2022_06_30 «Accelerating Self-Regulated Learning with AI: Opportunities and Challenges»
0 likes371 views
The document discusses the integration of artificial intelligence and advanced learning technologies to enhance self-regulated learning (SRL) through various methodologies and theoretical frameworks. It outlines current projects and research opportunities, emphasizing the importance of multimodal data and adaptive instructional interventions for fostering SRL in learners. Additionally, it highlights the role of motivation and the need for effective training of educators to support students' learning in dynamic virtual environments.
2022_06_30 «Accelerating Self-Regulated Learning with AI: Opportunities and Challenges»
- 1. Roger Azevedo, Ph.D. University of Central Florida School of Modeling, Simulation, and Training Learning Sciences Faculty Cluster Initiative Departments of Computer Science and Internal Medicine Laboratory for the Study of Metacognition and Advanced Learning Technologies Accelerating Self-Regulated Learning with AI: Opportunities and Challenges
- 2. • Science of learning with technology and multimodal self- regulated learning (SRL) trace data • Current theoretical, methodological, and analytical advances • Current and planned projects • Opportunities for future research using AI to accelerate SRL • Implications for researchers, learners, educators, and advanced learning technologies to promote SRL Overview
- 3. Measuring and Fostering Self-Regulated Learning (SRL) with Advanced Learning Technologies (ALTs) Theories, models, and frameworks of Self- Regulated Learning (SRL) Context—SRL with advanced learning technologies (ALTs) Measurement of SRL prior to, during, and following learning, reasoning, problem solving, performance, etc. Analyses of multimodal multichannel (e.g., eye tracking, log files) SRL data Intelligent and adaptive instructional interventions to foster self- regulation and learning, problem solving, etc. with ALTs Across Humans, Artificial Agents, Tasks, Domains, and Contexts
- 4. Learning, problem solving, reasoning, understanding, etc. • Psychological constructs • What is learning? • Operational definition(s) • What are the underlying neural, cognitive, affective, metacognitive, motivational, social, and contextual processes? • When, where, how, and why is learning occurring? • How do we measure it? • Research methods • When, where, how, and why do we measure it? • How do we analyze it? • Quantitative, qualitative techniques, mixed methods, computational modeling • When, where, how, and why do we analyze learning? • How do we model it? • Diagrams, human, artificial human, etc. • When, where, how, and why do we model learning? • How do we simulate it? • Simulation, immersive virtual environments, etc. • When, where, how, and why do we simulate learning?
- 5. Models of Self-Regulated Learning (SRL) (Azevedo, Bandura, Bjork, D’Mello, Dunlosky, Efklides, Graesser, Greene, Gross, Hadwin, Järvelä, Kappas, Koriat, Lajoie, Pekrun, Pintrich, Scherer, Schunk, Winne, Zimmerman)
- 6. Advanced Learning Technologies for Self-Regulated Learning (SRL)
- 7. MetaTutor (see Azevedo et al., 2022)
- 8. Experimental Set-Ups: From Lab to Real, and Virtual World Contexts
- 9. Cognitive Processes and Metacognitive Monitoring (using gaze behavior from an eye tracker; Cloude et al., 2020; Dever et al., in press)
- 10. DICE Project (Statistical Reasoning and Misconceptions)
- 11. Integration of Multimodal Multichannel Data with MetaTutor (gaze behavior, cognitive strategies, metacognitive monitoring and judgments, affective responses, social interactions, context from screen recording)
- 12. Detecting, Measuring, and Inferring SRL Processes in Real-Time (learner AND researcher, teacher, tutor, or trainer) Learner Researcher (or Teacher, Tutor, or Trainer)
- 13. MetaMentor: A System Designed to Enhance Tutors’ and Teachers’ Understanding of SRL Based on Learners’ Real- Time Multimodal Data (Azevedo, Lester, et al., 2018)
- 14. MetaMentor: A System Designed to Enhance Tutors’ and Teachers’ Understanding of SRL Based on Learners’ Real-Time Multimodal Data (Azevedo, Lester, et al., 2018)
- 15. Serious Games and Open Learner Models (OLMs) Inspectable Editable Negotiable
- 16. Virtual Learning, Research, Teaching, Training, and Assessment Platform for SRL • Explore virtual environments (e.g., Virbela) to detect, track, model, measure, infer, support, and foster SRL processes of learners across tasks, domains, and contexts • Used to teach and train students with embedded intelligent SRL agents to detect, model, track, support, and foster SRL • For example, have a Metacognition virtual room with virtual metacognitive agents capable of: • Teaching and supporting students’ learning and use of SRL • Collecting self-report, performance, and trace data on the timing, frequency of use, efficacy of use, conditions of use, application (e.g., success, efficacy), transfer to other tasks, over time, etc. • Communicating and coordinating with other SRL cognitive, motivational, and affective agents in their respective virtual rooms to foster SRL • Articulating and explaining their own and others’ (i.e., students and agents) SRL knowledge and skills while living in virtual learning environments
- 17. ZOOMBIES—Simulation of Biological Outbreak (Dr. Barrie Robinson @ University of Idaho)
- 18. Contributions and Limitations of Multimodal SRL Trace Data Azevedo & Gasevic, 2019; Azevedo & Dever, 2022; Azevedo & Wiedbusch, 2022) • Most research focuses on log-files as single channel of SRL process data • Time-scale of milliseconds to seconds to sometimes minutes • Provides mostly static post-hoc analysis but not the dynamics of SRL processes • Sequence, frequencies, and durations of activities, events, interactions, interventions, etc. • Probability of occurrence for next event • Mine sequences for dyads, triads, etc. of events • Generate hypotheses about possible underlying SRL mechanisms currently not explicit in models of SRL • Inferring cognitive and affective processes; but can we infer metacognitive and motivational processes? • Assumes equidistance between events (e.g., in log files), but what about processes with different durations and are measured at different sampling rates (e.g., 30Hz vs. 250Hz)? • Does not capture the parallel nature of SRL process • Challenging to infer high-level constructs, assumptions, processes, and mechanisms (e.g., adaptivity, dysregulation, self-efficacy, flexibility, etc.) • Limited use in real-time intelligent interventions (e.g., adaptive scaffolding, student modeling) with ALTs • Not used to measure/detect/infer qualitative and quantitative changes in SRL over time, tasks, and contexts
- 19. • SRL takes time to develop and needs to be acquired, internalized, practiced over time with the assistance of human and artificial agents to enhance transfer • Adaptive (intelligent) scaffolding is key to supporting students’ SRL with learning technologies • Multimodal multichannel SRL data is key to understanding the dynamics of SRL during learning, problem solving, reasoning, understanding, etc. • MetaLearning is key to acquiring, internalizing, using, and transferring SRL knowledge and skills across tasks, domains, and contexts • Data visualizations of students’ multimodal SRL processes are key to enhancing their understanding of SRL and the similar data visualizations are key in designing teacher dashboards that provide actionable data for effective instructional decision-making • Cognition, metacognition, and emotions are important for SRL but more attention needs to be paid to the role of motivation (as states that also fluctuate during task performance) • Training teachers to learn and use SRL in their classrooms is key in fostering their students’ SRL • AI-based immersive virtual environments hold great promise to enhance students’ SRL especially with the use of AI, NLP, computer vision, and machine learning and nanomaterials (e.g., sensors) Lessons Learned (Azevedo et al., in press)
- 20. Current Interdisciplinary Work—UCF SmartLab • Conceptual and Theoretical Issues • Define constructs, mechanisms, and CAMM SRL processes • Integrate current interdisciplinary frameworks, models, and theories of CAMM SRL processes with multimodal multichannel data (e.g., Azevedo et al., 2019; Azevedo & Dever, 2022; Azevedo & Gasevic, 2019; D’Mello et al., 2018; Efklides, 2018; Gross, 1015; Järvelä & Bannert, 2021; Lajoie, Pekrun, Azevedo & Leighton, 2020; Panadero, 2017; Pekrun et al., 2011; Scherer & Moors, 2019; Schunk & Greene, 2018; Winne, 2018; Winne & Azevedo, 2022) • Methodological and Analytical Issues • Process-oriented detection, measurement, and analytical methods • Temporally align and analyze multichannel data but balance theory vs. data-driven approaches • Temporal dynamics and synchronicity for individual learners and between agents • Quantitative and qualitative changes in SRL over time • Continue exploring data mining and machine learning techniques (inferences from high dimensional, and massive and noisy data sets, chaos theory, etc.) • Use, design, and test multimodal visualizations for learners, teachers, trainers, and researchers • Role of Human and Artificial External Regulating Agents • Role of external regulating agents (e.g., intelligent virtual humans, cyberhumans, nanobots) • Measure their impact on the acquisition, retention, use, and transfer of learners’ SRL knowledge and skills across topics, tasks, and contexts
- 21. Acknowledgements • Funding Agencies • NSF, IES, NIH, DOE, UCF, SSHRC, NSERC, CRC, CFI, CCR, Fulbright, EARLI, and Jacobs Foundation • Current and former members of the SMART Lab • Elizabeth Cloude, Megan Wiedbusch, Daryn Dever, Allison Macey, Nikki Ballelos, Dr. Nicholas Mudrick, Megan Price, Dennis Hernandez, Carina Tudela, Mitchell Moravec, Alex Haikonen, Pooja Ganatra, Sarah Augustine, Daniel Baucom, Franz Wortha, Kimani Hoffman, Lahari Revuri, Rosy Almanzar, and Jonathan Schertz • National and international collaborators • Engin Ader, Anila Asghar, Maria Bannert, Reza Feyzi Behnagh, Gautam Biswas, François Bouchet, Rafael Calvo, Analia Castigliani, Min Chi, Cristina Conati, Jennifer Cromley, Shane Dawson, Lisa Dieker, Melissa Duffy, Ian Garibay, Dragan Gašević, Arthur Graesser, Jeffrey A. Greene, Alexander Groeschner, Varadraj Gurupur, Jason Harley, Caridad Hernandez, Bari Hoffman, Charles Hughes, Eunice Jang, Sanna Järvelä, Joseph Kider, Susanne Lajoie, Joseph LaViola, Ronald Landis, James Lester, Amanda Major, Rebeca Cerezo Menéndez, Tova Michalsky, Inge Molenaar, Daniel Moos, Krista Muis, Susanne Narciss, Mark Neider, Soonhye Park, Reinhard Pekrun, Jose Carlos Núñez Pérez, Jonathan Rowe, Michael Serra, Mindy Shoss, George Siemens, Gale Sinatra, Robert Sottilare, Michelle Taub, Dario Torre, Damla Turgut, Gregory Trevors, Philip Winne, and Joerg Zumbach Thank you for your attention Questions? Collaborations? roger.azevedo@ucf.edu