Instructional Science 30: 31–45, 2002.© 2002 Kluwer Academic Publishers. Printed in the Netherlands.

31

Beyond intelligent tutoring systems:Using computers as METAcognitive tools to enhance learning?

ROGER AZEVEDOUniversity of Maryland, College of Education, Department of Human Development, CollegePark, MD 20742-1131, USA (E-mail: ra109@umail.umd.edu)

Received: 24 April 2001; accepted in final form: 7 June 2001

Abstract. Framed by the existing theoretical and empirical research on cognitive and intelli-gent tutoring systems (ITSs), this commentary explores two areas not directly or extensivelyaddressed by Akhras and Self (this issue). The first area focuses on the lack of conceptualclarity of the proposed constructivist stance and its related constructs (e.g., affordances, situa-tions). Specifically, it is argued that a clear conceptualization of the novel constructivist stanceneeds to be delineated by the authors before an evaluation of their ambitious proposal tomodel situations computationally in intelligent learning environments (ILEs) can be achieved.The second area of exploration deals with the similarities between the proposed stance andexisting approaches documented in the cognitive, educational computing, and AI in educationliterature. I believe that the authors are at a crossroads, and that their article presents aninitial conceptualization of an important issue related to a constructivist-based approach tothe computational modeling of situations in ILEs. However, conceptual clarity is definitivelyrequired in order for their approach to be adequately evaluated and used to inform the designof ILEs. As such, I invite the authors to re-conceptualize their framework by addressing howtheir constructivist stance can be used to address a particular research agenda on the use ofcomputers as metacognitive tools to enhance learning.

Keywords: intelligent learning environments, intelligent tutoring systems, metacognition,self-regulated learning

Learning involves more than just shifts in cognitive states or objectsembedded in a computer-based learning environment. Learning is a complexphenomenon that includes an intricate and complex interaction betweenneural, cognitive, motivational, affective, and social processes. Most educa-tional researchers have traditionally adhered to a specific theoretical frame-work (e.g., information processing theory) or a philosophical stance (e.g.,constructivism). This theoretical attachment has led to several debates amongresearchers (e.g., Greeno, Anderson, Simon, Reder) regarding the operationaldefinitions of constructs (e.g., symbols, affordances), theoretical perspect-ives and philosophical stances (e.g., IPT vs. social constructivism), unitsof analysis (e.g., individual knowledge states vs. sociohistorical accounts oflearning), methodological approaches (e.g., verbal protocols and cognitive

32

modeling vs. discourse analysis) and uses of technology for learning(e.g., modelers, non-modelers, and middle-campers; and, learning from andlearning with technology). We must acknowledge that these debates are acritical part of the evolution of learning theories and reflect the developmentaland cyclical nature of theory building, testing, and refinement.

My commentary on Akhras and Self’s (this issue) article is framed bythe existing theoretical and empirical literature on cognitive and intelligenttutoring systems (ITSs), and explores two areas not directly or extensivelyaddressed by the authors. The first issue focuses on the lack of conceptualclarity of the proposed constructivist stance and its related constructs (e.g.,affordances, situations). I will argue that a clear conceptualization of theirnovel constructivist stance needs to be delineated by the authors before anevaluation of their ambitious proposal to model situations computationallyin intelligent learning environments (ILEs) can be achieved. The secondarea of exploration deals with the similarities between the proposed stanceand existing approaches documented in various cognitive, educational, andcomputing literatures. Their article (this issue) presents an initial concep-tualization of an important issue related to the computational modeling ofsituations in ILEs. However, conceptual clarity is definitively required inorder for their approach to be adequately evaluated and used to inform thedesign of ILEs. As such, I invite the authors to re-conceptualize their frame-work by addressing how their constructivist stance can be used to address aparticular research agenda on the use of computers as metacognitive tools toenhance learning.

Cognitive research and Intelligent Tutoring Systems (ITSs)

One of the few groups to successfully adopt the traditional ITS approachand empirically demonstrate the effectiveness of their theory and tutors is theACT-R group (e.g., Koedinger & Anderson, 1997). Anderson and colleagues’ACT-R theory of cognitive skill acquisition (Anderson, 1983, 1993) is basedon extensive psychological experimentation and has been used to inform thedesign of the ACT-R computer tutors (Anderson et al., 1995). Their cognitiveresearch and tutor development is well documented in the literature andreflects the evolution of theory building, testing, and refinement based onempirical evidence from laboratory studies and studies of the effectiveness ofthe ACT-R tutors. They have been extremely successful in translating theirtheory into instructional principles for their computer tutors (e.g., Koedinger& Anderson, 1998).

Part of the cyclical nature of theory building is exemplified by the group’srecent extension of the ACT-R theory to include both cognitive and “atomic”

33

aspects of learning. The ACTR/PM (Anderson & Labiere, 1998) theory ofcognition now includes learning mechanisms that can account for attentionalas well as motor aspects of learning. Their new theory is being tested withboth real-world tasks (e.g., air traffic controllers) and computer tutors (e.g.,using eye-tracking data to examine how students examine the interface oftheir computerized tutors). The cycle of theory building, refinement, andevaluation, and revision will continue. Their experimental and tutor datawill drive them to refine their theory, which will subsequently be used toinform the design of their tutors, and so on. Perhaps some day the ACT-RP/M theory may include Newell’s (1989) neural and social bands of humanlearning. It is critical to highlight the ACT-R group as an “ITS success story”in light of the criticisms raised by Akhras and Self (this issue). First, it isimportant to recognize the role of the ACT-R group in the development ofearly cognitive research that was the impetus for the ITS development byfocusing on student modeling, knowledge tracing, knowledge representation,tutoring interventions (e.g., flagging), and the role of immediate feedback.As such, I assert that ITSs have been an important part of early cognitivetheory and have contributed to our understanding of cognitive processesand learning mechanisms. They have also contributed to the computationalmodeling of situations (e.g., algebra, computer programming), and have hada tremendous impact on early and contemporary ITS research. It is thereforeunfair for Akhras and Self (this issue) to fail to acknowledge the contributionsof cognitive theory and ITS research within the educational computing andAI in education community.

Second, the ACT-R group’s approach represents a theoretically-drivenand empirically-based approach used to inform the design of ITSs. Incontrast, the authors (this issue) focus heavily on the computational aspects ofmodeling situations in ILEs, without identifying a firm theoretical or philo-sophical grounding. This is a critical issue that stands in stark contrast ofthe theory-building by the ACT-R group. In fact, their approach is akin toother atheoretical, technological-driven and intuition-based approaches to thedesign of CBLEs.

Theoretical and empirical basis for the design of CBLEs

Akhras and Self (this issue) distance themselves from the traditional ITSapproach by proposing a constructivist philosophy of learning, and they raisecertain issues which, according to them, should determine the foundationfor intelligent learning environments (ILEs). It is difficult to evaluate their“novel” philosophical stance because as I will argue below, lacks conceptualclarity, lacks operational definitions of critical constructs, and is presented as

34

an unfair and incomplete comparison between what has been accomplished(e.g., ACT-R’s cognitive research and tutor effectiveness) and their proposedposition (Akhras & Self, this issue).

Their attempt to distance themselves from traditional ITS, by focusingon modeling the domain in terms of situations instead of knowledge struc-tures, and evaluating learning process rather than the product, and the notionthat opportunities for learning arise from affordances of situations ratherthan being provided on the basis of teaching strategies, is unclear and lacksconceptual clarity.

First, it is difficult to evaluate the adequacy of Akhras and Self’s(this issue) constructivist philosophy of learning because it is incomplete.Although not acknowledged by the authors, their approach is in many waysvery similar to the “modelers” approach (see Derry & Lajoie, 1993; Lajoie,2000) to computer-based learning environments. The “modelers” (e.g., ACT-R group) approach is based on the fact that a student’s learning can bedetected, traced, monitored, and evaluated by following hi/her problem-solving steps. In the case of ACT-R, this approach is not based on aphilosophy of learning, but rather on several mechanisms underlying theACT-R theory of skill acquisition which make explicit predictions aboutlearning, knowledge representation and use, performance, and transfer. Inthe both the theory and tutors, these mechanisms work in coordination withan underlying cognitive model of a well-structured domain (e.g., algebra) toensure that the student learns all the prerequisite skills and ultimately mastersthe domain.

Due to the limitations found with the traditional ITS approach, Akhrasand Self (this issue) propose a new constructivist philosophical stance. Theirstance emphasizes “different values and may require an entirely differentarchitecture of intelligent system to support its philosophy of learning.” Itis my position that their stance needs to be re-conceptualized and testedbefore it is presented as a novel constructivist approach for ILEs. First, it isunclear why the authors (this issue) fluctuate between the different “flavors”of constructivism, including von Glasersfeld, Piaget, Greeno, Brown, Collins,Duguid, and Jonassen, to name just a few. It is also unclear why we needa “new” constructivist approach, and even if we did, why would one eventry to amalgamate certain components of existing constructivist frameworks?The existing constructivist frameworks are quite diversified and do notalways share the same view of learning, including the uses of technology forlearning. For example, Jonassen’s (2000) constructivist view of computers asmindtools is different from Lajoie’s (1993, 2000; Lajoie & Azevedo, 2000)view of using computers as cognitive tools for enhancing learning. Thesedifferences between frameworks and uses of technology for learning must

35

be acknowledge and not overlooked. Similarly, new theoretical approachesand philosophical stances (e.g., Akhras & Self, this issue) need to build onprevious existing research and ILEs.

Akhras and Self’s (this issue) presentation of their stance reflects severalproblems mentioned above. First, they borrow terms from several existingframeworks and do not operationally define them. For example, they incor-porate several constructs/assertions/statements including, “knowledge is indi-vidually constructed from what learners do . . . and cannot be objectivelydefined, . . . autonomous role of the learner, focus on process.” They need toacknowledge that most of the terms they borrow are derived not only fromother constructivist frameworks but also from cognitive theories and modelsof learning and instruction. For example, many cognitive researchers as wellas constructivists view the learner as a constructor of his/her own knowledge(e.g., Mayer & Wittrock, 1996; Chi, 2000); cognitive theory too focuses onthe process of learning (e.g., Anderson & Labiere, 1998; Newell & Simon,1972); and many ITSs model students based on an a detailed analysis of thelearning process (e.g., Anderson et al., 1995).

Second, the authors need to explicitly define terms and clarify severalstatements, such as “an ILE should be attuned to features of the learner, theenvironment, and the interaction between learner and environment that differin fundamental ways from the features that are relevant to ITSs”. How is“attuned” defined within their framework? What does it mean for an ILE tobe attuned to the learner? How is their proposed ILE attunement to the learnerdifferent from the type provided by traditional ITSs?

Third, it is unclear why the authors (this issue) assert the need to modelsituations, evaluate learning, and ways of promoting learning by the system.Why should all this be accomplished if it is not theoretically driven, andeven it is were, how could the present technological limitations (e.g., naturallanguage processing, interpretation of intonation and gestures) be overcomein order to model situations to their fullest extent?

Overall, I would disagree with the authors’ opinion that their constructiviststance (this issue) presents an alternative to the traditional ITS architecture.On the contrary, their stance overlaps with much of what has been calledthe “modelers” and “middle campers” approaches to using computers ascognitive tools for enhancing learning (see Lajoie & Derry, 1993; Lajoie,2000, for an extensive review of these two perspectives).

From modeling knowledge structures to modeling situations

As with other aspects of their argument, Akhras and Self (this issue) find itnecessary to go from modeling knowledge structures to modeling situations.

36

What is a situation? What constitutes a situation? Which components and/oraspects of a situation should be modeled and why? What are the differentlevels of a situation? How do the different levels of a situation reflect theirconstructivist stance? Are neural, aperceptual, physical, cognitive, socio-cultural, socio-historical, affective, metacognitive, motivational, linguistic,gestural, and self-regulation factors part of a situation? Which of these factorsshould be modeled and why? How does a constructivist theory account forthe different levels of human learning? How do we detect, trace, monitor,and model the complex interactions between factors in a situation? How dothese factors interact over time (i.e., during learning “outside” the ILE – inother situations and with other agents) and with repeated utilization of theILE? These are extremely complex issues which Akhras and Self (this issue)need to begin to address, clarify and explicitly discuss so that we can trulyappreciate the contribution of their framework and research on ILEs. Other-wise, it is extremely difficult to determine the value of their contribution andproperly evaluate it vis-a-vis existing constructivist frameworks and researchon ILEs (for an extensive review see Jacobson & Kozma, 2000; Jonassen,2000; Lajoie & Derry, 1993; Lajoie, 2000; Shute & Psotka, 1996).

Is a sentence, a math problem, a clinical case, or a modeling and simu-lation tool (e.g., Stella, Model-It) a situation, or is it part of a situation?Does the learner have to be working alone or collaboratively to be part ofa situation? What other parts of the “situation” make it a situation? How doesthe computational modeling of the (part of the) situation correspond to theauthors (this issue) philosophical stance? Which parts of a situation can becomputationally modeled and why? Does a situation include an individuallearner diagnosing a medical case using a computer tutor which is housedin his/her office at a hospital (e.g., Azevedo & Lajoie, 1998)? Does a situa-tion include an experienced nurse solving a complex trauma case using asimulation-based computer tutor housed in the chaotic confines of a surgicalintensive care unit (Lajoie, Azevedo & Fleiszer, 1998; Lajoie & Azevedo,2000)? Does it include a dynamic and evolving external representation of astudent’s (internal) mental model of the cardiovascular system? If so, shouldthe ILE also model the variables associated with the student’s ability to regu-late his/her own learning of the cardiovascular system (Azevedo et al., 2001;Azevedo, Guthrie, Seibert & Wang, in prep.)? Does a situation include adozen dyads in a high school science class using a Web-based simulationenvironment to solve ecology problems, where they have access to severalteachers and are surrounded by educational resources (Azevedo, Verona &Cromley, 2001). What role should the teachers play according to Akhras andSelf’s stance? Which aspects of the “situation” should the teachers model,and which aspects of the “situation” should the ILE model? How can the

37

ILE model such a complex situation with various levels of complexity? Butwait, we still don’t have a theoretically-motivated operational definition of asituation that could begin to address these questions.

In addition, I find it peculiar that the authors (this issue) chose a non-academic, impractical task such as salad-making to illustrate their construc-tivist stance. Are we reverting back to an earlier generation of cognitiveand ITS research where we computationally modeled toy-tasks – meaning-less tasks that are easy to model by a computer scientist, but irrelevant tothe real-world? I would have rather preferred to see that authors use theirconstructivist stance to design an ILE that would be relevant to people wholive outside the lab – teachers, students, tutors, trainers. For example, howabout an ILE that can tutor struggling adolescent readers, and therefore modelthe tutoring situation which includes the cognitive, motivational, affectivestates of the tutee, instructional scaffolding techniques used by the tutor, andco-joint knowledge construction activities based on the tutor-tutee interac-tions (Cromley, 2001)? This would pose a real challenge to the authors’ stanceand “bring them out of the lab” to collaborate synergistically with educatorsand psychologists to tackle the complex issue of modeling of situations(however defined) in addressing educational and professional concerns.

Akhras and Self (this issue) revert back to the traditional ITS approach byhighlighting the need to incorporate cognitive structures after earlier statingthat knowledge cannot be objectively defined. In addition, they also state thatan ILE should include a domain model in the form of situations. How is theAkhras and Self (this issue) approach to modeling situations different thanexisting ILEs which do not model cognitive structures or learning processes(e.g., Ericsson & Lehrer, 2000; Kozma et al., 2000). Furthermore, designapproaches to ILEs tend not to include a “modeling” component (Jonassen& Land, 2000) to their environments, mainly because “modeling” is seen asantithetical to the constructivist framework. These issues need to be clarifiedby Akhras and Self (this issue) in order for one to properly evaluate theirconstructivist philosophy of ILEs.

It seems that the authors have reverted to a traditional ITS approach bystating that domain knowledge structures will be part of situation modelsand that they will be modeled in terms of objects, relations, and other kindsof structures. It should therefore be noted that this approach has been usedextensively by other ITS researchers (including “modelers” and “middle-campers”) where they have modeled domain knowledge as part of a situationmodel (e.g., algebra problems in ACT-R tutors, medical cases in the SICUNtutor) embedded in computer-based learning environments.

In sum, Akhras and Self (this issue) present an initial conceptualizationbased on a constructivist-based approach to modeling situations in ILEs.

38

However, their ambitious proposal is typical of novel philosophical frame-works – conceptual clarity and construct validation remain elusive. Thisstands in contrast to the theory-driven and empirically-based approach takenby others such as the ACT-R group. Nevertheless, I do agree with Akhras andSelf (this issue) that perhaps ILEs should reason about interactions betweenthe content and the dynamics of learning situations. However, like the ACT-Rgroup, I use theory and empirical data to inform the design of my CBLEs.What is not clear from their article (this issue) is the “what, when, how, andwhy” related to this issue. My first suggestion is for the authors to clearlyand explicitly define their constructivist position and not to try to constructone that is based on an amalgamation of various “flavors” of construct-ivism (e.g., von Glasersfeld, Piaget, etc.). Conceptual clarity is required inorder for their approach to be adequately evaluated and used to inform thedesign of ILEs. As such, I invite the authors to re-conceptualize their frame-work by addressing how their constructivist stance can be used to addressa particular research agenda that focuses on the use of computers as meta-cognitive tools for enhancing learning. I briefly sketch the issues in the nextsection.

The use of computers as metacognitive tools to enhance learning:A theoretically-based and empirically-derived approach

As part of my commentary I would like to invite Akhras and Self (this issue)to re-conceptualize their framework by challenging them think about howILEs can be used as metacognitive tools to enhance learning. My colleaguesand I are currently grappling with issues similar to those raised by Akhrasand Self (this issue). More specifically, we are interested in the “what, when,how, and why” questions related to modeling a situation – i.e., the phasesand processes used by students to regulate their learning with hypermediaand web-based environments. More specifically, we are investigating the roleof self-regulation during learning of complex systems (e.g., circulatory andecological systems) with hypermedia and web-based environments (Azevedoet al., 2001; Azevedo et al., in prep.; Azevedo, Verona & Cromley, 2001).

Self-regulated learning has recently been viewed as an emerging issue ineducational and psychological research. There are several outstanding theore-tical and empirical issues related to learning and the use of adaptive hyper-media systems designed to foster self-regulated learning (SRL). The purposeof this section is to briefly outline a theoretically-based and empirically-driven research agenda which examines the role of self-regulation in students’learning with hypermedia and web-based environments. These environ-ments are designed to foster mental model progression of complex systems

39

(e.g., circulatory system) by detecting, tracing, modeling, and fosteringself-regulatory skills.

Self-regulated learners are generally characterized as active learners whoefficiently manage their own learning in many different ways (Winne, 1998;Winne & Perry, 2000; Schunk & Zimmerman, 1994). Self-regulated learningis an active constructive process whereby learners set goals for their learningand then attempt to monitor, regulate, and control their cognition, motiva-tion, and behavior (Pintrich, 2000). Models of self-regulation (e.g., Winne& Perry, 2000; Pintrich, 2000; Zimmerman, 2000) describe a recursive cycleof cognitive activities central to learning and knowledge construction activ-ities (e.g., using a hypermedia environment to learn about the circulatorysystem). Most of these models propose four phases of self-regulated learning(Pintrich, 2000). The first phase includes planning and goal setting, activationof perceptions and knowledge of the task and context, and the self in rela-tionship to the task. The second phase includes various monitoring processesthat represent metacognitive awareness of different aspects of the self, taskand context. Phase three involves efforts to control and regulate differentaspects of the self, task, and context. Lastly, phase four represents variouskinds of reactions and reflections on the self and the task and/or context. Ourresearch on learners’ SRL provides a critical, but yet unexplored issue relatedto learning with adaptive computer-based learning systems.

Foundations for research of self-regulation and hypermedia

My colleagues and I have recently begun to investigate the effects ofgoal-setting conditions (e.g., learner-generated versus experimenter-set), onlearners’ ability to self-regulate their learning with hypermedia (Azevedo etal., 2001; in prep.). So far, our research addresses three specific research ques-tions, including: (1) Do different goal-setting conditions influence students’ability to shift to a more sophisticated mental model of the circulatorysystem? (2) How do goal-setting conditions influence students’ regulation in ahypermedia environment? (3) What are the qualitative differences in students’self-regulatory learning in the three goal-setting conditions?

Methods. Our studies combine true experimental designs (where studentsare randomly assigned to several instructional conditions) with a think-aloud protocol methodology, where participants are asked to verbalize theirthinking processes as they learn about the circulatory system using a hyper-media environment. The use of a mixed methodological strategy allowsus to determine the effects of various instructional interventions on SRLand to examine the dynamic nature of SRL variables during learning withhypermedia.

40

Results. The results from out initial research-based study aimed at investi-gating the nature of self-regulated learning (SRL) with hypermedia focuson: (1) shifts in mental models (of the circulatory system) from pretest toposttest, (2) role of multiple representations during learning with hypermedia,(3) coding scheme developed to analyze learners’ self-regulatory behavior,(4) establishing a model of SRL with hypermedia, and (5) understand thedynamics of SRL variables during learning. We have found five clustersof SRL variables used by learners while using a hypermedia environmentto learn about the circulatory system including: (1) Planning (planning,sub-goaling, prior knowledge activation, and recycling a goal in workingmemory), (2) Monitoring (judgement of learning, feeling of knowing, self-questioning, content evaluation, and identifying the adequacy of informationavailable in the hypermedia environment), (3) Strategy use (selecting anew informational source, searching, summarization, copying information,re-reading, making inferences, hypothesizing, knowledge elaboration, andevaluating the content as the answer to a question), (4) Task difficulty anddemands (time and effort planning, help-seeking behavior, task difficulty,control of context, and expectation of adequacy of information), and (5)Interest statement (the learner has a certain level of interest in the task orin the content domain of the task).

This brief demonstration of how we as psychologists, working withdomain experts, have a theoretically-driven approach to extend existingtheories and methods of SRL to study SRL when students learn from hyper-media and web-based environments. Our results will be subsequently used toinform the design of adaptive hypermedia and web-based systems. This leadsto the role of computer scientists and AI research in the next phase of ourresearch program – how can we address the issues presented below whichhave been derived from our empirical data?

Implications of SRL research for the design of adaptive hypermediaenvironments: Issues and challenges

To re-conceptualize their constructivist framework, I invite Akhras and Selfto explain how their evolving constructivist stance would apply to ourresearch on self-regulation and learning and design of hypermedia and web-based environments aimed at detecting, modeling, monitoring, and fosteringlearners’ self-regulated learning. How would they model a situation involvinga student using a hypermedia environment to learning about complexsystems? Existing SRL models (for an extensive review refer to Boekaerts,Pintrich & Zeidner, 2000) and our research indicate that there are severalphases and processes that a student uses when regulating his/her learning

41

about complex systems. How can we model a situation which includes severalphases (e.g., planning, monitoring, controlling, and reflections) and processes(e.g., cognitive, affective, motivational) related to self-regulation? What aboutchanging the situation to a dyad learning collaboratively or a tutoring situa-tion? How does this change the nature of a situation? How can the emergingknowledge structures that are being co-jointly constructed be modeled aspart of the situation? How would the ILE model nonverbal (gesture and toneused by a tutor and student), internal mental representations of the individuallearner (mental model of the circulatory system), and shared emerging repre-sentations between student-student or teacher-student? How would the ILEdetect, trace, model, and monitor these components of the situation in orderto reason about the situation? What other components would be necessary foran ILE to adequately model the different levels of a situation (e.g., what arethe implications for the instructional and motivational planners)?

How would a constructivist-based ILE detect, trace, and monitor thecritical SRL variables used by high self-regulated learners, e.g., plan-ning, sub-goaling, prior knowledge activation, self-questioning, coordinationof multiple representations, re-reading, knowledge elaboration, intentionalcontrol of time on task, taking advantage of the tools embedded in thehypermedia environment to enhance learning of the instructional material,and motivational aspects related to the learner’s interest in the topic? WhichAI techniques could be used to detect, monitor, and model these variables?Akhras and Self’s framework would need account for how the ILE couldhandle the complexity involved in detecting, tracing, and monitoring thesevariables during learning.

What types and levels of scaffolding methods should be designed for lowself-regulating learners? According to our research results, these studentstypically do not plan their learning activities, fail to set instructional goals,fail to monitor their learning, use ineffective learning strategies, and mangetheir learning by engaging in lots of help-seeking behavior, since they havedifficulty judging task difficulty and fail to integrate new information withexisting prior knowledge. So, how do we “expand” the ILE’s components(e.g., student model, instructional model, interface, etc.) to determine if alearner is a low- or high self-regulator and what effects will this determina-tion have on the detection, monitoring, and fostering of learners’ overallself-regulation? How do we make our SRL model “visible” to the learnersand flexible enough to allow learners to explore advanced topics related tothe circulatory system, including its content and structure. How does theenvironment adapt and exhibit flexibility during learning?

What are the implications of our SRL model in designing the studentmodel, instructional planner, motivational planner, and other system compon-

42

ents which may be needed for the system to detect, trace, monitor, model,and foster self-regulated learning? For example, do we need to build an SRLpalette similar to a help system which allows learners to indicate that they donot know how to plan their learning of the cardiovascular system? In this case,should the ILE present a student with a planning net with display a sequenceof possible sub-goals that he/she should attempt?

What about if the student indicates low motivation (e.g., interest in thetask)? How can the ILE detect low self-motivation? Should it ask the studentexplicitly about his/her motivational state (Lepper et al., 1993; duBoulay etal., 1999) on a regular basis or should the student be aware that there isan on-line motivational palette (part of the SRL pallete), which he/she canaccess and use to modify his/her current motivational level during learning?And even if the ILE is successful in detecting the learner’s motivationallevel, then how should the instructional planner and student model react?Should the student be challenged? How do these decisions affect subsequentlearning (including learning “outside” the ILE)? None of these questions canbe addressed without a theory and evidence.

How can we design ways of detecting, monitoring, and fostering shiftsin learners’ mental models of the circulatory system? Can we have studentscreate concepts maps and/or drawings which can be used to dynamic-ally assess their existing mental model and which will interact with theother system components? For example, what kind of instructional decisionsshould be made in the case where the ILE has determined that a studenthas a sophisticated mental model of the circulatory system but has expressedlow interest in the task, versus a learner who has a less-sophisticated mentalmodel but has indicated high interest in the topic, yet lacks the ability to planhis/her learning and is using ineffective strategies (e.g., “blindly” searchingthe hypermedia environment without any goals)? Would making the learnerconstruct an “external” visual representation of his/her emerging mentalmodel of the domain allow him/her to self-regulate? Can this informationprovide the system with another “variable” with which to make informedinstructional decisions? Would this external representation, which is visible,and accessible, allow the user and others (e.g., peers, teachers) to share,inspect, critique, modify, and assess the learner’s understanding of a complexsystem? Again, Akhras and Self’s framework needs to include a model of theevolving internal knowledge structures and make them accessible to both theindividual learner and the other agents participating in the situation.

In sum, Akhras and Self (this issue) need to clearly conceptualize theirnovel constructivist stance before we can begin to appreciate the contri-bution of their ambitious proposal to model situations computationally inintelligent learning environments (ILEs). The authors raise several interesting

43

ideas. However, conceptual clarity is definitively required in order for theirapproach to be adequately evaluated and used to inform the design of ILEs.An invitation has been put forth, based on my research on SRL and learningwith hypermedia and web-based environments, to stimulate the authors in re-conceptualizing their framework. How can their constructivist stance be usedto address a research agenda that focuses on the use of computers as meta-cognitive tools to enhance learning? These kinds of academic exchanges arefruitful in stimulating collaboration between educators, learning scientists,computer scientists and AI researchers to solve shared problems.

Acknowledgments

I would like to thank Patricia Alexander for giving me the opportunity toreview the original manuscript and write this commentary. I would also liketo thank Fabio Akhras and John Self for the opportunity to comment on theinitial conceptualization of their ideas regarding the computational modelingof situations in intelligent learning environments (ILEs). Lastly, I would alsolike to thank Jennifer Cromley for comments on a previous version of thismanuscript.

References

Anderson, J.R. (1983). The Architecture of Cognition. Cambridge, MA: Harvard UniversityPress.

Anderson, J.R. (1993). Rules of the Mind. Hillsdale, NJ: Erlbaum.Anderson, J.R., Corbett, A.T., Koedinger, K.R. & Pelletier, R. (1995). Cognitive tutors:

Lessons learned. The Journal of the Learning Sciences 4(2): 167–207.Anderson, J.R. & Lebiere, C. (1998). The Atomic Components of Thought. Mahwah, NJ:

Erlbaum.Azevedo, R., Guthrie, J.T., Seibert, D. & Wang, H. (in prep.). The Role of Learner-generated

Goals in Regulating Learning from Hypermedia. Manuscript in preparation.Azevedo, R., Guthrie, J.T., Wang, H. & Mulhern, J. (2001). Do Different Instructional Inter-

ventions Facilitate Students’ Ability to Shift to more Sophisticated Mental Models ofComplex Systems? Paper to be presented at the Annual Conference of the AmericanEducational Research Association, Seattle, WA.

Azevedo, R. & Lajoie, S.P. (1998). The cognitive basis for the design of a mammographyinterpretation tutor. International Journal of Artificial Intelligence in Education 9(1/2):32–44.

Azevedo, R., Verona, M.E. & Cromley, J.G. (2001, May). Fostering Students’ CollaborativeProblem Solving with RiverWeb. Paper to be presented at the 10th International conferenceon Artificial intelligence in Education, San Antonio, TX.

Boekaerts, M., Pintrich, P, & Zeidner, M. (2000). Handbook of Self-regulation. San Diego,CA: Academic Press.

44

Chi, M.T.H. (2000). Self-explaining: The dual processes of generating inference and repairingmental models. In R. Glaser, ed., Advances in Instructional Psychology: EducationalDesign and Cognitive Science (vol. 5), pp. 161–238. Mawah, NJ: Erlbaum.

Cromley, J.G. (2001, May). Effective Human Tutoring in Reading: Precursor to the Design ofan ITS. Paper to be presented at the 10th International Conference in Artificial Intelligenceon Education, San Antonio, TX.

Derry, S.J. & Lajoie, S.P. (1993). A middle camp for (un)intelligent instructional computing:An introduction. In S.P. Lajoie & S.J. Derry, eds, Computers as Cognitive Tools, pp. 1–11.Hillsdale, NJ: Erlbaum.

Du Boulay, B., Luckin, R. & del Soldato, T. (1999). The plausibility problem: Human teachingtactics in the “Hands” of a machine. In S.P. Lajoie & M. Vivet, eds., Frontiers in ArtificialIntelligence and Applications. Open Learning Environments: New Computational Tech-nologies to Support Learning, Eploration and Collaboration, pp. 225–232. Amsterdam:IOS Press.

Erickson, J. & Lehrer, R. (2000). What’s in a link? Student conceptions of the rhetoric ofassociation in hypermedia composition. In S.P. Lajoie, ed., Computers as cognitive ToolsII: No more Walls: Theory Change, Paradigm Shifts and Their Influence on the Use ofComputers for Instructional Purposes, pp. 197–226. Mawah, NJ: Erlbaum.

Jacobson, M.J. & Kozma, R.B. (2000). Innovations in Science and Mathematics Education:Advanced Designs for Technologies of Learning. Mahwah, NJ: Erlbaum.

Jonassen, D. (2000). Computers as Mindtools for Schools: Engaging Critical Thinking (2nded.). Englewood Cliffs, N.J.: Merrill.

Jonassen, D. & Land, S.M. (Eds). (2000). Theoretical Foundations of Learning Environments.Mahwah, NJ: Erlbaum.

Koedinger, K.R. & Anderson, J.R. (1997). Intelligent tutoring goes to school. InternationalJournal of Artificial Intelligence in Education 8: 30–43.

Koedinger, K.R. & Anderson, J.R. (1998). Illustrating principled design: The early evolutionof a cognitive tutor for algebra symbolization. Interactive Learning Environments 5: 161–179.

Kozma, R., Chin, E., Russell, J. & Marx, N. (2000). The roles of representations and toolsin the chemistry laboratory and their implications for chemistry learning. Journal of theLearning Sciences 9(2): 105–144.

Lajoie, S.P. (1993). Computer environments as cognitive tools for enhancing learning. InS.P. Lajoie & S.J. Derry, eds, Computers as Cognitive Tools, pp. 261–288. Hillsdale, NJ:Erlbaum,.

Lajoie, S.P. (2000). Computers as Cognitive Tools II: No more Walls: Theory Change,Paradigm Shifts and Their Influence on the Use of Computers for Instructional Purposes.Mahwah, NJ: Erlbaum.

Lajoie, S.P. & Azevedo, R. (2000). Cognitive tools for medical informatics. In S.P. Lajoie,ed., Computers as Cognitive Tools II: No more Walls: Theory Change, Paradigm Shiftsand Their Influence on the Use of Computers for Instructional Purposes, pp. 247–271.Mahwah, NJ: Erlbaum.

Lajoie, S.P., Azevedo, R. & Fleiszer, D.M. (1998). Cognitive tools for assessment and learningin a high flow information environment. Journal of Educational Computing Research18(3): 203–233.

Lajoie, S.P. & Derry, S.J. (1993). Computers as Cognitive Tools. Hillsdale, NJ: Erlbaum.Lepper, M., Woolverton, M., Mumme, D. & Gurtner, J. (1993). Motivational techniques of

expert human tutors: Lessons for the design of computer-based tutors. In S. Lajoie & S.Derry, eds, Computers as Cognitive Tools, pp. 75–105. Hillsdale, NJ: Erlbaum.

45

Mayer, R.E. & Wittrock, M.C. (1996). Problem solving transfer. In D. Berliner & R. Calfee,eds, Handbook of Educational Psychology, pp. 45–61. New York: Macmillan.

Newell, A. (1989). Putting it all together. In D. Klahr & K. Kovosky, eds., ComplexInformation Processing: The Impact of Herbert A. Simon, pp. 399–440. Hillsdale, NJ:Erlbaum.

Newell, A. & Simon, H.A. (1972). Human Problem Solving. Englewood Cliffs, NJ: PrenticeHall.

Pintrich, P.R. (2000). The role of goal orientation in self-regulated learning. In M. Boekaerts,P. Pintrich & M. Zeidner, eds, Handbook of Self-regulation, pp. 451–502. San Diego, CA:Academic Press.

Schunk, D. & Zimmerman, B. (1994).Self-regulation of Learning and Performance: Issuesand Educational Applications. Hillsdale, NJ: Erlbaum.

Shute, V. & Psotka, J. (1996). Intelligent tutoring system: Past, present, and future. In D.Jonassen, ed., Handbook of Research for Educational Communications and Technology,pp. 570–600. New York: Macmillan.

Winne, P. (1998). Experimenting to bootstrap self-regulated learning. Journal of EducationalPsychology 89: 397–410.

Winne, P.H. & Perry, N.E. (2000). Measuring self-regulated learning. In M. Boekaerts, P.Pintrich & M. Zeidner, eds., Handbook of Self-regulation, pp. 531–566. San Diego, CA:Academic Press.

Zimmerman, B. (2000). Attaining self-regulation: A social-cognitive perspective. In M.Boekaerts, P. Pintrich & M. Zeidner, eds, Handbook of Self-regulation, pp. 13–35. SanDiego, CA: Academic Press.