Presentation on theme: "The Ecological Approach to E-Learning Gord McCalla ARIES Laboratory Department of Computer Science University of Saskatchewan Saskatoon, Saskatchewan."— Presentation transcript:
The Ecological Approach to E-Learning Gord McCalla ARIES Laboratory Department of Computer Science University of Saskatchewan Saskatoon, Saskatchewan CANADA
My Research Perspectives My background –37 years in AI research (I started when I was 4!) –first 10 years in natural language dialogue and knowledge representation –since then mostly artificial intelligence in education (AIED) and user modelling (UM) Current research areas –AIED –user modelling –multi-agent systems –recommender systems –some natural language pragmatics stuff –virtual learning communities
Talk Outline AIED as a crucible for research Overview of my research projects Finding coherence in my research projects the ecological approach Four ecological projects I-Help active modelling of learners research paper recommender LORNET Theme 3 What does it all mean for AI and AIED?
Artificial Intelligence in Education My research is situated in the area of artificial intelligence in education (AIED): advanced systems to support human learning AIED –is an applied area of AI (and education) –draws from a wide variety of disciplines: education, psychology, sociology, anthropology, computer science (AI): need advanced technology and advanced social science –emphasizes building working systems to be used with real users (learners) –usually puts the learner at the centre: learner modelling –is not concerned with formal issues of soundness, completeness and consistency, but with the practical issues of robustness effectiveness context change resource constraints
AIED is a Crucible for AI Research AIED is AI-complete, perhaps human knowledge-complete Is it tractable? YES –the domain is naturally limited –the focus is on information not the physical world –the learner is naturally constrained –the learner is naturally forgiving –there are many humans already involved in supporting learners, including teachers and the learners themselves –there is much research to draw on from a wide variety of disciplines
My Current Research: Apparent Chaos? My current research projects –LORNET (Learning Object Repository Network): NSERC network of centres of excellence: major national project (Simon Fraser U, TelUQ, Montreal, Saskatchewan, Waterloo, Ottawa) Theme 3: active and adaptive learning objects (with Greer, Vassileva, Deters, Cooke) –research paper recommender system (Tang) –capturing user goals in purpose hierarchies for just in time active user modelling (Niu, with Vassileva) –open learner modelling in an active context (Hansen) –new agent negotiation paradigms (non-monotonic offers, strategic delay, ignorance-based counter argument) (Winoto, with Vassileva) –impeding spread of delusion in agent models (Olorunleke) –enhancing social capital in virtual learning communities (Daniel, with Schwier) –data mining patterns of learner interaction with an e-learning system (Liu) –mapping folksonomies of meta-tags on learning objects (Bateman) Is there some whole emerging from these parts??
Bringing Order out of Chaos! A number of forces are driving systems that support learning: there is increasing fragmentation of –culture each learner embedded in cyberspace, has local perspectives connecting to huge global world of information and other people –learning knowledge flows through virtual communities to/from the learner, and transforms en route much learning happens just in time, when learner needs to know –teaching teaching becomes support for learning, in context of learners goals –technology boundaries of software blur: importing/exporting computation behaviour of such software systems will be emergent, like an ecosystem, fundamentally unpredictable
Bringing Order out of Chaos! Need to build AIED systems that are consistent with the fragmented perspective –software architecture multi-agent –knowledge base dynamic, oriented around change not consistency –learner modelling just in time understand learners purpose track changes model communities, not just individuals –pedagogical strategy nuanced, supportive, context sensitive take advantage of communities –research sources look broadly in computer science and to the social sciences and beyond
The Ecological Approach I have been working on an AIED architecture consistent with the fragmented perspective: the ecological approach It has the following characteristics: –the learning environment all learning materials are created as learning objects learning objects can range from relatively inert text objects through fully interactive immersion environments learning objects may be at various grain sizes, with one learning object potentially breaking down into subsidiary learning objects the learning objects are in a learning object repository new learning objects can be incorporated into, and old objects retired from, the repository the learning objects can have many associative links to each other and to the outside world learners have final control over which learning objects they select and how they interact with them
The Ecological Approach Characteristics of the ecological architecture –the AIED system learners are represented in the learning object repository by personal agents each personal agent advises its learner on how best to interact with the learning object repository, essentially the custodian of pedagogical advice; many types of advice –recommend a learning object or a sequence of objects –provide diagnostic advice to the learner –find a helper for the learner, a human tutor or peer –help the learner find a learning community each personal agent has on board a model of their learner and possibly models of other learners as a learner interacts with a learning object, the personal agent is always in the loop, advising the learner according to the learners goals and the agents pedagogical purposes, and actively updating its model(s) after a learner has interacted with a learning object, a copy of the learners model, as kept by the personal agent, is attached to the learning object over time, learning objects will be adorned with learner models of many learners (and even, possibly, the same learner many times) these learner model instances can be mined for useful information
personal affective learning/cognitive style current goal(s) previous learning objects CHARACTERISTICS EPISODIC trace of learners interactions learners evaluation of object learners view of content outcomes Learner Model Instance
The Ecological Approach Two key technologies –active modelling each personal agent tries to keep track of the learners current purpose(s) it then mediates its interactions with the learner in ways appropriate to the learners purpose(s) and its own pedagogical goals –it only uses (or computes) information about the learner that it actually needs –the learner model is actually just a residue of many such purpose-based active computations –context is thus central: the learner, other humans, resources, purposes and goals –mining learner model instances to find out which learning objects are relevant to a learner for their purpose(s): learning object recommender system to find a sequence of such objects: instructional planning to find out which learning objects are useful, not useful, or no longer useful: intelligent garbage collection to find peers with appropriate characteristics: help finding to find groups of learners with appropriate shared attributes: building learning communities to find out what happened to a learner or learners: empirical evaluation
The Ecological Approach The approach is ecological –the environment is populated by many agents and learning objects (possibly changing over time) –the agents and objects constantly accumulate more and more information –there is natural selection as to which objects are useful: could prune useless objects –there are ecological niches based on purposes: certain agents and learning objects are useful for a given purpose, others arent –the whole environment evolves and changes naturally through interaction among the agents and on-going attachment of learner models to learning objects
The Ecological Approach The ecological approach impacts many computational issues in AI and other areas of CS –various traditional AIED topics, especially learner modelling and instructional planning –various application level agent topics, especially agent negotiation and agent modelling –various system level agent topics, especially scalability and adaptivity –data mining and clustering, especially to actively compute patterns connecting particular types of learner to particular types of outcomes –collaborative filtering and case-based reasoning, which essentially underlie much of the active decision making
Current Ecological Research Projects I-Help: the font –Greer, McCalla, Vassileva, Deters, Cooke, Kettel, Bull, Collins, Meagher, graduate and summer students Active learner modelling: the paradigm –Vassileva, McCalla, Greer, graduate students Research paper recommender: the prototype –Tiffany Tang, McCalla (supervisor) LORNET: the critical mass –McCalla, Greer, Vassileva, Deters, Cooke, Brooks, Winter, graduate students
I-Help: Supporting Peer Help Two components –I-Help Pub: open peer forum –I-Help 1-on-1: find a ready, willing, able peer Agent-based –personal agents representing learners and applications Fragmented learner modelling –each agent keeps models of other agents Testing –wide-scale deployment of Pub (1000s of users) –pilot studies of 1-on-1 Current and future directions –mining Pub to supply information for 1-on-1 –full integration and effective performance
Active Learner Modelling learner models arent stored, but are computed in context main context elements: learners, purposes current investigations: –purpose hierarchies in e-commerce domain: purpose is to match a user to a stock broker agent (Niu) can the domain be covered? can you get purpose re-use? –open active modelling: in domain with many purposes: supporting learners and teachers (Hansen) how and when do you open a learner model that doesnt exist?
Research Paper Recommender Tiffany Tangs Ph.D. thesis recommending papers to graduate students preparing for research in a domain (eg. data mining) learner models of readers attached to papers recommendations made by clustering learners according to these models and predicting usefulness of papers for the student based on the cluster they map to most of the research has been investigating what pedagogical features should underlie the recommendation
LORNET Project Five year NSERC-sponsored research network investigating learning object repositories: –theme 1: interoperability (SFU) –theme 2: aggregation (TelUQ) –theme 3: active and adaptive learning objects (U. Sask.) –theme 4: learning object mining (U. Waterloo) –theme 5: multi-media and learning objects (Ottawa U.) –theme 6: integrative theme: telelearning operations system (TelUQ, and the rest)
LORNET - Theme 3 explore ecological approach to capturing and using information about learners (McCalla) MUMS user modelling middleware (Brooks, Winter) instructional planning and recommending through agent negotiation (Vassileva) –personal agents and agents representing learning objects granularity of learning and learning objects (Greer) privacy (Greer) learning object (agent) reliability and scalability (Deters) design, construction, deployment, and evaluation of application systems –in partnership with industrial sponsors (TRLabs, Parchoma Ltd.) –two entirely on-line courses with 1000s of learning objects: CS service course; CS readiness course –module of first year CS service course fully wired for ecological data collection: will be mined (Liu) and issues in meta-tagging will be explored (Bateman)
The Appeal of the Ecological Vision learning objects are activated: they are not passive, but take on responsibilities for their use in support of learning learners are in the loop: personal agents allow learners to be part of the educational environment focus is on end use: essentially learning objects are tagged by models of the learners who use them, not by context-independent content tags from a pre- defined ontology approach is ecological: as end use experience accumulates, there can be an ever more refined understanding of what works for whom
The Appeal of the Ecological Vision decision making is contextual: information is actively interpreted in context and as needed for more appropriate reactions approach is extensible and adaptable: the agent- based approach allows new learning objects and learners to be added, old ones to be deleted approach is modular: agent approach localizes decision making and improves robustness approach supports diversity: learners, applications, and learning objects can be integrated into one system, unified by the agent metaphor
Is the Ecological Approach Tractable? computational issues –how much can be done actively –space-time trade-offs –can purposes and learner models constrain the mining, clustering, and filtering algorithms –can purposes cover a domain and be re-used in other domains –can learner models be standardized and shared social issues –what kinds of pedagogy can be supported –advantages of e-learning application environment can be constrained learner can be constrained feedback from learner is natural and serves a pedagogical purpose
Déjà vu? Doesnt this seem somehow familiar? –active modelling: procedural approach –fragmented technology: frames/actors –associative links among learning objects: semantic networks –looking outside of AI for other paradigms –building big systems and seeing if anybody salutes! These were big AI issues in the 1970s –good old fashioned AI (GOFAI) –what goes around, comes around: the cycle of research Isnt it somehow different? –data-centric: machine learning was not central then –emphasis on end-use context: context was usually ignored then –needs powerful computational engine: not available then
Conclusion What works for AIED may work for many AI application areas –computer games, natural language understanding, AI-based e- commerce, even computer vision AIED forces deep issues to be grappled with –much current AI is exploration of algorithm space or theoretical issues without the reality check provided by applications such as e-learning –precision in a vacuum is indeed a vice! AIED is thus a crucible for AI research Can AIED once again be a mainstream area of AI, feeding ideas into AI as well as vice versa?
Questions, Comments, Interactions ? Acknowledgements –my graduate students past and present –my colleagues in the ARIES Laboratory –our research associates past and present –funding from the Natural Sciences and Engineering Research Council of Canada discovery grant LORNET networks grant –private sector support: TRLabs, Parchoma Consulting Ltd.
Some References –G. I. McCalla, The Ecological Approach to the Design of E-Learning Environments: Purpose-based Capture and Use of Information about Learners. Journal of Interactive Media in Education, Special Issue on the Educational Semantic Web (eds. T. Anderson and D. Whitelock), May 2004. http://www-jime.open.ac.uk/2004/1 http://www-jime.open.ac.uk/2004/1 –J. Vassileva, G.I. McCalla, and J.E. Greer, Multi-Agent Multi-User Modelling in I-Help. User Modeling and User-Adapted Interaction J., Special Issue on User Modelling and Intelligent Agents (E. André and A. Paiva, eds.), 13 (1), 2003, 1-31. –G.I. McCalla, The Fragmentation of Culture, Learning, Teaching and Technology: Implications for the Artificial Intelligence in Education Research Agenda in 2010. Special Millennium Issue on AIED in 2010, Int. J. of Artificial Intelligence in Education, 11, 2000, 177-196. Contact me at email@example.com