What the Research Says

Bridging historical insights with contemporary AI challenges

While freely accessible research specifically addressing generative AI in education remains limited, there exists a wealth of relevant insights from previous decades that could inform our present challenges. Join Professor Rose Luckin as she explores what the research says.

What the Research Says

"What the Research Says" - Bridging Historical Insights with Contemporary AI Challenges

A significant challenge has emerged in the conversations surrounding AI in education: despite decades of valuable research, there remains a notable scarcity of readily accessible, robust evidence regarding AI's current educational impact. This gap becomes particularly pronounced as we navigate the integration of generative AI into educational settings.

There is a disconnect between contemporary discussions and the rich historical understanding that has emerged over many decades of AI in education research. While freely accessible research specifically addressing generative AI in education remains limited, there exists a wealth of relevant insights from previous decades that could inform our present challenges.

Join Professor Rose Luckin as she explores what the research says...

What the Research Says: June 2025 - What the research says about: Emotions as the Foundation for Self-Directed Learning in an AI-Enhanced World

I’m often asked to talk about the importance of human intelligence in an AI world. Indeed, in the last few months I’ve been asked to do this by both the Ministry of Education in Finland and the Ministry of Education in Singapore. Whenever I’m asked to focus on human intelligence, I try to draw out the differences between human and AI and focus on our unique strengths too.

For example, two areas of important difference can be found in:

1. our emotional connection to the world and to other people, and

2. our ability to reflect on our own cognition and engage in reflection and self-directed learning.

Both these areas of difference are extremely important for learning and so I thought it would be interesting to focus for this ‘What The Research Says’ on the connection between emotions and self-directed learning.

I hope you enjoy reading this - there’s a lot that we can do to support students, including in the way that we use AI, to better enable the connection between emotions and learning, including and importantly self-directed learning.

Executive Summary

This synthesis examines the critical intersection of emotions, self-directed learning (SDL), and AI in education. Drawing on extensive research including meta-analyses there is compelling evidence that emotions are not peripheral to learning but fundamental to it.

Key Findings:

Emotions drive learning outcomes: Social-emotional learning interventions can improve academic performance, whilst emotional intelligence correlates strongly with academic success
Learning is inherently social: Tomasello's shared intentionality framework demonstrates that human learning evolved as a cooperative, emotionally engaged process requiring joint attention and shared goals
SDL requires emotional competence: Effective self-directed learning depends on emotional regulation, resilience, and the ability to navigate frustration and setbacks
AI presents both opportunities and risks: Whilst AI can provide personalised emotional scaffolding and safe spaces for learning, it risks creating over-reliance, laziness, superficial engagement, and loss of human connection
Early evidence is mixed: Recent studies show AI can reduce learning anxiety and support adaptive strategies, but may also diminish self-monitoring and metacognitive awareness

Critical Implications:

Research reveals that as we integrate AI into education, we must ensure technology amplifies rather than diminishes the emotional foundations of learning. Success requires:

Educators who balance technological tools with human connection and emotional support
Technology designers who centre emotional awareness and social connection in their systems
Policymakers who protect the fundamentally human aspects of learning whilst enabling innovation

In our rush to harness AI's power for education, we must not lose sight of what makes learning meaningful: emotional engagement, social connection, and shared purpose.

The most effective learning environments will use AI to enhance—not replace—these fundamentally human elements. Our educational future depends not on choosing between technology and humanity, but on thoughtfully integrating both to support learners' emotional and intellectual development.

Introduction

In our rapidly evolving educational landscape, a critical intersection has emerged between three transformative forces: the fundamental role of emotions in learning, the imperative for self-directed learning capabilities, and the potential of AI. While each domain has been extensively researched, their convergence presents both unprecedented opportunities and complex challenges for educators, learners, and technology developers.

The evidence is compelling: emotions are not peripheral to learning but fundamental to it. Meta-analyses demonstrate that social-emotional learning interventions can improve academic performance, and emotional intelligence shows robust correlations with academic success. Yet as we enter an era where AI promises to personalise and enhance learning at scale, we face a crucial question: How do we ensure that technology amplifies rather than diminishes the emotional foundations of effective learning?

This synthesis examines what decades of research tell us about the interplay between emotions, self-directed learning (SDL), and AI in education. We explore how emotional competencies serve as prerequisites for effective SDL, how AI can support or undermine these capabilities, and what this means for designing learning environments that honour our fundamentally social and emotional nature as learners.

The Emotional Architecture of Learning

Understanding the Emotion-Learning Connection

Research spanning the past few decades provides overwhelming evidence that emotions fundamentally shape learning outcomes. For example, according to Pekrun's Control-Value Theory, students' emotions arise from their perceived control over learning activities and the subjective value they assign to academic tasks. This framework categorises emotions along two critical dimensions:

Valence: Positive emotions (enjoyment, pride, hope) versus negative emotions (anxiety, shame, boredom)

Activation: Activating emotions (excitement, anxiety) versus deactivating emotions (relaxation, hopelessness)

The impact is measurable and significant. For example, Camacho-Morles et al. (2021) found that positive emotions correlate positively with academic performance (r = 0.24), while negative emotions show negative effects (r = -0.25). These aren't merely correlations—longitudinal studies using the PALMA dataset (3,425 German students) demonstrate reciprocal effects, where emotions predict subsequent achievement and achievement predicts subsequent emotions.

The Shared Intentionality Foundation

Michael Tomasello's groundbreaking work provides crucial theoretical grounding for understanding why emotions are so central to learning. Humans possess unique capacities for shared intentionality—the ability to share mental states with others and engage in collaborative activities with joint goals. This framework reveals that learning is not merely cognitive but inherently social and emotional.

From infancy, humans engage in triadic interactions (self-other-object) that require emotional coordination. This capacity for joint attention, emerging around 9-12 months, forms the foundation for all subsequent cultural learning, including formal education. As Tomasello (2008) demonstrates, human communication is fundamentally cooperative and prosocial, driven by motives to help and share information—processes that require emotional engagement and trust.

In educational contexts, this means that effective learning depends on:

Shared goals between teachers and students
Joint attention to learning materials
Cooperative problem-solving
Emotional signalling about the value and importance of knowledge

Emotion Regulation as a Critical Mediator

Research consistently identifies emotion regulation as the critical mechanism linking emotions to learning outcomes. Children with better emotion regulation skills demonstrate:

Higher teacher-reported academic success
Better standardised test scores in literacy and mathematics
Improved student-teacher relationships
Fewer behavioural problems

The relationship appears particularly strong in early childhood, with kindergarten marking a critical transition where emotion regulation skills significantly predict academic outcomes. This aligns with neuroscientific evidence showing that the prefrontal cortex regions responsible for emotion regulation continue developing through adolescence, making this a crucial window for intervention.

Self-Directed Learning: An Emotional Endeavor

The Emotional Prerequisites of SDL

Self-directed learning—the process where individuals take initiative and responsibility for their own learning journey—requires sophisticated emotional competencies. SDL encompasses identifying learning needs, establishing goals, locating resources, managing time, and evaluating progress. Each of these components has emotional dimensions:

1. Identifying learning needs requires honest self-assessment and the emotional resilience to acknowledge gaps in knowledge

2. Setting goals involves hope and optimism about future possibilities

3. Resource location demands persistence in the face of challenges

4. Time management requires delaying gratification and managing frustration

5. Progress evaluation necessitates accepting feedback and maintaining motivation despite setbacks

Zimmerman's cyclical model of self-regulated learning highlights how emotions permeate each phase:

· Forethought phase: Emotions influence goal-setting and strategic planning

· Performance phase: Emotional states affect attention, effort, and persistence

· Self-reflection phase: Emotional reactions to outcomes shape future learning behaviours

The Social Dimension of Self-Direction

Contemporary research recognises that self-directed learning is not a solitary endeavour. Hadwin, Järvelä, and Miller (2018) identify three levels of learning regulation:

1. Self-regulation: Individual management of learning processes

2. Co-regulation: One person supporting another's self-regulatory processes

3. Socially shared regulation: Collaborative regulation where individuals coordinate processes

This social dimension aligns with Tomasello's framework, suggesting that even "self-directed" learning emerges from our capacity for shared intentionality and cooperative engagement. The most effective SDL occurs within emotionally supportive communities where learners can share goals, seek help, and celebrate progress together.

AI's Promise and Peril for Emotionally Grounded SDL

Artificial intelligence, particularly generative AI, presents a complex challenge for emotionally grounded self-directed learning. While offering unprecedented opportunities to support personalised and emotionally aware learning experiences, emerging research reveals significant risks that could undermine the very foundations of human cognitive and emotional development.

The Potential for Enhancement

Generative AI offers several promising avenues for supporting emotionally aware self-directed learning:

1. Personalised Emotional Scaffolding

AI systems can analyse individual learning patterns and emotional states to provide tailored support. For example, research by Zhou et al. (2024) demonstrates that the ease of use of AI tools can enhance self-regulation, which in turn positively impacts both critical thinking and problem-solving abilities. When learners show signs of frustration, for example, AI can offer encouragement or suggest taking a strategic break. Alternatively, when confidence is high, it can present more challenging material adapted to the learner's emotional readiness.

2. Safe Spaces for Emotional Risk-Taking

Wang et al. (2024) found that learners appreciate how ChatGPT alleviates social pressure by providing instant, non-judgmental feedback. This creates emotionally safe environments for experimentation and learning from mistakes without the fear of social evaluation that can inhibit risk-taking in traditional learning settings.

3. Enhanced Critical Thinking and Creativity

When implemented thoughtfully, AI can support cognitive development. For example, Essel et al. (2024) conducted an experimental study with 125 Ghanaian undergraduate students, finding that those who used ChatGPT for Research Methodology tasks demonstrated significantly higher scores in critical thinking, creative thinking, and reflective thinking compared to control groups using traditional methods.

4. Improved Engagement and Confidence

Research by Chea and Xiao (2024) documented significant improvements in reading comprehension among university students using AI-assisted tools. The experimental group showed greater engagement, confidence, and motivation while developing better vocabulary acquisition and critical thinking skills, devoting considerably more time to learning activities. In terms of confidence, Ruiz-Rojas et al. (2024) found that 64% of surveyed university students believed generative AI tools significantly improved their critical thinking abilities. Of course this is a double-edge sword, because those same students may not actually have improved their critical thinking abilities.

The Risks of Cognitive and Emotional Disconnection

However, mounting research evidence reveals concerning risks that threaten both cognitive development and emotional engagement:

1. Erosion of Critical Thinking and Brain Engagement

Recent MIT Media Lab research provides particularly alarming findings about AI's impact on brain function. Using EEG monitoring across 32 brain regions, researchers found that ChatGPT users had the lowest brain engagement and "consistently underperformed at neural, linguistic, and behavioural levels" compared to those using Google search or working without AI assistance. Over several months, ChatGPT users became progressively lazier, often resorting to copy-and-paste by the study's end, with essays deemed "soulless" by evaluators.

2. Over-reliance and Reduced Independent Problem-Solving

Multiple studies confirm risks of unhealthy dependency. For example, Abbas et al. (2024) discovered that ChatGPT usage was associated with increased procrastination, reduced memory retention, and poorer academic performance. Zhang et al. (2024) identified key factors contributing to AI dependency: students with lower academic self-efficacy experience higher academic stress, leading to inflated expectations from AI technology and subsequently greater dependency.

Zhai et al. (2024) found that as students become more dependent on AI dialogue systems, their ability to engage in independent analysis diminishes, with users favouring quick solutions over thorough analytical processes. This dependency appears to weaken essential cognitive skills as students accept AI-generated information without critical evaluation.

3. Generational Differences in Perception and Risk

Research by Chan and Lee (2023) reveals significant generational divides in AI adoption. Generation Z students demonstrated considerably greater optimism about AI benefits and higher usage frequency compared to Generation X and Millennial teachers, who showed greater caution and emphasis on fact-checking. Notably, teachers were significantly more likely than students to believe that students would become over-reliant on AI.

4. Diminished Creative Experience and Cognitive Effort

Mei et al. (2025) found that while ChatGPT enhanced creative output and reduced task difficulty (particularly benefiting non-native English speakers), it significantly diminished participants' perceived value and enjoyment of the creative process. This suggests that while AI can improve outcomes, the reduction in cognitive effort may harm the intrinsic satisfaction and meaning derived from creative work.

5. Inverse Relationship Between AI Literacy and Human Skills

And it seems that the more we trust the AI, the less we really engage our sophisticated thinking processes. Research by Wijaya et al. (2024) identified an inverse relationship between AI literacy/trust and crucial 21st-century skills among mathematics teachers. As AI literacy and trust increased, so did AI dependency, while self-confidence, problem-solving, critical thinking, creative thinking, and collaboration significantly decreased. Teachers with the highest AI literacy showed the lowest levels of these essential human capabilities.

Factors Influencing Effective Integration

Research reveals that the impact of AI on learning depends heavily on the implementation approach and individual factors:

Task-Specific Effects

Lee et al. (2025) found that generative AI shifts critical thinking in knowledge work from content creation to oversight and verification. While this can free capacity for higher-order thinking, high confidence in AI often reduces users' critical engagement, whereas confidence in one's own abilities supports greater critical thinking.

The Importance of Human Capability

Vaccaro et al.'s (2024) meta-analysis reveals that human-AI combinations often performed worse than the best performer alone. Strong human capability appears to be a prerequisite for effective AI collaboration, with task type significantly affecting outcomes—decision-making tasks showing performance losses when combining humans and AI, while content creation tasks showed gains.

Current Evidence and Emerging Patterns

Early studies of AI-enhanced self-directed learning reveal mixed and concerning results:

Positive but Limited Findings: Li et al. (2024) found that language learners develop adaptive strategies with ChatGPT, with the AI serving as a patient practice partner that reduces anxiety around making mistakes.
Neutral Academic Outcomes: Chun et al. (2025) found no statistically significant differences in academic achievement between traditional and AI-enhanced learning groups, suggesting AI alone certainly doesn't guarantee improved outcomes.
Concerning Cognitive Patterns: Wang et al. (2024) found that survey respondents demonstrated relatively low self-monitoring in AI-assisted learning, raising questions about whether AI might inadvertently reduce metacognitive awareness essential for self-directed learning.

The Critical Need for Thoughtful Integration

The research evidence converges on a crucial finding: the relationship between AI and emotionally grounded self-directed learning is not inherently positive or negative, rather it depends entirely on how these tools are implemented and integrated with human development priorities.

Whether we get a positive or negative outcome from these stools is up to us. As we integrate AI into educational settings, we must ensure technology amplifies rather than diminishes the emotional foundations of learning. Success requires moving beyond the false choice between embracing or rejecting AI, toward a more nuanced understanding of when, how, and for whom these tools can support rather than substitute for human cognitive and emotional development.

The most effective learning environments will use AI to enhance—not replace—the fundamentally human elements of emotional engagement, social connection, and shared purpose that make learning meaningful and transformative.

The Educator's Evolving Role

Research consistently emphasises that educators remain central to emotionally grounded SDL, even in AI-enhanced environments. For example, Wu et al. (2024) describe a bidirectional relationship between "AI-supported Teacher" and "Teacher-supported AI" approaches. The most effective implementations occur when teachers:

· Model emotional regulation and learning strategies

· Help students develop emotional vocabulary for discussing their learning experiences

· Guide students in setting emotionally meaningful goals

· Facilitate peer connections and collaborative learning

· Provide the human warmth and genuine care that no AI can replicate

Designing for Emotionally Intelligent SDL in the AI Era

For Educators: Balancing Technology and Humanity

1. Prioritise Emotional Connection: Begin with relationship-building. Research shows that positive teacher-student relationships predict both emotional well-being and academic achievement. AI should enhance, not replace, these connections.

2. Teach Emotional Literacy Alongside AI Literacy: Help students name and understand their emotions during learning. Teach them to recognise when they're frustrated, curious, or confident, and how these states affect their learning choices.

3. Use AI to Amplify Emotional Support: Leverage AI to identify students who might be struggling emotionally and provide targeted support. Use learning analytics to spot patterns of frustration or disengagement early.

4. Create Collaborative AI Experiences: Design activities where students work together using AI tools, maintaining the social and emotional dimensions of learning while benefiting from technological enhancement.

5. Model Emotional Regulation: Demonstrate how to handle frustration when AI provides incorrect information, how to persist when learning is difficult, and how to celebrate progress appropriately.

For Technology Developers: Emotion in Design

1. Build Emotional Awareness into AI Systems: Develop features that help learners recognise and reflect on their emotional states during learning, not just their cognitive progress.

2. Design for Connection, Not Isolation: Create AI tools that facilitate rather than replace human interaction. Include features for collaborative learning, peer feedback, and sharing emotional experiences.

3. Respect the Rhythm of Emotions: Build in natural breaks, celebrate small victories, and recognise that learning involves emotional ups and downs. Avoid gamification that creates artificial emotional highs and lows.

4. Maintain Transparency About AI Limitations: Be clear that AI cannot truly understand or share human emotions, helping learners maintain realistic expectations about AI relationships.

5. Support Diverse Emotional Expressions: Recognise that emotional expression varies across cultures and individuals. Avoid one-size-fits-all approaches to emotional recognition or support.

For Policy Makers: Protecting the Human Heart of Learning

1. Consider Emotional Learning: Include social-emotional learning competencies, in including in relation to AI, in the curriculum. Thus, ensuring they're not marginalised in the rush to implement AI.

2. Fund Research on Emotional Impacts: Support longitudinal studies examining how AI-enhanced learning affects students' emotional development, resilience, and capacity for human connection.

3. Require Emotional Impact Assessments: Before approving AI tools for educational use, require evidence that they support rather than undermine emotional development and human relationships.

4. Invest in Teacher Development: Provide extensive professional development on supporting students' emotional needs in AI-enhanced environments, recognising this as a core teaching competency.

5. Address Equity in Emotional Support: Ensure all students have access to both human emotional support and appropriate technological tools, recognising that emotional neglect is as harmful as academic neglect.

The Path Forward: Integration, Not Replacement

Key Principles for Emotionally Grounded SDL in the AI Era

1. Emotions First, Technology Second: Always begin with understanding and supporting learners' emotional needs. Technology should serve these needs, not override them.

2. Shared Intentionality as the Gold Standard: Design learning experiences that maintain the fundamentally cooperative nature of human learning, using AI to enhance rather than replace shared goals and joint attention.

3. Emotional Resilience Through Supported Challenge: Use AI to provide scaffolding that helps learners work at the edge of their capabilities while developing the emotional strength to persist independently.

4. Community Over Isolation: Leverage technology to build learning communities where emotions can be shared, validated, and channelled toward growth.

5. Continuous Reflection and Adjustment: Regularly assess how AI implementation affects learners' emotional experiences and adjust accordingly.

Why This Matters Now More Than Ever

As we watch the rollout of AI, the stakes could not be higher. The choices we make today about how to integrate human emotion, self-directed learning, and artificial intelligence, will shape not just educational outcomes but the very nature of human development and potential.

The research evidence shows that emotions are not obstacles to learning but are its very foundation. Self-directed learning is not a purely cognitive skill but an emotional journey requiring resilience, hope, and connection. And while AI offers powerful tools for enhancing learning, its value depends entirely on how well it serves our fundamentally social and emotional nature as human beings.

In an era of rapid technological change, the ability to direct one's own learning while maintaining emotional balance and human connection becomes not just advantageous but essential for:

· Personal Fulfilment: Finding meaning and joy in lifelong learning

· Professional Adaptability: Navigating careers that require continuous emotional and intellectual growth

· Social Connection: Maintaining genuine human relationships in an increasingly digital world

· Collective Wisdom: Solving complex global challenges that require both analytical thinking and emotional intelligence

· Human Flourishing: Developing the full range of human capacities, not just those that can be automated

The Heart of Learning in the Age of AI

The convergence of research on emotions, self-directed learning, and AI points to an important finding:

Effective learning is irreducibly human.

It involves not just the acquisition of knowledge but the development of wisdom, not just individual achievement but collective growth, not just cognitive processing but emotional engagement.

As we design educational futures enhanced by AI, we must resist the temptation to reduce learning to what can be easily measured, automated, or scaled. Instead, we must use these powerful tools to amplify what makes us most human: our capacity for shared intentionality, emotional connection, and collaborative meaning-making.

The path forward requires courage and persistence: the courage to maintain human relationships in an age of artificial ones, to value emotional development alongside academic achievement, and the persistence to insist that technology serve rather than subvert our deepest educational values. It requires wisdom to discern when AI enhances learning and when it diminishes it, when to provide support and when to allow struggle, when to connect digitally and when to insist on face-to-face presence.

Most importantly, it requires love, the love that motivated Tomasello's cooperative creatures to share knowledge across generations, that drives teachers to nurture their students' growth, and that inspires learners to persist despite challenges. No AI, no matter how sophisticated, can replicate this fundamentally human capacity for care.

So, let’s ensure that our educational technologies honour rather than hollow out the emotional core of learning. Let’s design systems that strengthen rather than substitute human connections. And let’s prepare learners not just with knowledge and skills but with the emotional wisdom to navigate an uncertain future with resilience, compassion, and hope.

We have research to guide us; the challenge now is to listen—with both our minds and our hearts.

- Professor Rose Luckin, June 2025

References

Abbas, M., Jam, F. A., & Khan, T. I. (2024). Is it harmful or helpful? Examining the causes and consequences of generative AI usage among university students. International Journal of Educational Technology in Higher Education, 21(1), 10. https://doi.org/10.1186/s41239-024-00444-7

Camacho-Morles, J., Slemp, G. R., Pekrun, R., Loderer, K., Hou, H., & Oades, L. G. (2021). Activity achievement emotions and academic performance: A meta-analysis. Educational Psychology Review, 33, 1051-1095.

Chan, C. K. Y., & Lee, K. K. W. (2023). The AI generation gap: Are Gen Z students more interested in adopting generative AI such as ChatGPT in teaching and learning than their Gen X and millennial generation teachers? Smart Learning Environments, 10(60), 1-23. https://doi.org/10.1186/s40561-023-00269-3

Chea, P., & Xiao, Y. (2024). Artificial intelligence in higher education: The power and damage of AI-assisted tools on academic English reading skills. Journal of General Education and Humanities, 3(3), 287-306. https://doi.org/10.58421/gehu.v3i3.242

Chow, A. R. (2025, June 18). ChatGPT may be eroding critical thinking skills, according to a new MIT study. TIME. https://time.com/7295195/ai-chatgpt-google-learning-school/

Chun, J., Kim, J., Kim, H., Lee, G., Cho, S., Kim, C., Chung, Y., & Heo, S. (2025). A comparative analysis of on-device AI-driven, self-regulated learning and traditional pedagogy in university health sciences education. Applied Sciences, 15(4), 1815.

Cipriano, C., et al. (2024). A systematic review and meta-analysis of the effects of universal school-based SEL programs in the United States: Considerations for marginalized students.

Durlak, J. A., Weissberg, R. P., Dymnicki, A. B., Taylor, R. D., & Schellinger, K. B. (2011). The impact of enhancing students' social and emotional learning: A meta-analysis of school-based universal interventions. Child Development, 82(1), 405-432.

Essel, H. B., Vlachopoulos, D., Essuman, A. B., & Amankwa, J. O. (2024). ChatGPT effects on cognitive skills of undergraduate students: Receiving instant responses from AI-based conversational large language models (LLMs). Computers and Education: Artificial Intelligence, 6, 100198. https://doi.org/10.1016/j.caeai.2023.100198

Hadwin, A., Järvelä, S., & Miller, M. (2018). Self-regulation, co-regulation, and shared regulation in collaborative learning environments. In D. H. Schunk & J. A. Greene (Eds.), Handbook of self-regulation of learning and performance (pp. 83-106). Routledge/Taylor & Francis Group.

Lee, H., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., & Wilson, N. (2025). The impact of generative AI on critical thinking: Self-reported reductions in cognitive effort and confidence effects from a survey of knowledge workers. In CHI Conference on Human Factors in Computing Systems (CHI '25), April 26–May 01, 2025, Yokohama, Japan. ACM, New York, NY, USA. https://doi.org/10.1145/3706598.3713778

Li, B., Bonk, C. J., Wang, C., & Kou, X. (2024). Reconceptualizing self-directed learning in the era of generative AI: An exploratory analysis of language learning. IEEE Transactions on Learning Technologies.

Mei, P., Brewis, D. N., Nwaiwu, F., Sumanathilaka, D., Alva-Manchego, F., & Demaree-Cotton, J. (2025). If ChatGPT can do it, where is my creativity? Generative AI boosts performance but diminishes experience in creative writing. Computers in Human Behavior: Artificial Humans, 4, Article 100140. https://doi.org/10.1016/j.chbah.2025.100140

Pekrun, R. (2006). The control-value theory of achievement emotions: Assumptions, corollaries, and implications for educational research and practice. Educational Psychology Review, 18, 315-341.

Pekrun, R., Lichtenfeld, S., Marsh, H. W., Murayama, K., & Goetz, T. (2017). Achievement emotions and academic performance: Longitudinal models of reciprocal effects. Child Development, 88(5), 1653-1670.

Ruiz-Rojas, L. I., Salvador-Ullauri, L., & Acosta-Vargas, P. (2024). Collaborative working and critical thinking: Adoption of generative artificial intelligence tools in higher education. Sustainability, 16(13), 5367. https://doi.org/10.3390/su16135367

Tomasello, M. (2008). Origins of human communication. MIT Press.

Tomasello, M. (2014). A natural history of human thinking. Harvard University Press.

Tomasello, M. (2019). Becoming human: A theory of ontogeny. Harvard University Press.

Tomasello, M., Carpenter, M., Call, J., Behne, T., & Moll, H. (2005). Understanding and sharing intentions: The origins of cultural cognition. Behavioral and Brain Sciences, 28(5), 675-691.

Vaccaro, M., Almaatouq, A., & Malone, T. (2024). When combinations of humans and AI are useful: A systematic review and meta-analysis. Nature Human Behaviour, 8, 2293-2303. https://doi.org/10.1038/s41562-024-02024-1

Wang, C., Li, Z., & Bonk, C. (2024). Understanding self-directed learning in AI-assisted writing: A mixed methods study of postsecondary learners. Computers and Education: Artificial Intelligence, 6, 100247.

Wijaya, T. T., Yu, Q., Cao, Y., He, Y., & Leung, F. K. S. (2024). Latent profile analysis of AI literacy and trust in mathematics teachers and their relations with AI dependency and 21st-century skills. Behavioral Sciences, 14(11), 1008. https://doi.org/10.3390/bs14111008

Wu, D., Zhang, S., Ma, Z., Yue, X.-G., & Dong, R. K. (2024). Unlocking potential: Key factors shaping undergraduate self-directed learning in AI-enhanced educational environments. Systems, 12(9), 332.

Zhai, C., Wibowo, S., & Li, L. D. (2024). The effects of over-reliance on AI dialogue systems on students' cognitive abilities: A systematic review. Smart Learning Environments, 11, 28. https://doi.org/10.1186/s40561-024-00316-7

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

Zhou, X., Teng, D., & Al-Samarraie, H. (2024). The mediating role of generative AI self-regulation on students' critical thinking and problem-solving. Educational Sciences, 14(12), 1302. https://doi.org/10.3390/educsci14121302

Zimmerman, B. J. (2000). Attaining self-regulation: A social cognitive perspective. In M. Boekaerts, P. R. Pintrich, & M. Zeidner (Eds.), Handbook of self-regulation (pp. 13-39). Academic Press.

What the Research Says: April 2025 - What the research says about: Self-Directed and Self-Regulated Learning in an AI-Driven World

Introduction

In today's rapidly evolving educational landscape, two interconnected concepts have emerged as essential for effective learning: self-directed learning (SDL) and self-regulated learning (SRL). These approaches to learning are not merely academic constructs but vital skills for navigating a world where knowledge is abundant, constantly changing, and increasingly mediated by artificial intelligence.

Self-directed learning refers to the process where individuals take initiative and responsibility for their own learning journey. This encompasses identifying learning needs, establishing goals, locating appropriate resources, managing time effectively, and evaluating progress. As research spanning several decades has shown, SDL represents more than just independent study—it embodies a proactive mindset toward personal development that extends well beyond formal education settings.

Self-regulated learning, while closely related, focuses more specifically on the cognitive, metacognitive, and motivational processes learners employ within specific learning activities. SRL involves planning how to approach tasks, monitoring comprehension, evaluating progress, and adapting strategies when necessary.

Central to self-directed learning is the concept of self-regulation, which encompasses cognitive, metacognitive, and motivational dimensions. Self-regulated learning consists of:

Cognitive elements: The intentional implementation of learning processes and strategies
Metacognitive elements: Components that require learners to consider the process of learning, such as task analysis, goal setting, strategy selection, and monitoring
Motivational elements: Factors that impact and are affected by self-regulated learning, such as persistence and confidence

These components interact cyclically through phases of forethought, performance/volitional control, and self-reflection (Zimmerman, 2000), forming a comprehensive framework for understanding how learners regulate their learning processes. The importance of social contexts in learning regulation is also recognised, identifying three levels:

Self-regulation: Individual management of learning processes
Co-regulation: One person supporting another's self-regulatory processes
Socially shared regulation: Collaborative regulation where individuals coordinate regulation processes

This social dimension is particularly relevant for technology-enabled learning environments.

The significance of these sophisticated, thinking and learning capabilities has grown in recent years. In a world where information is ubiquitous and career trajectories rarely follow linear paths, the ability to direct and regulate one's learning has shifted from advantageous to essential. Traditional educational models, designed for knowledge scarcity and predictable career paths, struggle to prepare learners for environments characterized by information abundance and rapid technological change.

The emergence of artificial intelligence only accelerates this trend. As AI systems increasingly automate routine cognitive tasks, the premium on higher-order thinking skills and adaptability grows. The future belongs not to those who possess static knowledge but to those who can continuously identify what they need to learn, efficiently acquire new skills, and thoughtfully integrate new information with existing understanding.

As we design learning environments, create educational technologies, and shape educational policy for an AI-infused future, we face a fundamental question: Will our approaches empower learners to become more self-directed and self-regulated, or will they foster dependency and diminish these essential capabilities?

The answers to these questions will shape not just educational outcomes but the very nature of human agency and potential in a technology-mediated world. By understanding the mechanisms of SDL and SRL, we can develop approaches that harness technology—including AI—to enhance rather than replace human learning capabilities, creating a future where technology amplifies rather than diminishes human potential.

Key Researchers and Contributions

Research into SDL and SRL has evolved significantly over recent decades:

Barry Zimmerman developed a cyclical model of SRL, highlighting phases of forethought, performance, and self-reflection.
Paul Pintrich integrated cognitive, metacognitive, and motivational strategies within SRL frameworks.
Winne and Hadwin introduced a detailed model focusing on conditions, operations, products, evaluations, and standards (COPES), showing how learners interact with dynamic contexts.
Monique Boekaerts emphasised the importance of personal goals, emotional regulation, and socio-cognitive aspects of learning.
Roger Azevedo and colleagues advanced research on technology-supported SRL, using real-time data and adaptive feedback.
Hadwin, Järvelä, and Miller led research into social aspects of learning regulation, including co-regulation and socially shared regulation of learning.

Technology and SDL

Technology has transformed how self-directed learning can be implemented and supported. The integration of technology into SDL environments has evolved from simple computer-based learning modules to sophisticated AI-powered systems that can adapt to individual learner needs and behaviours.

Early applications of technology in SDL focused primarily on providing access to learning materials and resources. However, as technology has advanced, more interactive and adaptive systems have emerged, offering personalized learning experiences and real-time feedback.

Current Technological Approaches

Recent research reveals several technological approaches to supporting SDL:

Adaptive Learning Systems: These technologies use algorithms to adjust content difficulty and presentation based on learner performance (Voss et al., 2011)
Learning Analytics Platforms: These tools collect and analyse data on learner behaviours to provide insights into study patterns and effectiveness (Winne & Hadwin, 2013)
Virtual and Augmented Reality: Immersive technologies that allow learners to explore complex concepts in three-dimensional, interactive environments
Mobile Learning Applications: Apps designed to facilitate on-the-go learning, providing flexibility in when and where learning occurs
Intelligent Tutoring Systems: Computer-based systems that mimic human tutors, offering personalized instruction and feedback
Web 2.0 Tools: As highlighted by Sumuer (2018), these collaborative tools play a significant role in supporting self-directed learning with technology, enabling greater learner autonomy and engagement

Artificial Intelligence in SDL

Artificial intelligence has emerged as a particularly powerful tool for enhancing self-directed learning. Several AI approaches have been developed to support various aspects of SDL:

Pedagogical Agents: These virtual characters interact with learners, providing guidance and support during the learning process. As noted by Graesser and McNamara (2010), pedagogical agents can "more consistently promote SRL versus human tutors" (p. 67) by accurately tracking dialogue moves and computing optimal next steps.
Active Learning Algorithms: Inspired by machine learning, these algorithms select training examples that maximize learning efficiency. As Castro et al. (2008, cited in Gureckis & Markant, 2012) demonstrated, such approaches can significantly reduce the amount of training needed to achieve proficiency.
Recommendation Systems: AI-powered systems that suggest relevant resources based on learner preferences, past behaviour, and learning goals.
Natural Language Processing: Technologies that understand and respond to learner questions and comments, facilitating more natural interactions with learning systems.

Benefits of AI for SDL

Research has identified several advantages of using AI to support self-directed learning:

Personalisation: AI can analyse individual learning patterns and preferences to provide tailored content and feedback. This addresses what Atkinson (1972) identified as a key challenge in education: the need to optimize learning sequences for individual learners.
Scaffolding and Fading: AI systems can provide guidance when needed and gradually withdraw support as learners develop competence, supporting the development of self-regulatory skills.
Real-time Feedback: AI can offer immediate responses to learner actions, helping them understand their progress and make adjustments to their learning strategies.
Metacognitive Support: Advanced AI systems can prompt reflective thinking and self-assessment, fostering metacognitive development that is crucial for effective SDL.

A key advantage of AI in supporting SDL relates to how it can help learners gather information effectively. Gureckis and Markant (2012) highlight that AI algorithms can help learners optimise their information sampling by:

Uncertainty Reduction: Guiding learners toward information that reduces their uncertainty about a topic
Intervention-Based Learning: Supporting causal learning through experimentation and hypothesis testing
Exploration-Exploitation Balance: Helping learners navigate the trade-off between exploiting known information and exploring new areas

Effectiveness and Outcomes

Studies examining the effectiveness of technology-enhanced self-directed learning have generally shown positive results, though the findings are not uniform:

Knowledge Acquisition: Several studies have demonstrated improved content knowledge when learners use technology to support self-directed learning (e.g. Voss et al., 2011)
Retention and Transfer: Research by Markant and Gureckis (2010) found that self-directed learning supported by technology can lead to better retention and transfer of knowledge to new contexts
Self-Regulatory Skills: Technology-supported SDL has been shown to enhance metacognitive abilities and self-regulation strategies (Winne & Hadwin, 2013)
Learner Engagement: Multiple studies report increased motivation and engagement when learners have control over their learning process through technology tools

The effectiveness of technology and AI in supporting SDL appears to be mediated by several factors:

Learner Characteristics: Individual differences in prior knowledge, metacognitive abilities, and learning preferences impact how effectively learners can utilize technology for SDL
Technology Design: The usability, accessibility, and adaptability of the technology significantly influence its effectiveness in supporting SDL
Implementation Context: The social, cultural, and institutional context in which the technology is implemented shapes how it is used and its impact on learning
Task Complexity: The nature and complexity of the learning task affect how helpful technology support can be for self-directed learners
Self-Efficacy: As demonstrated by Sumuer (2018), both computer self-efficacy and online communication self-efficacy play significant roles in determining how effectively students can engage in self-directed learning with technology

The Impact of Generative AI on SDL

The rapid evolution of Generative Artificial Intelligence (GenAI) has created unprecedented opportunities for transforming educational paradigms. However, it is early days and there is no solid body of evidence or longitudinal research studies from which we can learn. But there are an emerging set of research reports that can be informative. By synthesising findings from these recent studies, published between 2022 and 2025, we can start to identify key themes, controversies, and knowledge gaps in this dynamic field.

Key Themes

Theme 1: GenAI as a Potential Enhancement for SDL

GenAI tools show potential to enhance SDL processes. Traditional SDL activities such as seeking resources, setting learning goals, and designing learning plans can be assisted by tools like ChatGPT (Li et al., 2024, cited in Roe & Perkins, 2024). Studies have demonstrated positive relationships between GenAI use and improved learning motivation in SDL contexts.

Wang et al. (2024) found that postsecondary learners primarily use ChatGPT for brainstorming and seeking inspiration for writing. Their study noted that ChatGPT alleviates social pressure by providing instant feedback, encouraging more writing practice and revision. Li et al. (2024) revealed that language learners develop adaptive strategies with ChatGPT, continuously adjusting their interactions to optimize learning outcomes.

Theme 2: The Educator as a GenAI Guide

The literature emphasizes educators' continued importance in AI-enhanced SDL environments. Wu et al. (2024) describe a bidirectional relationship, categorizing it into "AI-supported Teacher" and "Teacher-supported AI" approaches. In GenAI-enhanced environments, teachers must "relinquish authority and take a facilitative role" while developing skills to guide students in establishing learning goals and utilizing AI tools effectively.

Educators should adopt facilitative roles in AI-enhanced learning environments, guiding learners in optimizing their interactions with GenAI while fostering SDL skills. There is a need to cultivate learners' AI literacy, digital skills, and strategic approaches to integrating AI into learning processes.

Theme 3: Personalisation of Learning

GenAI's capacity to deliver personalised learning experiences emerges as a significant advantage for SDL. Multiple studies highlight that GenAI can generate adaptive content tailored to individual learning needs and preferences. Wu et al. (2024) found that technology acceptance significantly impacts self-directed learning ability, and personalized learning experiences allow highly motivated students to more effectively harness resources for autonomous learning.

Wang et al. (2024) describe how learners develop strategies such as designing specific prompts, pre-prompting with context, and fine-tuning questioning techniques to better direct their learning with AI. Learners who view AI as a useful tool demonstrate higher levels of engagement with these technologies.

Theme 4: Approaching with Caution

Researchers consistently emphasise the need for cautious implementation of GenAI in SDL contexts. Challenges include preventing over-reliance on technology and addressing the hallucination effect, where AI generates incorrect information. Wu et al. (2024) acknowledge that "overreliance on technology without critical thinking can detrimentally affect students' learning," and emphasize developing critical AI literacy.

Wang et al. (2024) found that while survey respondents demonstrated relatively low self-monitoring in AI-assisted learning, interviewees emphasized the importance of critically validating AI-provided information. Li et al. (2024) identify challenges including the "learning optimisation gap" and the "knowledge comprehension gap" between AI-generated content and learners' capacity to integrate it into their existing knowledge framework.

Challenges and Limitations of AI and Technology in SDL

Technological Challenges

Despite the potential benefits, several technological challenges limit the effectiveness of AI and technology in supporting self-directed learning (SDL):

Algorithm Limitations: As Gureckis and Markant (2012) note, "if the learning model is incorrectly specified for the domain... the information samples acquired will be severely biased" (p. 471). This can lead to ineffective or even harmful learning experiences.
Data Privacy Concerns: Collection of learner data raises significant privacy issues that must be addressed.
Digital Divide: Unequal access to technology creates disparities in who can benefit from technology-enhanced SDL.
Technical Reliability: Issues with software, hardware, or connectivity can disrupt the learning process.

Pedagogical Concerns

Several pedagogical challenges also emerge when implementing technology and AI for SDL:

Over-reliance on Technology: Learners may become overly dependent on technological scaffolding, potentially reducing their active intellectual engagement in the learning process and limiting the development of autonomous learning skills.
Superficial Engagement: Technology-enhanced environments may encourage surface-level interaction rather than deep learning.
Confirmation Bias: As highlighted by Gureckis and Markant (2012), self-directed learners may seek information that confirms existing beliefs rather than challenging them.
Cognitive Overload: Complex technological environments may overwhelm learners' cognitive resources, particularly for novices in a domain.
Lack of Training: Students need comprehensive support to effectively engage in self-directed learning with technology, specifically :
1. Self-directed learning skills: Helping students develop the ability to identify learning needs, set goals, select appropriate strategies, and evaluate outcomes.
2. Technical competencies: Building students' proficiency with the tools and platforms used for learning.
3. Self-efficacy development: Fostering students' confidence in their ability to use technology for learning and to communicate effectively in online environments.

Controversies and Tensions

AI Implementation vs. Educational Value

A significant controversy surrounds whether Generative AI (GenAI) truly enhances educational outcomes or merely changes the learning process without substantive improvements. Chun et al. (2025) found no statistically significant differences in academic achievement between traditional and AI-enhanced learning groups.

The tension between learning and assistance is notable, with concerns that learners may become overly reliant on AI tools, potentially reducing their active intellectual engagement in the learning process.

Human-AI Balance in Learning Environments

Another controversy concerns the optimal balance between human instruction and AI assistance. While educators remain central to the learning process, their role is transforming significantly. This creates tension around preserving valuable human elements of education while leveraging AI capabilities.

Equity and Access Concerns

Significant tensions exist regarding equitable access to AI-enhanced learning opportunities. Studies have not adequately addressed "the inbuilt biases, cultural orientation, and equity concerns that are part of AI in education" (Roe & Perkins, 2024). This raises questions about whether GenAI might exacerbate existing educational disparities.

Significant Gaps in Current Knowledge

One notable gap is the disproportionate focus on higher education applications of GenAI for SDL, with relatively few studies investigating K-12 settings. K-12 students face unique challenges in SDL compared to adults: "effective self-directed learning for younger students has two major influences: internal 'self' influences and external 'other' influences." Younger learners "need more guidance on what, why, where, and how of learning" (Ali et al., 2023).

Most current studies provide snapshots of implementation effects rather than tracking developmental trajectories over extended periods. Roe and Perkins (2024) explicitly call for "longitudinal studies to assess the long-term impact of the GenAI on SDL outcomes across multiple educational and cultural contexts."

The literature predominantly focuses on text-based GenAI applications, neglecting the broader possibilities of multimodal GenAI.

While multiple studies theorize about GenAI's potential benefits for SDL, robust empirical evidence demonstrating improved learning outcomes remains limited. More rigorous experimental studies are needed to establish causal relationships between GenAI use and improved learning outcomes.

The current literature lacks sufficient exploration of how GenAI-enhanced SDL functions across diverse cultural contexts and learner populations.

Guidelines for Educators

Educators seeking to support SDL through technology and AI should:

Balance facilitation with guidance: Relinquish some authority while still providing essential guidance. Help students establish clear learning goals and develop strategies for utilizing AI tools effectively.
Design AI-enhanced SDL environments: Create structured learning spaces that promote autonomy while incorporating AI tools, allowing learners to set goals, plan learning paths, and reflect on progress with technological support.
Foster AI literacy and critical thinking: Explicitly teach students to critically evaluate AI-generated content, recognize potential biases or inaccuracies, and develop validation strategies that compare AI outputs with established sources.
Differentiate implementation by educational level: Recognize that K-12 students require more structured guidance in SDL compared to higher education learners. Younger students need additional support on the "what, why, where, and how of learning."
Integrate metacognitive scaffolding: Incorporate prompts that encourage learners to think about their strategies and learning processes, especially when interacting with AI tools.
Employ learning analytics thoughtfully: Use data from AI-enhanced learning to help learners visualize their progress and identify areas for improvement, while being transparent about how this data is collected and used.
Promote collaborative AI use: Encourage co-regulation and socially shared regulation of learning by designing collaborative activities where students can collectively interact with and learn from AI tools.
Adapt to evolving roles: Develop skills in "AI-supported teaching" and "teacher-supported AI" approaches, focusing on how to enhance both teaching effectiveness and student learning experiences.
Address equity concerns proactively: Recognize and mitigate potential barriers to equitable access, providing additional support for students who may have limited access to required technology.
Evaluate long-term impact: Regularly assess how AI implementation affects learning outcomes over time, rather than focusing solely on immediate results.

Guidelines for Technology Developers

Developers creating technology and AI tools to support SDL should:

Design for comprehensive SDL cycles: Support the complete self-directed learning process, including planning, monitoring, control, and reflection phases, rather than focusing on isolated components.
Enhance personalisation capabilities: Develop systems that deliver truly adaptive content tailored to individual learning needs, preferences, and contexts, allowing for customized learning pathways.
Bridge learning optimization gaps: Create interfaces and features that help learners fully leverage the technical affordances of GenAI, addressing the gap between AI capabilities and users' ability to utilize them effectively.
Incorporate evidence-based prompts: Include metacognitive scaffolds and prompts that guide learners through the SDL process, with adaptive fading as learners develop greater self-regulation capabilities.
Create transparent learning analytics: Develop analytics that help learners understand their own processes and progress while making the limitations of AI-generated insights clear.
Support multimodal learning: Extend beyond text-based interactions to incorporate visual, audio, and interactive elements that enhance understanding across different learning domains.
Balance automation with learner agency: Ensure AI provides suggestions and support rather than prescriptive pathways, preserving learner decision-making and ownership of the learning process.
Address cultural and contextual diversity: Design tools that can adapt to diverse cultural contexts and learner populations, recognising that SDL approaches may vary across cultural traditions.
Prioritize ethical data use: Collect only necessary data and be transparent about its use, addressing concerns about algorithmic biases, data privacy, and ideological influences.
Enable collaborative learning: Include features that support co-regulation and socially shared regulation of learning, facilitating collaborative SDL experiences.
Develop K-12 specific implementations: Create tools specifically designed for younger learners that provide appropriate levels of guidance and structure while still fostering SDL capabilities.

Guidelines for Policy Makers and Educational Leaders

Policy makers and educational leaders should:

Recognise SDL as a core competency: Update curriculum frameworks to position SDL as an essential 21st-century skill, particularly in AI-enhanced educational environments.
Invest in comprehensive teacher development: Provide substantial professional development focused on technology-enhanced SDL, helping educators transition to facilitative roles in AI-enhanced learning environments.
Develop ethical frameworks: Create comprehensive guidelines addressing privacy, bias, learner autonomy, and responsible AI use in educational settings.
Prioritise equity and access: Ensure technologies supporting SDL are available to all learners, addressing both technical access and support needs across diverse populations.
Support longitudinal research: Fund studies examining long-term impacts of AI-enhanced SDL across various educational contexts, age groups, and cultural settings.
Balance innovation with evidence: Encourage educational innovation while requiring evidence of improved learning outcomes, not just technological implementation.
Address K-12 to higher education continuity: Develop policy frameworks that support SDL development across the full educational spectrum, recognising different needs at various educational levels.
Promote cross-sector collaboration: Facilitate partnerships between educators, technology developers, researchers, and policymakers to ensure alignment of SDL approaches.
Consider sustainability dimensions: Evaluate AI implementations through environmental, social, and economic sustainability lenses, as outlined in recent research.
Create accountability frameworks: Establish mechanisms to evaluate whether AI tools are enhancing rather than diminishing SDL capabilities and educational quality

Why Supporting SDL Matters

The ability to direct one's own learning is essential in today's fast-changing world. Careers increasingly require continuous learning, complex problem-solving, and adaptability. Technology, particularly AI, offers significant potential to support SDL, but its effectiveness depends on thoughtful, research-informed implementation.

The case for the importance of SDL in today's world:

Rapid knowledge evolution: In fields where knowledge changes quickly, the ability to direct one's own learning is essential for ongoing relevance.

Lifelong learning demands: Career trajectories increasingly require continuous learning beyond formal education.

Complex problem-solving: Modern challenges require the ability to identify learning needs and acquire new knowledge and skills.

Digital transformation: Technology is transforming how learning happens, making SDL capabilities more important than ever.

Learner empowerment: Developing SDL shifts education from passive consumption to active ownership of learning.

Evolving educational roles: While educators remain central to the learning process, their role is transforming significantly toward facilitation and guidance in AI-enhanced environments.

Critical approach necessities: Research consistently emphasizes the need for developing critical AI literacy, addressing concerns about overreliance, misinformation, and algorithmic biases.

Equity considerations: Studies highlight tensions regarding equitable access to AI-enhanced learning opportunities, requiring deliberate policy and implementation approaches.

By aligning educational practices, technology design, and policy with established research on SDL and SRL, we can create environments that empower learners to take ownership of their learning journeys, preparing them for lifelong learning and success in an evolving digital landscape.

References

Ali, F., Choy, D., Divaharan, S., Tay, H. Y., & Chen, W. (2023). Supporting self-directed learning and self-assessment using TeacherGAIA, a generative AI chatbot application: Learning approaches and prompt engineering. Learning: Research and Practice. https://doi.org/10.1080/23735082.2023.2258886

Alshahrani, A. (2023). The impact of ChatGPT on blended learning: Current trends and future research directions. International Journal of Data and Network Science, 7(6), 2029–2040. https://doi.org/10.5267/j.ijdns.2023.6.010

Azevedo, R., Johnson, A., Chauncey, A., & Graesser, A. (2015). Use of hypermedia to assess and convey self-regulated learning. In Handbook of self-regulation of learning and performance (pp. 1-1237). https://doi.org/10.4324/9780203839010.ch7

Boekaerts, M. (2002). Bringing about change in the classroom: Strengths and weaknesses of the self-regulated learning approach—EARLI Presidential Address, 2001. Learning and Instruction, 12(6), 589–604. https://doi.org/10.1016/S0959-4752(02)00010-5

Hadwin, A. F., Järvelä, S., & Miller, M. (2011). Self-regulated, co-regulated, and socially shared regulation of learning. In B. J. Zimmerman & D. H. Schunk (Eds.), Handbook of self-regulation of learning and performance (pp. 65–84). Routledge/Taylor & Francis Group.

Knowles, M. S. (1975). Self-directed learning: A guide for learners and teachers. Association Press.

Pintrich, P. R. (2000). The role of goal orientation in self-regulated learning. In M. Boekaerts, P. R. Pintrich, & M. Zeidner (Eds.), Handbook of self-regulation (pp. 451–502). Academic Press. https://doi.org/10.1016/B978-012109890-2/50043-3

Pintrich, P. R., Smith, D. A., Garcia, T., & McKeachie, W. J. (1993). Reliability and predictive validity of the Motivated Strategies for Learning Questionnaire (MSLQ). Educational and Psychological Measurement, 53(3), 801-813. https://doi.org/10.1177/0013164493053003024

Pintrich, P. R., & Zusho, A. (2002). The development of academic self-regulation: The role of cognitive and motivational factors. In A. Wigfield & J. S. Eccles (Eds.), Development of achievement motivation (pp. 249–284). Academic Press. https://doi.org/10.1016/B978-012750053-9/50012-7

Roe, J., & Perkins, M. (2024). Generative AI in self-directed learning: A scoping review [Preprint]. arXiv. https://doi.org/10.48550/arXiv

Winne, P. H., & Hadwin, A. F. (1998). Studying as self-regulated learning. In D. J. Hacker, J. Dunlosky, & A. C. Graesser (Eds.), Metacognition in educational theory and practice (pp. 277–304). Lawrence Erlbaum Associates Publishers.

Winne, P. H., & Hadwin, A. F. (2013). nStudy: Tracing and supporting self-regulated learning in the Internet. In R. Azevedo & V. Aleven (Eds.), International handbook of metacognition and learning technologies (pp. 293-308). Springer. https://doi.org/10.1007/978-1-4419-5546-3_20

Zimmerman, B. J. (1986). Becoming a self-regulated learner: Which are the key subprocesses? Contemporary Educational Psychology, 11(4), 307-313. https://doi.org/10.1016/0361-476X(86)90027-5

Zimmerman, B. J. (2000). Attaining self-regulation: A social cognitive perspective. In M. Boekaerts, P. R. Pintrich, & M. Zeidner (Eds.), Handbook of self-regulation (pp. 13-39). Academic Press. https://doi.org/10.1016/B978-012109890-2/50031-7

What the Research Says: Late March 2025 - What the research says that we can learn from Roger Azevedo’s research on Metacognition and AI in Education

From 2005 to 2025 - what can we learn from the 20 years of work from Roger Azevedo on Metacognition and AI in Education.

Looking at the collection of papers that I have explored as part of the ‘What the Research Says’ project, I can trace a clear evolution in Azevedo's thinking about metacognition and the use of technological tools to enhance learning. I detail this in the second part of this article. But firstly, because I am sure this is what most readers really want to know – what can we learn and apply from his work and its development over the years?

The Individual Learner and Basic Metacognitive Processes

In his early papers such as "Using Hypermedia as a Metacognitive Tool for Enhancing Student Learning" (2005), Azevedo focused primarily on individual learners interacting with technology. His research identified key self-regulatory processes—planning, monitoring, and strategy use—that significantly affect learning outcomes.

What's striking about this early work is how Azevedo established self-regulated learning (SRL) as a theoretical framework for understanding how students learn with computer-based learning environments, particularly hypermedia. This approach was revolutionary at a time when educational technology was often evaluated simply on engagement or basic knowledge acquisition.

Implications for Today's Educators:

Focus on Process, Not Just Content: Azevedo's work reminds us that teaching students how to learn is as important as what they learn. Consider dedicating class time to explicitly teaching metacognitive strategies.
Scaffolding is Critical: Early Azevedo studies highlighted how human tutors could provide external regulation to support students' learning. Today's teachers should view themselves as metacognitive coaches, gradually releasing responsibility as students develop self-regulation skills.
Think-Aloud Protocols as Assessment: Azevedo's methodological approach using think-aloud protocols can be adapted as a classroom assessment technique. Ask students to verbalise their thinking process while solving problems to gain insight into their metacognitive strategies.

For Educational Leaders and Policymakers:

Professional Development Focus: Invest in training teachers to understand and foster metacognitive development, not just content delivery.
Technology Evaluation Criteria: When selecting educational technologies, go beyond engagement metrics and consider how tools support metacognitive development.
Long-Term Vision: Azevedo's early work reminds us that the true value of educational technology lies not in immediate test score improvements but in developing lifelong, self-regulated learners.

From Single to Multimodal Data

Another notable shift in Azevedo's work is his move from relying primarily on think-aloud protocols to integrating multiple data streams for a more comprehensive understanding of metacognition. This multimodal approach incorporated eye tracking, log files, facial expressions, and physiological sensors to create a richer picture of learners' metacognitive processes.

Simultaneously, Azevedo's conception of technological support evolved from fixed scaffolds to intelligent, adaptive systems that could respond in real-time to learners' needs. Rather than one-size-fits-all support, he began envisioning technologies that could detect metacognitive states and provide personalised assistance.

Implications for Educators:

Multiple Forms of Assessment: Azevedo's multimodal approach reminds us that student understanding can be expressed in many ways. Consider implementing diverse assessment strategies that capture different dimensions of student thinking.
Adaptive Teaching: Just as Azevedo's systems began to adapt in real-time, effective teachers adjust their approach based on ongoing student feedback. Look for signals beyond just verbal responses—body language, engagement patterns, and work products all provide insights.
Specific Metacognitive Processes: Azevedo's work became more precise about different metacognitive processes. Teachers can benefit from this specificity by targeting particular aspects of metacognition:
- Judgments of Learning (Can students accurately assess what they know?)
- Feelings of Knowing (Do students recognise when they have relevant knowledge?)
- Content Evaluation (Can students critically assess information quality?)
- Monitoring Progress Towards Goals (Do students track their advancement?)
- Monitoring Use of Strategies (Can students evaluate the effectiveness of their approaches?)

For AI Developers and EdTech Companies:

Inclusive Data Collection: Azevedo's multimodal approach highlights the importance of gathering diverse data points that capture the full range of learning experiences.
Privacy-Conscious Design: As we collect more types of data, ethical considerations become paramount. Design systems that protect student privacy while still providing valuable insights.
Transparent Algorithms: Educators and students should understand how AI systems are interpreting multimodal data and making recommendations.
Cultural Responsiveness: Ensure that multimodal data interpretation accounts for cultural differences in learning behaviours and expressions.

For Policymakers:

Investment in Teacher Training: Prepare teachers to work alongside adaptive technologies, interpreting data and making informed instructional decisions.
Ethical Guidelines: Develop comprehensive frameworks for the ethical use of multimodal learning data, especially as AI systems become more prevalent.
Equity Focus: Ensure that the benefits of adaptive learning technologies reach all students, not just those in well-resourced settings.

The Future is Now

Roger Azevedo's most recent work, particularly his 2023-2024 publications, represents the cutting edge of thinking about metacognition, learning technologies, and artificial intelligence in education. His latest ideas offer a roadmap for the future of education that educators, policymakers, and AI developers would be wise to follow.

From Individual to Contextual

Perhaps the most significant evolution in Azevedo's thinking has been moving from focusing primarily on individual cognitive processes to considering broader social, cultural, and contextual factors. His recent work emphasizes that metacognition doesn't develop in isolation but is shaped by learning environments, social interactions, and cultural contexts.

Game-Based Learning Environments

Azevedo has shifted his focus from hypermedia to serious games as powerful contexts for metacognitive development. His 2024 paper "A Taxonomy for Enhancing Metacognitive Adaptivity and Personalization in Serious Games Using Multimodal Trace Data" outlines how games can create engaging environments where metacognition can be practiced and tracked in authentic contexts.

Human-AI Co-evolution

A fascinating aspect of Azevedo's recent work is his emphasis on the co-evolutionary relationship between human learners and AI systems, where both adapt to each other. This represents a profound shift from seeing AI as simply a tool to viewing it as a partner in the learning process.

Implications for Educators:

Games as Learning Tools: Consider how serious games might be integrated into your curriculum as spaces for metacognitive development, not just content delivery.
Observable Metacognitive Behaviours: Look for specific behaviours that indicate metacognitive processes:
- Pausing during complex tasks to consider options (self-questioning)
- Checking progress indicators (monitoring progress toward goals)
- Changing strategies in response to challenges (strategy monitoring)
Personalised Metacognitive Support: Different students have different metacognitive profiles. The strongest students may need support in different areas than struggling learners.

For AI Developers:

Integration of Multiple Data Streams: Design systems that can seamlessly integrate eye-tracking, log files, verbal data, and other indicators to create comprehensive metacognitive profiles.
Adaptive Scaffolding: Create AI systems that can adjust support based on real-time metacognitive data:
- Provide more challenging content for learners who demonstrate strong feelings of knowing
- Offer simplified tasks for those struggling with strategy monitoring
- Develop personalised goal-setting features based on progress monitoring data
Ethical and Inclusive Design: Ensure that AI systems respect cultural differences in metacognitive expression and support diverse learning approaches.

For Educational Leaders and Policymakers:

Long-term Investment: Support the development and implementation of sophisticated learning technologies that foster metacognitive growth, even if immediate standardized test gains aren't evident.
Comprehensive Assessment: Develop assessment frameworks that value metacognitive development alongside content knowledge.
Teacher as Metacognitive Coach: Redefine teaching roles to emphasize facilitating metacognitive development in partnership with technology.
Equity Focus: Ensure that all students, regardless of background, have access to technologies that support metacognitive development.

PART 2 – Changes over time

Early Work (2002-2005)

In his earlier papers, such as "Using Hypermedia as a Metacognitive Tool for Enhancing Student Learning" (2005) and "Computer Environments as Metacognitive Tools for Enhancing Learning" (2005), Azevedo focuses on:

Fundamental Framework Development: He establishes SRL as a theoretical framework for understanding how students learn with computer-based learning environments (CBLEs), particularly hypermedia.
Basic Metacognitive Processes: He identifies key self-regulatory processes like planning, monitoring, and strategy use that affect learning outcomes.
Individual Learning Focus: His early work primarily examines individual learners interacting with technology, with less emphasis on social or contextual factors.
Human Tutoring as External Regulation: He explores how human tutors can provide external regulation to support students' learning with hypermedia.
Empirical Testing: His research methodology focuses on comparing conditions (e.g., scaffolded vs. non-scaffolded learning) using mixed methods, particularly think-aloud protocols.

Later Work (2018-2025)

In his more recent work, especially the 2024 paper "A Taxonomy for Enhancing Metacognitive Adaptivity and Personalisation in Serious Games Using Multimodal Trace Data," Azevedo's thinking has evolved to include:

Multimodal Data Integration: His recent work emphasizes integrating multiple data channels (eye tracking, log files, concurrent verbalizations, facial expressions, physiological sensors) to gain richer insights into learners' metacognitive processes.
Dynamic and Adaptive Systems: There's a shift from static scaffolding to dynamic, AI-based systems that can adapt in real-time to learners' metacognitive states.
Expanded Metacognitive Processes: His taxonomy now includes more specific metacognitive processes:
- Judgments of Learning (JOL)
- Feelings of Knowing (FOK)
- Content Evaluation (CE)
- Monitoring Progress Towards Goals (MPTG)
- Monitoring Use of Strategies (MUS)
- Self-Questioning (SQ)
Game-Based Learning Environments: His focus has shifted from hypermedia to serious games as powerful contexts for metacognitive development.
Human-AI Co-evolution: He now emphasises the co-evolutionary relationship between human learners and AI systems, where both adapt to each other.
Personalisation and Adaptivity: There's much greater emphasis on how systems can personalise learning experiences based on individual metacognitive patterns.
Inclusive and Equitable Approaches: His recent work considers cultural responsiveness and equity in using multimodal data and metacognitive support.

Key Shifts in Thinking

From Individual to Contextual: Azevedo has moved from focusing primarily on individual cognitive and metacognitive processes to considering broader social, cultural, and contextual factors.
From Single to Multimodal Data: His methodological approach has evolved from relying primarily on think-alouds to integrating multiple data streams for a more comprehensive understanding of metacognition.
From Static to Dynamic Support: His conception of technological support has shifted from fixed scaffolds to intelligent, adaptive systems that respond in real-time to learners' needs.
From General to Specific: His conceptualization of metacognitive processes has become more detailed and nuanced, with specific taxonomies and frameworks.
From Description to Intervention: His early work was more descriptive of metacognitive processes, while his recent work focuses more on designing interventions to enhance these processes.
From Theoretical to Applied: While maintaining theoretical rigor, his work has become increasingly applied, with concrete recommendations for designing adaptive serious games.

Azevedo's trajectory reflects the broader evolution in the field of learning technologies, moving from basic computer-based tools to sophisticated AI-driven systems that can detect, trace, model, and foster students' metacognitive processes in real-time, creating more personalised and effective learning experiences.

- Professor Rose Luckin, March 2025

What the Research Says: Mid-February 2025 - Bridging the Science-Practice Chasm to Enhance Robust Student Learning

The next research for analysis is by researchers in Pittsburgh: Kenneth Koedinger, Albert T. Corbett, and Charles Perfetti. Their research aims to bridge the gap between cognitive science research and educational practice. A framework is presented that provides a systematic approach to understanding how students learn by analysing three key components:

1. Knowledge Components (KCs) - Units of cognitive function or structure that can be inferred from student performance
2. Learning Events (LEs) - Processes through which students acquire knowledge
3. Instructional Events (IEs) - Activities designed to facilitate learning

Research discussed:

The Knowledge-Learning-Instruction Framework: Bridging the Science-Practice Chasm to Enhance Robust Student Learning

What the Research Says: Late January 2025 - The 2 Sigma Problem: the Search for Methods of Instruction as Effective as One-to-One Tutoring

Following the full-on UK Bett Show 2025, for this week's 'What the Research Says', Professor Rose Luckin explores the seminal paper by Benjamin Bloom.

Research discussed:

The 2 Sigma Problem: the Search for Methods of Instruction as Effective as One-to-One Tutoring

What the Research Says: January 2025 - Early Lessons from AI Tutoring that Matter Today

For our first 'What the Research Says', Professor Rose Luckin explores early lessons from AI tutoring that matter today. Rose covers this in the January 2025 issue of the Skinny and is holding a live talk on the material at the UK Bett Show in London on 24th January. Please note this talk will not be recorded but future sessions will be held online.

Research discussed:

What the Research Says

What the Research Says

Key Findings:

Critical Implications:

Introduction

The Emotional Architecture of Learning

Understanding the Emotion-Learning Connection

The Shared Intentionality Foundation

Emotion Regulation as a Critical Mediator

Self-Directed Learning: An Emotional Endeavor

The Emotional Prerequisites of SDL

The Social Dimension of Self-Direction

AI's Promise and Peril for Emotionally Grounded SDL

The Potential for Enhancement

1. Personalised Emotional Scaffolding

2. Safe Spaces for Emotional Risk-Taking

3. Enhanced Critical Thinking and Creativity

4. Improved Engagement and Confidence

The Risks of Cognitive and Emotional Disconnection

1. Erosion of Critical Thinking and Brain Engagement

2. Over-reliance and Reduced Independent Problem-Solving

3. Generational Differences in Perception and Risk

4. Diminished Creative Experience and Cognitive Effort

5. Inverse Relationship Between AI Literacy and Human Skills

Factors Influencing Effective Integration

Task-Specific Effects

The Importance of Human Capability

Current Evidence and Emerging Patterns

The Critical Need for Thoughtful Integration

The Educator's Evolving Role

Designing for Emotionally Intelligent SDL in the AI Era

For Educators: Balancing Technology and Humanity

For Technology Developers: Emotion in Design

For Policy Makers: Protecting the Human Heart of Learning

The Path Forward: Integration, Not Replacement

​

Key Principles for Emotionally Grounded SDL in the AI Era

Why This Matters Now More Than Ever

The Heart of Learning in the Age of AI

References

Key Researchers and Contributions

Technology and SDL

​

Current Technological Approaches

​

Artificial Intelligence in SDL

Benefits of AI for SDL

​

Effectiveness and Outcomes

​

The Impact of Generative AI on SDL

​

Key Themes

Theme 1: GenAI as a Potential Enhancement for SDL

​

Theme 2: The Educator as a GenAI Guide

​

Theme 3: Personalisation of Learning

Theme 4: Approaching with Caution

​

Challenges and Limitations of AI and Technology in SDL

Technological Challenges

​

Pedagogical Concerns

Controversies and Tensions

AI Implementation vs. Educational Value

​

Human-AI Balance in Learning Environments

​

Equity and Access Concerns

​

Significant Gaps in Current Knowledge

​

Guidelines for Educators

​

Guidelines for Technology Developers

​

Guidelines for Policy Makers and Educational Leaders

​

Why Supporting SDL Matters