows that expert learners are often able to manage greater complexity without detriment, while novices benefit more from structured, step-by-step guidance (Kalyuga, 2007; van Merriënboer & Sweller, 2005).

The theory’s generalizability is further complicated by the needs of neurodiverse learners. Individuals with attention-deficit/hyperactivity disorder (ADHD), dyslexia, or autism spectrum conditions may experience and process cognitive load in qualitatively distinct ways (Alloway & Alloway, 2010; Boekaerts, 2011). Instructional designs that do not account for such variation risk excluding or under-serving these learners. Adaptive learning technologies offer a way forward by enabling real-time adjustments to instructional content, pacing, and modality, based on learners’ evolving cognitive profiles (Kizilcec et al., 2020; Walkington, 2013).

While CLT continues to provide a valuable framework for designing cognitively efficient instruction, it must be refined to reflect the realities of contemporary learning—realities characterized by heterogeneity, interactivity, and the growing integration of intelligent systems. A more adaptive, context-sensitive interpretation of CLT is necessary to support diverse learners in increasingly complex and technologically mediated educational environments.

7. Future Directions in CLT: Toward Biometrically Adaptive and Ethically Responsive Learning

As education becomes increasingly digital, multimodal, and personalized, Cognitive Load Theory (CLT) must evolve to meet the challenges posed by these shifts. While CLT has served as a foundational framework in instructional design for decades, its continued relevance depends on its capacity to engage with emerging pedagogical, technological, and ethical imperatives (Mayer, 2009; Sweller et al., 2011a). This section outlines key directions for the advancement of CLT, emphasizing developments in cognitive load measurement, expansion into new learning contexts, and its integration with intelligent and responsive educational systems.

A central issue in contemporary CLT research is the accurate assessment of cognitive load. Traditional self-report instruments, such as the Paas mental effort scale (Paas, 1992), though widely adopted, have been criticized for their subjectivity and limited temporal sensitivity (de Jong, 2010). In response, recent research has turned to biometric methodologies—including electroencephalography (EEG), eye-tracking, heart rate variability, and pupillometry—as objective, real-time indicators of cognitive effort (Buettner et al., 2021; Gkintoni et al., 2025; Juliano et al., 2022; Y. Liu et al., 2023). The growing accessibility of wearable biosensors (e.g., Tobii Pro Glasses, Muse EEG headbands) has enabled the use of such tools in authentic educational settings, opening up new possibilities for dynamic, evidence-based instructional modulation (Sharma et al., 2021).

Parallel to these methodological developments is the theoretical broadening of CLT into increasingly complex and authentic learning environments. Although originally formulated in the context of well-structured tasks in STEM education (Sweller & Chandler, 1991), CLT is now being adapted to areas such as second-language learning (Roussel et al., 2022), medical simulation (Andersen et al., 2016), and workplace safety training. These domains require instructional strategies that can support cognitive efficiency without oversimplifying the inherent complexity of real-world tasks.

Immersive technologies such as virtual and augmented reality present a further frontier for CLT. While these environments can enhance learner engagement, they often impose elevated cognitive demands due to their perceptual richness and attentional complexity. If poorly designed, such experiences risk inducing overload. However, when combined with CLT-informed principles—such as signaling, segmentation, and pre-training—VR can become a powerful medium for learning (Makransky & Petersen, 2021; Parong & Mayer, 2020).

Another promising direction lies in the integration of CLT with complementary frameworks such as self-regulated learning (SRL). SRL emphasizes learners’ abilities to monitor, plan, and adjust their own learning strategies—a process that is inherently linked to the management of cognitive load (Azevedo et al., 2010; Gorbunova et al., 2024). Instructional designs that incorporate metacognitive prompts and cognitive scaffolds have been shown to facilitate deeper engagement and improved performance, particularly in complex problem-solving contexts (Song et al., 2023; Q. Zhang & Lockee, 2022).

Perhaps the most significant transformation is being driven by artificial intelligence (AI). AI-powered adaptive learning systems can tailor content presentation, pacing, and difficulty based on continuous learner input (Aleven et al., 2017; Kizilcec et al., 2020). Tools such as Smart Sparrow and Carnegie Learning already operationalize CLT principles to personalize learning in real time (Baker & Inventado, 2014). When paired with biometric data—such as gaze behavior or EEG patterns—these platforms can detect indicators of overload and adjust instructional sequences accordingly (D’Mello & Graesser, 2015; Zu et al., 2020; Y. Zhang et al., 2024; Zhu & Lv, 2023).

Despite their potential, these technologies raise important ethical considerations. The use of biometric monitoring, especially in educational settings, introduces questions of data privacy, transparency, and consent. As Holmes et al. (2021) note, AI in education must be governed by principles of fairness, accountability, and explainability. The integration of CLT into such systems must therefore be accompanied by ethical guidelines that ensure learner autonomy and protect individual rights (Bower et al., 2024; Zawacki-Richter et al., 2019).

Simultaneously, CLT is being expanded to populations and disciplines traditionally underrepresented in cognitive load research. In the humanities, where learning often involves abstract reasoning and interpretive analysis, segmenting content and reducing extraneous complexity has been shown to enhance comprehension and engagement (Sweller, 2020). For neurodiverse learners, CLT-aligned strategies—such as multimodal content presentation and reduction in redundant information—can facilitate access and reduce unnecessary processing demands (Castro-Alonso & Sweller, 2020; Le Cunff et al., 2024).

Taken together, these developments point to an emerging vision of CLT as a dynamic, context-sensitive framework capable of informing instruction across increasingly diverse and technologically sophisticated learning landscapes. As the boundaries of education continue to expand, CLT must remain responsive not only to cognitive efficiency, but also to inclusivity, emotional engagement, and the ethical dimensions of learning design.

What distinguishes this work is the proposal of the CLAM model as a forward-looking framework that extends cognitive load theory into new instructional territory. Rather than applying CLT principles retrospectively, CLAM is designed to operate in real time, integrating biometric feedback, adaptive decision-making, and ethical safeguards. To our knowledge, no existing model organizes these elements within a unified structure specifically tailored for today’s multimodal, data-rich, and ethically complex learning environments.

The next section introduces the Cognitive Load-Aware Modulation (CLAM) strategy—a novel instructional framework that synthesizes these advances into a cohesive model for real-time, ethically grounded, and learner-responsive pedagogy.

8. The CLAM Strategy: A Next-Generation Model for Cognitive Load-Informed Adaptive Learning

In response to the increasing complexity of digital learning environments and the limitations of static instructional approaches, this section introduces the Cognitive Load-Aware Adaptive Modulation (CLAM) Strategy. Rooted in the principles of cognitive load theory (CLT; Sweller et al., 2011a) and Mayer’s Cognitive Theory of Multimedia Learning (CTML; Mayer, 2009), CLAM also draws on advances in affective computing (D’Mello & Graesser, 2015), biometric monitoring (Y. Liu et al., 2025), and adaptive learning systems (Aleven et al., 2017). The model offers a dynamic framework for designing instruction that adapts in real time to the learner’s cognitive and emotional state.

CLAM is built on five interconnected components—Dynamic Load Monitoring (DLM), Learner Profile Matrix (LPM), Multimodal Adaptive Instruction (MAI), Cognitive Load Zoning (CLZ), and Feedback-Driven Knowledge Construction (FDKC)—each of which addresses core challenges in instructional design for contemporary learning contexts. See

Table 1. References supporting CLAM model components.

CLAM Component	Supporting References (APA 7)
DLM (Dynamic Load Monitoring)	Ekin et al. (2025). https://doi.org/10.1038/s41598-025-89336-y Guo (2025). https://ojs.sin-chn.com/index.php/mcb/article/view/935 Ahmad et al. (2023). https://doi.org/10.1007/s00779-020-01455-7 X. Liu and Cui (2025). https://doi.org/10.1016/j.compedu.2025.105263
LPM (Learner Profiling Module)	Fuchs and Henning (2022). https://doi.org/10.1201/9781003337690-3 Ronkainen et al. (2025). https://doi.org/10.1016/j.lindif.2025.102698 R. Wong et al. (2024). https://par.nsf.gov/servlets/purl/10545057
MAI (Multimodal Adaptive Instruction)	Aslan et al. (2025). https://doi.org/10.1016/j.caeai.2025.100393 V. Liu et al. (2025). https://doi.org/10.48550/arXiv.2503.09748 Makransky and Mayer (2022). https://doi.org/10.1007/s10648-022-09675-4
CLZ (Cognitive Load Zoning)	Ekin et al. (2025). https://doi.org/10.1038/s41598-025-89336-y Ferguson et al. (2022). https://doi.org/10.1016/j.caeai.2022.100089 Halkiopoulos and Gkintoni (2024). https://doi.org/10.3390/electronics13183762 X. Liu and Cui (2025). https://doi.org/10.1016/j.compedu.2025.105263
FDKC (Feedback-Driven Knowledge Construction)	Azevedo et al. (2022). https://doi.org/10.3389/fpsyg.2022.813632 Ferguson et al. (2022). https://doi.org/10.1016/j.caeai.2022.100089 Newar et al. (2023). https://doi.org/10.1016/j.scico.2023.102933

Note: Some studies—such as X. Liu and Cui (2025) and Ekin et al. (2025)—support more than one module, reflecting the integrative architecture of the CLAM model.

for supporting references corresponding to each CLAM component.

Figure 1 illustrates the dynamic architecture of the CLAM model. The process begins with incoming data streams—including biometric indicators (e.g., HRV, pupil dilation, EEG) and behavioral patterns (e.g., latency, clickstream)—which are analyzed in real time by the Dynamic Load Monitoring (DLM) module. Based on this analysis, the Cognitive Load Zoning (CLZ) mechanism classifies learners into effort-based zones. These zones then inform the selection of adaptive instructional strategies via the Multimodal Adaptive Instruction (MAI) component, in coordination with learner profiles maintained by the Learner Profiling Module (LPM). Finally, feedback loops driven by learner performance and physiological changes are captured through the Feedback-Driven Knowledge Construction (FDKC) component, allowing the system to continuously recalibrate instruction and personalize the learning experience over time.

8.1. Dynamic Load Monitoring (DLM)

DLM captures learners’ moment-to-moment cognitive and affective states using multimodal indicators such as eye-tracking, electroencephalography (EEG), heart rate variability, pupillometry, and facial expression analysis. These indicators have been empirically validated as reliable proxies for cognitive load, attention, and emotional arousal (Alshanskaia et al., 2024; Juliano et al., 2022; Y. Liu et al., 2023). For example, pupil dilation has been shown to correlate with increased mental effort during problem-solving tasks (van der Wel & van Steenbergen, 2018), while EEG markers such as theta and alpha oscillations provide insight into attentional states and working memory load (Ekin et al., 2025). DLM leverages these signals to initiate modulated changes in instruction before overload impedes learning (William & Murugesh, 2020). These mechanisms build directly upon prior research reviewed in Section 5 and Section 6, particularly on biometric load monitoring and real-time physiological markers (Y. Liu et al., 2025; Juliano et al., 2022).

8.2. Learner Profile Matrix (LPM)

The LPM incorporates data on learners’ prior knowledge, working memory capacity, neurodiversity, cognitive style, and motivational orientation to guide instructional decisions (O. Chen et al., 2023; Plass & Kalyuga, 2019). The personalized learning literature consistently highlights the value of personalization in scaffolding cognitive effort (Kalyuga, 2007; Kirschner et al., 2006), while inclusive design research underscores the importance of accounting for neurodiversity in instructional planning (Alloway & Alloway, 2010; Boekaerts, 2011). Within CLAM, the LPM aligns instructional complexity with the learner’s zone of proximal development (van Merriënboer & Sweller, 2005; Vygotsky, 1978; Fuchs & Henning, 2022), supporting both accessibility and cognitive engagement. The foundations of LPM reflect the learner-centered approaches highlighted in Section 4, emphasizing the importance of adaptive scaffolding based on cognitive profiles and neurodiversity considerations (O. Chen et al., 2023; Alloway & Alloway, 2010).

8.3. Multimodal Adaptive Instruction (MAI)

MAI regulates the delivery of instructional content through responsive selection and sequencing of media channels—text, audio, animation, video, or haptic feedback—based on real-time assessments of the learner’s state. It applies CTML principles such as coherence, segmentation, and modality (Mayer & Moreno, 2003; Schroeder & Cenkci, 2018), translating them into responsive instructional behaviors. Empirical findings suggest that strategic management of modality can enhance both retention and affective engagement, especially under cognitively demanding conditions (Makransky & Lilleholt, 2018; Parong & Mayer, 2020). MAI functions as the operational mechanism through which the system maintains cognitive efficiency without compromising motivation. As outlined in Section 3 and Section 5, the principles governing MAI derive from CTML and are reinforced by empirical evidence on modality and engagement effects in multimedia learning (Mayer & Moreno, 2003; Makransky & Lilleholt, 2018).

8.4. Cognitive Load Zoning (CLZ)

CLZ introduces a tripartite framework to categorize learner experience into three cognitive zones: Safe Load, Effort, and Overload. This conceptualization reflects evidence that moderate cognitive challenge facilitates learning, while excessive cognitive demands are detrimental to performance (Kalyuga, 2011; Sweller, 2010). It also aligns with the concept of “desirable difficulties,” which posits that appropriately calibrated struggle can foster long-term retention (Bjork & Bjork, 2020). In applied settings such as immersive simulation or problem-solving environments, CLZ offers a practical schema for dynamic load balancing (Makransky et al., 2019). The conceptualization of CLZ is informed by the desirable difficulties framework and the expertise reversal effect, both of which are critically discussed in Section 6 (Bjork & Bjork, 2020; Kalyuga, 2007). Within the CLAM framework, CLZ does not operate in isolation; rather, it directly informs the instructional behavior of the MAI system. Once a learner is classified into a zone, this zoning status functions as an instructional switch that regulates the type, complexity, and modality of content delivery via MAI. This modulation occurs dynamically and recursively, meaning that MAI continuously receives zoning updates from CLZ to inform real-time instructional decisions. In its current conceptualization, CLZ operates through rule-based heuristics and threshold modeling rather than machine learning classification. These heuristics are grounded in prior biometric research (e.g., pupil dilation thresholds, EEG theta/alpha ratios), allowing for interpretable and context-sensitive zoning without requiring extensive training data. In essence, CLZ and MAI operate as a tightly coupled feedback loop within CLAM’s architecture. For example, learners in the Overload zone receive reduced-extraneous-load materials such as step-by-step visuals with narrated explanations. Those in the Effort zone are presented with richer, dual-modality or interactive formats that optimize germane processing. Learners in the Safe Load zone are nudged toward mild challenge via exploratory or problem-based tasks. In this way, CLZ enables MAI to respond adaptively to learner needs in real time, ensuring that instruction remains both cognitively efficient and emotionally sustainable.

Cognitive Load Zoning is measured through the interpretation of real-time biometric and behavioral indicators, many of which have been validated in recent cognitive neuroscience and educational technology studies. Specifically, pupil dilation has been shown to correlate with mental effort in problem-solving and executive control tasks (van der Wel & van Steenbergen, 2018; Kosel et al., 2024). EEG oscillatory activity, particularly theta and alpha band patterns, reveals fluctuations in working memory load and attentional focus (Y. Liu et al., 2025). Heart rate variability (HRV) is another reliable indicator, as it is typically reduced under high cognitive load and elevated stress (Buettner et al., 2021). In addition, eye-tracking metrics such as fixation duration and dispersion are recognized measures of visual cognitive processing (William & Murugesh, 2020; Zu et al., 2020), while facial expression analysis supports the detection of affective states that modulate cognitive efficiency (Sharma et al., 2021). These multimodal data streams are interpreted using rule-based thresholds or machine learning classifiers that categorize learners into zoning categories in real time. For example, a pattern of high pupil dilation, low HRV, and dispersed visual fixations may indicate cognitive overload, prompting instructional simplification through MAI. This integrative measurement approach ensures that zoning remains both empirically grounded and instructionally responsive.

8.5. Feedback-Driven Knowledge Construction (FDKC)

FDKC supports metacognitive development and schema formation by prompting learners to reflect, explain, and regulate their learning processes. Research consistently shows that metacognitive prompting enhances learning outcomes, particularly when integrated with personalized feedback mechanisms (Azevedo et al., 2010; Dignath & Büttner, 2008). In CLAM, FDKC enables intelligent tutoring systems to analyze learner-generated inputs—such as summaries, self-evaluations, or clarification requests—and tailor subsequent instructional steps (Zhu & Lv, 2023). This feedback loop structure echoes trends in learning analytics and AI-enhanced education, offering personalized and inclusive support for neurodiverse and adult learners (Baker & Inventado, 2014). As explored in the review’s analysis of SRL and learning analytics (Section 7), FDKC draws on metacognitive prompting, AI-based tutoring, and data-informed feedback loops to enhance learning outcomes (Azevedo et al., 2010; Lin et al., 2023).

8.6. Contribution and Application

The CLAM (Cognitive Load-Aware Model) framework addresses well-documented limitations of classical Cognitive Load Theory (CLT), particularly its insufficient consideration of emotional engagement, learner motivation, and individual learner profiles (Holmes et al., 2021; Sweller, 2020; Xu et al., 2021). By incorporating real-time learner data and data-driven instructional decision-making, CLAM enables more granular and personalized pedagogical responses, aligning with advancements in adaptive learning and learner modeling (Walkington, 2013). Its modular design supports implementation across a range of educational contexts, including K–12 classrooms, higher education, workforce development, and medical simulation (Makransky et al., 2021). As such, CLAM bridges foundational cognitive principles with contemporary educational technologies, offering a flexible and ethically responsive framework for designing cognitively aligned learning environments.

In summary, CLAM does not replace existing cognitive load theory; rather, it operationalizes and extends its principles in ways that address the demands of personalized, data-informed, and ethically attentive instruction in the digital age.

One potential implementation of the CLAM model can be envisioned in a secondary-level history course that incorporates immersive digital storytelling or historical simulations. Students interact with VR-based reenactments of historical events, while biometric sensors (e.g., EEG headbands, eye-trackers) collect real-time data on cognitive and emotional engagement (Makransky et al., 2019). The Dynamic Load Monitoring (DLM) system processes these signals, and the Cognitive Load Zoning (CLZ) component classifies learners into zones of Safe Load, Effort, or Overload. These classifications trigger adjustments in the Multimodal Adaptive Instruction (MAI) module, which modifies the complexity and modality of content in real time (Sweller et al., 2019).

The Learner Profile Matrix (LPM) complements this adaptivity by incorporating static learner characteristics such as prior knowledge and learning preferences. After each activity, students complete reflective or analytic tasks—such as written justifications, self-explanations, or historical argumentation—which are processed by the Feedback-Driven Knowledge Construction (FDKC) module. This module provides metacognitive scaffolding and tailored feedback (Azevedo et al., 2010), which feeds back into CLZ and LPM to refine the learner’s zoning status and profile for the next instructional phase.

These recursive feedback loops—DLM → CLZ → MAI and FDKC → CLZ → LPM—enable a responsive and reflective instructional system that supports both cognitive processing and historical empathy. By aligning cognitive complexity with real-time learner readiness, CLAM facilitates a pedagogical shift from rote memorization toward emotionally grounded, critical engagement with the past.

While the CLAM model is theoretically grounded and technologically feasible, its implementation in real-world classrooms requires both infrastructural and ethical readiness. Biometric-based adaptivity depends on access to sensors, software integration, and educator training, which may not be equally available across educational systems. Moreover, the use of real-time physiological data raises critical concerns regarding informed consent, data privacy, and potential misuse (Stahl et al., 2016). Transparent data policies, opt-in frameworks, and anonymization protocols are essential for ethical deployment. Finally, as a conceptual model, CLAM should be subjected to controlled field testing in diverse educational contexts to assess its instructional impact, operational scalability, and social acceptability (Ifenthaler & Schumacher, 2016). This model thus offers strategic insights not only for classroom-level innovation but also for policy-level decisions in technology integration and inclusive curriculum planning.

To ensure the relevance and academic integrity of the sources supporting the CLAM framework, the literature review was conducted using two primary databases: Google Scholar and Scopus. Only studies published in indexed, peer-reviewed journals between 2020 and 2025 were considered eligible for inclusion. Articles published prior to 2020 were intentionally excluded in order to reflect the most current developments in intelligent learning, instructional technology, and cognitive science.

The review focused exclusively on works situated within the broader context of learning and teaching, with particular attention to studies addressing concepts such as cognitive load, Dynamic Load Monitoring, responsive feedback, learner modeling, multimodal instruction, and feedback-driven knowledge construction. Studies involving technical implementations (e.g., AI-based tutors, biometric systems) were included only if they demonstrated pedagogical relevance and were clearly situated within an educational context.

9. Future Implications of Cognitive Load Theory in Contemporary Learning Environments

As Cognitive Load Theory (CLT) continues to evolve, it finds renewed relevance within a rapidly transforming educational landscape shaped by artificial intelligence, responsive platforms, and multimodal learning environments. Beyond its traditional applications in STEM education, CLT is now informing instructional practices across the humanities, social sciences, and professional training programs (Mayer, 2009; Sweller et al., 2011a). These shifts reflect a growing recognition that cognitive load principles—when dynamically applied—can enhance both the accessibility and effectiveness of learning, particularly in complex and individualized contexts.

One of the most promising developments is the integration of CLT into AI-driven tutoring systems, which use real-time learner data to adjust content difficulty, pacing, and modality. These adaptive systems are increasingly supported by biometric technologies such as electroencephalography (EEG), eye-tracking, and heart rate variability monitoring, which offer objective insight into learners’ cognitive and emotional states (Y. Liu et al., 2023; Sharma et al., 2021). Such tools facilitate continuous load regulation, supporting learners through timely scaffolding or simplification when signs of overload are detected.

Moreover, CLT is being recontextualized through its convergence with self-regulated learning (SRL), a framework that emphasizes goal setting, metacognitive monitoring, and strategic engagement (Azevedo et al., 2010). This synergy creates opportunities for designing learning environments that are both cognitively efficient and learner-driven.

Importantly, these advances also address the growing need for inclusive and neurodiversity-aware instruction. CLT’s principles, when combined with multimodal content delivery and personalized pacing, can better accommodate learners with ADHD, dyslexia, or sensory processing differences (Plass & Kalyuga, 2019). This is particularly relevant in hybrid, mobile, and immersive learning environments, where maintaining cognitive efficiency across diverse user profiles is essential.

To frame these future directions more concretely, we propose a synthesis of implications drawn from CLT and extended through the CLAM framework. At the theoretical level, this includes refining cognitive load constructs in relation to multimodal and affective data streams. At the research level, it involves developing more ecologically valid methods for measuring cognitive load in dynamic learning environments, especially through real-time biometric tracking. Finally, at the practical level, CLAM highlights the need for scalable, ethically responsible systems that support adaptive instruction, neurodiversity, and learner agency across formal and informal settings. Together, these directions reinforce CLT’s continued relevance while positioning CLAM as a guiding framework for cognitively aligned and ethically aware innovation in education.

These future implications underscore the versatility and continued relevance of CLT as a foundational yet evolving framework. As educational technologies advance and learner diversity becomes more visible and valued, CLT’s principles remain a stable anchor for cognitively informed instructional design. Its ongoing integration with AI, affective computing, and inclusive pedagogies points toward a new frontier in the learning sciences—one that is not only technologically enhanced, but also ethically grounded. These shifts signal a pressing need to move from abstract theory to actionable design models that can guide educators in real-world contexts.

9.1. Instructional Implications of the CLAM Model

We fully acknowledge the importance of validating conceptual frameworks through empirical research. Although the present manuscript offers a theoretical model rather than reporting on a data-driven study, it is imperative to consider pathways for testing its practical relevance and generalizability. As such, we envisage a multi-phase roadmap for future validation of the CLAM model.

Initial efforts could involve pilot studies conducted in digitally mediated classroom environments—both at the secondary and post-secondary level—where biometric tracking technologies (e.g., HRV, EEG, eye-tracking) are already in use. These could be followed by comparative experimental or quasi-experimental studies contrasting personalized instruction informed by real-time cognitive load data with conventional, non-adaptive approaches. In parallel, mixed-methods designs that combine biometric indicators with qualitative interviews and interaction log data could yield richer insights into both learning outcomes and user experience.

Ultimately, replication studies across diverse educational contexts—including formal K–12 settings, higher education, and professional training environments—would be necessary to assess the model’s scalability and contextual robustness. While conceptual in nature, the model is thus accompanied by a clear trajectory for empirical grounding and refinement.

9.1.1. Comparative Positioning and Theoretical Novelty of the CLAM Framework

Existing adaptive learning systems—such as MetaTutor (Azevedo et al., 2022) and AutoTutor (D’Mello & Graesser, 2015)—have made substantial contributions to metacognitive scaffolding and affect-aware interaction. However, these models generally rely on self-regulation prompts or emotional state detection and do not incorporate biometric diagnostics as part of real-time instructional regulation.

Similarly, recent AI-driven personalization frameworks (e.g., Bosch et al., 2015) often emphasize performance or affect-based adaptation while lacking a principled cognitive foundation grounded in load management.

In contrast, CLAM distinguishes itself by integrating continuously monitored physiological data (e.g., EEG, HRV, pupillometry) into pedagogical decision-making, supported by a dual theoretical foundation in Cognitive Load Theory (CLT) and the Cognitive Theory of Multimedia Learning (CTML). It builds upon contemporary advances in neuroadaptive and multimodal learning (Makransky & Mayer, 2022), enabling moment-to-moment adaptation based not only on performance metrics but on the learner’s actual cognitive effort.

Moreover, CLAM unifies five functionally distinct components—load monitoring, profiling, adaptive delivery, zoning, and feedback loops—into a coherent, domain-general framework that spans educational levels and promotes ethically aligned personalization. While hybrid approaches to adaptivity are emerging (e.g., Ahmad et al., 2023; Szulewski et al., 2020), none, to date, operationalize biometric-driven cognitive zoning and feedback-driven personalization as systematically or as theoretically grounded as CLAM.

The CLAM model offers a way for educators and designers to respond more deliberately to learners’ needs. By incorporating real-time data and adapting instruction dynamically, it helps maintain cognitive balance while supporting engagement. Monitoring cognitive load through DLM enables timely adjustments when learners are at risk of overload, while CLZ turns these signals into meaningful categories that inform how content should be delivered or paced (Sweller et al., 2019).

MAI takes that a step further, adjusting not just the difficulty, but also the format—whether through visual, verbal, or interactive modes—depending on what learners can handle in the moment (Makransky & Mayer, 2022). At the same time, LPM brings in a longer view, helping instruction take into account learners’ histories, preferences, and profiles (M. Chen et al., 2007). FDKC closes the loop by using what learners do—how they reflect, perform, and respond—to refine the system’s next steps (Azevedo et al., 2022).

Together, these components work not in isolation, but as part of a responsive, ethically aware system—one that evolves with the learner and reflects a broader shift toward data-informed, cognitively grounded educational practice (Ifenthaler & Schumacher, 2016).

9.1.2. Applicability Across Educational Levels and Implementation Challenges

While the CLAM framework is conceptually designed for broad applicability across educational contexts, we recognize that its implementation must be carefully calibrated to each level’s pedagogical, cognitive, and infrastructural characteristics. In Section 8.6, we offered illustrative applications across K–12, higher education, and corporate learning. Here, we extend this by clarifying that each component of the model may take distinct forms depending on the setting.

For instance, in primary education, real-time biometric integration may not be feasible or ethically appropriate in its full form. In such contexts, simplified proxies for cognitive effort (e.g., observational data, teacher-facilitated input) combined with semi-automated feedback loops may offer a more developmentally aligned approach. In contrast, secondary and post-secondary classrooms increasingly adopt learning analytics systems and, in some cases, biometric tools, enabling more sophisticated responsive instruction based on real-time data.

Professional training environments, particularly in high-stakes or simulation-based domains (e.g., aviation, healthcare), may offer the greatest potential for deploying the full CLAM suite, as these contexts often support rich learner profiling, multimodal input, and performance-based feedback cycles.

Nonetheless, we remain cautious regarding feasibility. Beyond ethical considerations, especially in relation to consent and data sensitivity in younger populations, there are logistical and cost-related barriers. The required infrastructure for biometric monitoring and data-driven delivery remains unevenly distributed across institutions, and teacher training or algorithmic transparency may limit uptake even in technologically advanced settings. These constraints underline the need for contextual piloting before scaling implementation. In this sense, CLAM should be viewed not as a prescriptive solution, but as a flexible architecture adaptable to varying degrees of technical capacity and pedagogical readiness.

10. Theoretical and Technological Limitations of Cognitive Load Theory

While Cognitive Load Theory (CLT) has been instrumental in structuring effective instructional design, its foundational assumptions increasingly reveal limitations when applied to real-world, technology-enhanced learning. Although originally developed for structured tasks in controlled settings, CLT proves less robust in complex, ill-defined environments where ambiguity, novelty, and emotional regulation are integral. The core principle of minimizing extraneous load, although pedagogically sound in many contexts, may hinder deeper engagement in tasks requiring flexible expertise or creativity (Bjork & Bjork, 2020).

Moreover, CLT does not fully account for motivational and emotional influences on cognitive processing. Evidence suggests that factors such as stress, curiosity, and task value dynamically affect learning efficiency (Boekaerts, 2011; D’Mello & Graesser, 2015). This gap has particular implications for inclusive and affective design, as learners differ not only in cognitive capacity but also in their emotional regulation and engagement patterns.

The generalizability of CLT strategies also remains problematic. The “expertise reversal effect” underscores that instructional techniques beneficial for novices, e.g., worked examples or segmented instruction, may impair learning for experts, who benefit more from problem-based approaches (Kalyuga, 2007, 2009b). Similarly, CLT’s uniform application across learner populations can neglect neurodiverse profiles, adult learners in workplace settings, and those in informal or hybrid contexts (Plass & Kalyuga, 2019).

Although the integration of biometric monitoring tools such as EEG, eye-tracking, or heart rate variability holds promise for improving the granularity of cognitive load measurement (Juliano et al., 2022; Y. Liu et al., 2023), these technologies remain financially and logistically inaccessible to many educational settings. CLAM’s Dynamic Load Monitoring (DLM) attempts to bridge this gap by proposing scalable, multimodal proxies for real-time cognitive state estimation—but empirical validation in large-scale deployment remains a challenge.

10.1. Contradictory Effects of AI-Driven and Digital Learning Environments

AI-enhanced platforms and personalized learning environments hold the potential to refine instructional personalization and increase scalability. However, recent empirical studies have highlighted several paradoxical and unintended effects that must temper overly optimistic implementation (UNESCO, 2019):

• Cognitive Overload and Fragmented Attention: Digital interfaces, despite being designed to scaffold learning, often contain distractions—pop-ups, multitasking demands, and non-essential visual elements—that elevate extraneous load and undermine deep processing (Q. Chen & Yan, 2016).
• Algorithmic Bias and Opacity: AI models trained on biased datasets may perpetuate inequities, disproportionately affecting underrepresented groups (Binns, 2018). Moreover, the “black box” architecture of most systems obscures how instructional decisions are made, complicating both transparency and pedagogical agency (Holmes et al., 2021).
• Emotional Disengagement and Isolation: While automation improves scalability, it can reduce opportunities for social presence and affective support—elements crucial for sustained motivation, especially in asynchronous environments (Zawacki-Richter et al., 2019).
• Ethical and Privacy Concerns: The use of affective computing and biometric data raises critical questions about data governance, informed consent, and learner autonomy. Inference of internal states from gaze patterns or neural signals without user transparency may breach ethical thresholds (Cukurova et al., 2022; Luckin, 2017).
• Digital Divide and Unequal Access: AI-enabled personalization presupposes access to robust technological infrastructure. Socioeconomic disparities continue to limit the equitable distribution of these innovations (Van Deursen & Van Dijk, 2019).

These tensions necessitate a cautious integration of CLT and CTML principles within AI-mediated environments. While CLAM offers a conceptual framework for ethical adaptivity, its success depends on transparent design, inclusive infrastructure, and responsiveness to social-emotional learning needs.

Beyond the technical sensitivity of biometric data, the ethical deployment of the CLAM framework demands clear and proactive safeguards. These include robust protocols for informed consent, meaningful opt-in mechanisms, and transparent communication regarding how biometric inputs contribute to instructional decisions (du Boulay, 2022).

Data minimization and anonymization should be standard practice, and data governance must adhere to relevant legal frameworks such as the General Data Protection Regulation (GDPR) and the Family Educational Rights and Privacy Act (FERPA). Particular attention must also be paid to storage duration, third-party access, and algorithmic accountability.

Importantly, educational institutions and developers share joint responsibility in ensuring that participation is voluntary, equitable, and that learners are not penalized—explicitly or implicitly—for declining to engage with biometric tracking. These principles are essential not only for ethical compliance but also for fostering trust in intelligent learning environments.

10.2. Challenges in Real-World Problem Solving

CLT also encounters limitations in guiding learners through authentic, real-world problem-solving, particularly in domains characterized by ill-structured tasks, open-ended inquiry, and dynamic decision-making. Such learning experiences typically require flexible reasoning, tolerance for ambiguity, and integration of emotional with cognitive regulation (Plass & Kalyuga, 2019).

Rigid adherence to minimizing cognitive load may constrain learners’ opportunity to develop cognitive resilience and innovation. Instead, research supports the value of “productive struggle” and moderate complexity, especially for expert learners engaged in transfer or application (Bjork & Bjork, 2020; Kalyuga, 2009b). The CLAM framework accommodates this need through the Cognitive Load Zoning (CLZ) mechanism, which distinguishes between Safe, Effort, and Overload zones, allowing for targeted instructional modulation based on learner readiness.

Furthermore, real-world tasks benefit from instructional strategies that promote self-regulation and metacognition, such as goal-setting, reflective questioning, and iterative feedback (Azevedo et al., 2010; Zimmerman, 2002). Feedback-Driven Knowledge Construction (FDKC) within CLAM operationalizes this by enabling learners to co-direct their progression through complex material, while AI systems interpret learner inputs for responsive sequencing.

11. Conclusion: Rearticulating Cognitive Load Theory for a Responsive and Inclusive Future

This review has explored the evolution of Cognitive Load Theory (CLT), tracing its foundations, refinements, and current challenges in light of contemporary educational demands. CLT, particularly when integrated with the Cognitive Theory of Multimedia Learning (CTML), has long provided a solid base for instructional design, helping educators manage the balance between intrinsic, extraneous, and germane load (Sweller et al., 2011b; Mayer & Moreno, 2003). Its principles continue to offer valuable guidance, especially in technology-rich environments where multimedia and interactivity are the norm.

Yet, as learning contexts grow more complex, individualized, and emotionally charged, the traditional model begins to show its limits. Motivational and affective factors—like curiosity, anxiety, or mental fatigue—are not fully captured in classical CLT frameworks, despite their clear impact on learning (Boekaerts, 2011). Similarly, the theory does not yet speak fully to the needs of neurodiverse learners or account for how instructional demands should shift based on learner readiness or background (Plass & Kalyuga, 2019).

At the same time, developments in biometric technology, AI-driven adaptivity, and data-informed instruction have opened up new possibilities for supporting learners in real time. Eye-tracking, EEG, heart rate variability, and other physiological data can offer moment-by-moment insights into cognitive effort. Intelligent systems, in turn, can use these data to adjust instruction dynamically—simplifying or enriching content to maintain optimal cognitive balance. These tools hold great promise, but they also raise pressing questions about privacy, consent, and fairness (Holmes et al., 2021).

In response to these shifts, we proposed the Cognitive Load-Aware Modulation (CLAM) Strategy. CLAM builds on CLT and CTML but goes a step further, offering a model for adapting instruction responsively based on biometric and behavioral indicators. The framework brings together five key components—Dynamic Load Monitoring (DLM), Cognitive Load Zoning (CLZ), Multimodal Adaptive Instruction (MAI), Learner Profile Matrix (LPM), and Feedback-Driven Knowledge Construction (FDKC). Together, these elements create a system that responds not just to how much effort a learner is exerting, but to what kind of support they need at that moment. Importantly, CLAM does not view cognitive load as something to be eliminated, but as something to be monitored, interpreted, and modulated for growth.

Beyond theory, this model has practical relevance. In Section 8.6, we described how CLAM might work in a secondary history classroom—an environment often overlooked in discussions of adaptive learning. Through tools like VR and biometric sensing, instruction could shift in real time to meet learners where they are, both cognitively and emotionally. In Section 9.1, we explored how this model can support inclusive practice and guide instructional design across various educational levels and formats, from school classrooms to simulation training environments.

Naturally, CLAM remains a conceptual framework. Future research will need to test its components in practice, assess its usability for educators, and examine its impact on learning outcomes. Further work is also needed to develop strong ethical frameworks around data use, ensuring that adaptivity enhances rather than undermines learner autonomy.

To conclude, CLT remains a cornerstone of instructional design, but its future depends on its ability to adapt. The CLAM model offers one way forward—preserving what works in CLT while responding to new realities. If applied with care, it can help create learning environments that are not only more effective, but more human.