Conference Theme
Mining for Meaning: Human Agency, Values, and Learning in the Age of AI
Human agency is central to education. Learners establish goals, make choices, seek and reject help, explore alternatives, regulate their learning, develop strategies, and decide what matters to them. Teachers interpret evidence, exercise professional judgment, adapt instruction, and negotiate competing educational goals. Yet EDM has often modeled learners and teachers primarily through observable behaviors and outcomes such as performance, engagement, persistence, or task completion. As adaptive, generative, and increasingly agentic AI systems become embedded in educational environments, understanding how these systems interact with and shape human agency becomes increasingly important.
Educational data and models are also situated within cultural, institutional, and social contexts. Constructs such as engagement, participation, collaboration, self-regulation, successful learning, and appropriate help-seeking may manifest differently across populations and learning environments. Models trained in one context may not transfer to another, while seemingly universal measures of educational success may reflect particular assumptions about learning, teaching, and education. These challenges invite greater attention to whose behaviors our models represent, where they remain valid, and which educational and human values they encode.
At the same time, advances in AI create new opportunities for EDM to contribute to the scientific understanding of learning itself. Computational systems, including neural models, reinforcement learning agents, cognitive architectures, large language models, and hybrid approaches, can serve as explicit, testable models of learning processes, operationalizing assumptions about memory, knowledge acquisition, strategy use, motivation, self-regulation, and social learning. Yet behavioral similarity or predictive performance alone does not establish that an AI system explains human learning. Establishing such connections requires attention to theoretical grounding, construct validity, mechanisms, learning trajectories, characteristic errors, causal interventions, individual differences, and ecological validity.
This year’s theme thus emphasizes the opportunity for the EDM community to connect its strengths in data, computation, modeling, measurement, and experimentation to help explain learning processes; examine the assumptions and values embedded in educational models; investigate variation across contexts and populations; and explore how educational data and AI can expand the capacity of learners and teachers to understand, choose, and act. We invite work that helps explain learning processes, examines the assumptions and values embedded in educational models, investigates variation across contexts and populations, and explores how educational data and AI can expand the capacity of learners and teachers to understand, choose, and act.
Topics of Interest
Topics include, but are not limited to:
- Modeling Human Agency
- Computational definitions and measures of learner agency
- Modeling learner choice, autonomy, goal setting, and self-directed learning
- Detecting changes in agency from interaction and process data
- Modeling help-seeking, help avoidance, exploration, and resistance to recommendations
- Distinguishing productive assistance from dependency and cognitive substitution
- Modeling learners’ capacity to act independently following support
- Agency-aware learner models
- Modeling shared and distributed agency among learners, teachers, peers, and AI systems
- Longitudinal modeling of the development of learner agency
- EDM and the Science of Learning
- Using EDM to test and refine theories of learning
- Translating verbal theories into computationally explicit models
- Computational models of memory, knowledge acquisition, strategy use, motivation, and self-regulation
- Comparing competing theoretical mechanisms using educational data
- Moving from predictive models toward explanatory models of learning
- Connecting learner models to learning sciences and educational psychology
- Modeling learning trajectories rather than only learning outcomes
- Identifying mechanisms underlying observed patterns of educational behavior
- AI as a Model of the Learner
- Neural networks, reinforcement learning, cognitive architectures, LLMs, and hybrid models as models of learning
- Criteria for determining when AI systems can serve as cognitive models
- Comparing artificial and human learning trajectories
- Human-AI error-signature and strategy comparisons
- Representation analysis and probing for educational constructs
- Ablation and intervention studies for testing proposed learning mechanisms
- Construct validity in AI-based learner models
- Distinguishing behavioral prediction from mechanistic explanation
- Individual differences and heterogeneity in computational models of learning
- Learning and Co-Adaptation with Generative and Agentic AI
- Mining and modeling learner interactions with generative and agentic AI
- Modeling human-AI learning processes from dialogue and interaction traces
- AI-supported problem solving, writing, programming, and knowledge construction
- Effects of AI assistance on knowledge acquisition, transfer, and productive struggle
- Cognitive offloading, delegation, and productive versus unproductive reliance on AI
- AI-mediated help-seeking, feedback, and support
- Effects of AI on metacognition, self-regulation, motivation, and creativity
- How learners adapt their strategies, trust, and reliance in response to AI systems
- Feedback loops between learner models, personalization, recommendations, and learner behavior
- Longitudinal effects of AI assistance and human-AI co-adaptation
- Strategic and emergent behaviors in response to automated assessment, recommendation, and personalization
- Co-adaptive learner models and dynamic personalization
- Modeling interactions among learners, teachers, and multiple AI agents
- Teacher Agency and Teaching Analytics
- Data-driven models of teacher decision-making
- Analytics that support teacher professional judgment
- Teacher interactions with AI-generated recommendations
- Modeling when teachers accept, modify, or reject algorithmic advice
- Human-AI interaction in assessment, feedback, and instructional planning
- Classroom orchestration with AI
- Effects of automation on teacher agency and expertise
- Teacher-facing explainability and actionable analytics
- Data mining to understand changes in teaching practices associated with AI
- Social and Collaborative Learning
- Modeling student and group verbal and non-verbal interactions for collaborative and/or competitive problem-solving
- Social network analysis of student and teacher interactions
- Data mining to understand how learners interact in formal and informal educational contexts
- Peer-assessment modeling
- Social learner modeling
- Values, Culture, Context, and Model Validity
- Cross-cultural and cross-context validity and generalizability of EDM models across populations, educational systems, languages, institutions, and learning environments
- Cultural and contextual variation in learning behaviors and constructs such as engagement, collaboration, help-seeking, and self-regulation
- Transfer, adaptation, and failure of models across contexts and populations
- Culturally responsive EDM, including approaches that incorporate local and community knowledge
- Identifying and examining cultural assumptions embedded in educational datasets, constructs, measures, and models
- Accounting for meaningful contextual differences while avoiding demographic essentialism
- Values and definitions of successful or desirable learning embedded in model objectives, outcomes, and educational recommendations
- Modeling diverse learner and teacher goals and alternative definitions of success beyond performance and task completion
- Multi-objective and value-sensitive modeling that addresses tensions among educational goals such as achievement, autonomy, equity, efficiency, privacy, well-being, care, and belonging
- Participatory approaches to defining educational outcomes, model objectives, and criteria for success
- Methods for making the cultural and normative assumptions of EDM systems transparent, examinable, and contestable
- Educational Recommenders and Agency-Aware Personalization
- Learner choice in instructional sequencing
- Recommenders that preserve meaningful alternatives
- Goal-sensitive recommendation
- Learner-configurable personalization
- Modeling responses to recommendations
- Effects of recommendations on subsequent learner behavior
- Feedback loops between learner models, recommendations, and learner trajectories
- Detecting when personalization narrows educational opportunities
- Balancing optimized sequencing with exploration and learner choice
- Human-in-the-loop approaches to personalization
- Educational Measurements
- Measuring agency, autonomy, curiosity, creativity, belonging, and self-efficacy
- Computational measures of deeper learning and understanding
- Measuring changes in learner capability rather than only immediate performance
- Detecting dependency, deskilling, and over-reliance on AI
- Process measures of learning and teaching
- Validating computational proxies against educational constructs
- Multimodal and multi-resource measurement
- Combining behavioral traces with self-report, qualitative, and contextual data
- Culturally sensitive measurement
- Longitudinal measures of learning and human development
- Equity, Fairness, Privacy, and Human Agency
- Fairness across cultural and educational contexts
- Differential effects of AI assistance across learner populations
- Bias in learner models and educational datasets
- Privacy-preserving modeling
- Learner and teacher control over educational data
- Data governance and meaningful consent
- Transparency and contestability of automated educational decisions
- Relationships between privacy, surveillance, and learner agency
- Participatory and community-centered approaches to educational data governance
- Domain Knowledge Modeling
- Deriving representations of domain knowledge from data
- Algorithms for discovering relationships, associations, and prerequisite structures between learning resources
- Algorithms to improve existing domain models
- Human-in-the-loop methods to collect domain knowledge models, including crowd-sourcing and expert tagging
- Reproducibility
- Replicating previous studies with larger sample sizes, in different domains, and/or in more diverse contexts
- Facilitating accessible benchmarking systems and publishing educational datasets that are useful for the community
Questions for the EDM Community
The theme particularly encourages the community to engage with questions such as:
- How can EDM move from predicting learner performance toward explaining how learning unfolds, and what evidence is needed for computational models to serve as models of human learning?
- How can educational data distinguish learning with AI from task completion with AI, including whether AI assistance builds future capability rather than only improving immediate performance?
- How can learner models and adaptive systems account for human agency, including learners’ goals and choices, and for the ways personalization itself shapes subsequent behavior and opportunity?
- How can EDM support and augment teacher judgment rather than replace it?
- How do learning constructs and models vary across cultural and educational contexts, and how can we establish when models remain valid across populations and settings?
- Which values are embedded in EDM outcomes, objectives, and recommendations, and what should educational systems optimize when goals such as performance, agency, equity, and well-being do not align?