7. Conceptual Approach
7.2. Towards an Integrated Conceptual Model
Three main aspects of discussion are categorized to be: the learning environment, the IPA, and supporting models. The three main categories are discussed below.
The IPA
In reference to the review on pedagogical agents in Chapter 3, the requirements discussed in Chapter 6, and the visited conceptual frameworks in Section 7.1 the IPA as center of focus requires different abilities that can be categorized to be: appearance and animation, observation and decision, conversation, and affection.
The IPA appears in a visual environment, and as result it should have a visual embodiment component. As reported earlier, the IPA can utilize the persona for increased learner performance (Lester et al., 1997; Wang et al., 2008; Chapter 4).
The appearance element of the IPA is relevant to embodiment, visualization, and animation as well for the needed interaction with the user. In a relevant aspect, emotion elicitation by the pedagogical agent requires animation abilities. The animation aspect of the pedagogical agent gives further abilities to a channel of visual communication as described before. For example, and in similarity to the actual teacher, the virtual teacher pedagogical agent uses its animation abilities to point to relevant learning task location or provide gestures to show agreement or correctness of the task.
The observation function of the IPA is important not only to enable the IPA to
“sense” learner actions but also to be able to add contextualization (Machet, Lowe, & Guetl, 2012) by sensing the environment and tailor the actions to suit the context. The CycleTalk agent by Kumar and Rose (2009) is an example of an agent that is aware of the environment events and which can tailor its conversation to context. The observation component of the pedagogical agent is also important for the intelligence part to enable informed decisions that are not in isolation of what the learner is doing nor what is happening in the environment. As discussed in Section 7.1.3, the intelligent agent component requires sensing the environment for providing best reasoning. While the sensing component provides input for the decision making ability of the pedagogical agent, acting on the environment is also needed. I.e. to convey decisions to the environment and to the user. Apart from observation, the decision ability of the pedagogical agent is central to an effective realization for different pedagogical relevant reasons. Chapter 4 discussed intelligent methods in relevance to several pedagogical aspects. Lessons from the literature review are taken in considering the decision component to support the pedagogical agent. The decision component of the pedagogical agent is tied to it its behavior and is a very important component that is strongly tied to its pedagogical intelligent support.
Conversation is central to the IPA from the perspective that the IPA is the central point of interaction between the learner and the learning environment. A multi-model communication is necessary for any learner. Without communication, learning does not occur. The conversation ability is what distinguishes IPA value addition (Kim & Baylor, 2006). Variation of the type of communication is desirable to suit the learner differences and style. The type of communication can be by textual dialogues, voice conversations, or even by a visual gesture which can also convey a message. It is found by Weusijana, Kumar, and Rose (2008) that hiding text chat during speech is preferred than using dual channels of communication. Therefore, the implementation of conversation abilities of pedagogical agents should consider to the redundancy principle while giving multiple disjoint multi-modal communication interface abilities. As the IPA is an artificial entity, the type and realization of the conversation ability is bound to technological advancements in this research area while it positively contributes to the believability of the pedagogical agent.
The affection component suggests a pedagogical agent that can better understand the emotional state of the learner and contribute to pedagogy through affective support. This contributes to learner attitude and motivation to increase the rate of task completion in the virtual learning environment. In the found dialogues with the learner. Therefore, the affection element of the pedagogical agent has an emotion capture methods and control tactics to convey suited emotion to the learner. This aspect is tied to technological advancements and has relevance to the affective computing research domain.
Combining the above mentioned categories in the pedagogical agent realization can form its roles and functions. Considering the IPA as a virtual human like embodiment, those functions span the different layers of the virtual human reference architecture by Funge, Tu, and Terzopoulos (1999). The elements of the visual and intelligent pedagogical agent, discussed above, require behavioral functions that are mainly controlled by the top layer which is the cognitive layer. While the pedagogical agent stresses pedagogical oriented behavior, intelligent pedagogical reasoning focus on this cognitive layer as well.
Pedagogical intelligent functions require such a high layer that consequently controls the agent. In relation to Steve for example, it has the pedagogical function layer on top of SOAR cognitive architecture (Johnson et al., 1998;
Figure 36). With the facilitation of a cognitive layer and the mentioned functions, the roles of the pedagogical agent are better realized. Those roles can include providing a tutorial to a learner in the immersive learning environment, showing tasks, or monitoring and interacting with the learner while performing the task so
as to guide, correct errors, and make assessments. Different roles of the IPA should be targeted in depicting a conceptual framework.
The IPA has the potential to play different roles. Not necessarily the IPA will only interact with one learner. But it can act and interact with different avatars who can be teachers as well. Referenced above is the need for learners in the immersive environment to find peers; the IPA can also act as a peer which can provide other functions such as directing to other learners and other pedagogical agents such as in the work by Ashoori et al. (2009). An example depicting a different function of the IPA is found in Sklar, Parsons, and Stone (2004) who reported that incorporating multiple interacting educational robots in course settings brings benefits to learning. In interactive pedagogical drama (Marsella, 2003), multiple pedagogical agents are considered who are actors in front of immersed learners or a class presenting a story for a pedagogical objective, with ability of learners to interact as well. With pedagogical agents’ interaction and conversational abilities with learner, and in context, the virtual world is considered richer in pedagogical offering in comparison to learning with videos.
The Learning Environment
The environment should not only be rich in learning resources but should also support the suitable pedagogical methods of learning with these resources. It should supply the needs of pedagogical agents to be able to deliver effective instruction and support the needs of intelligent functions reasoning. Integrating a virtual world with an intelligent agent platform should convey the environment state and relevant events to the agent platforms and allow the intelligent agent to affect it for a pedagogical objective. The following questions are important: What are the learning elements needed in the learning environment, what is comprehensive? And what models suit our target learning environment?
While, the virtual world as a learning environment provides interesting characteristics for education, further support is required. It is discussed in Chapter 3 that the immersive environment provides a good opportunity for visual and contextualized collaboration. Basic affordances of the virtual world are also discussed in Section 3.2. It is further discussed that virtual worlds are not originally designed for learning purposes. The existence of such gap motivates the need for other functions and the addition of intelligent pedagogical agents to support its learning offerings. Considering the intelligent pedagogical agent interaction in the virtual world, IPA awareness and support are crucial to provide its services in a contextualized way accordingly.
Learning scenarios differ in a virtual world from other learning environments in 3D visual interaction with learning objects that are not anymore in 2D.
Learning events occur and there is a great potential for interaction with other learners in the immersive environment. It is discussed in Chapter 2 that virtual worlds give the opportunity for learning by doing. Furthermore, a distinguishing
aspect of virtual worlds is the authentic learning experience and the active explorative nature. Learning by doing and the gained learning upon experience with the immersive environment shift the focus from traditional learning and thus requires its awareness in the pedagogical strategy the IPA will undertake in the immersive environment. Details of interaction with the pedagogical agent, the environment, and the learner in relation to its assessment and goal achievement are important given the relation to the new virtual world characteristics.
The virtual world as an environment of learning is considered for its potential abilities of visualization that is combined with collaborative support to the learner. The IPA is viewed to act as a central point of interaction between the learner and the learning environment to provide learning services to the immersed learners. The components of those immersive learning services are related to the learning objects, spaces and modules that form the immersive learning environment. The conceptual functions adopting innovations of learning in the immersive world mandates supporting learning services from the environment. However, virtual worlds are not necessarily designed for learning, and therefore have two components to add. A supporting component from the environment and the other is from the IPA discussed above. The supporting component from the environment attempts to convert it from a virtual environment, or a virtual world to an immersive learning environment; a learning virtual world. Both dedicate resources, functionalities, and supporting models to be put to serve of the learner individual needs. Another objective is to make the environment pedagogically intelligent. For example, it is to support the IPA to reason and take decisions that achieve its pedagogical goals. Common to those in relation to the environment are the use of a perception module that links the events of the world to intelligent agent platform and to dispose reasoned actions to the environment. Reasoned actions received from the intelligent platform are forwarded to the environment to materialize the actions.
The importance of learning objects with pedagogical design is profound.
Considering a pedagogical agent in relation to the learning object, it has two objectives: to provide guidance to the resource, and to provide pedagogical guidance to the learner about the resource itself. Hence, the IPA requires understanding and ability of interaction with the learning objects. The semantic description and the ability to control the behavior of the learning object by the agent both facilitate better pedagogical scenarios. Therefore, the learning object model is required in a way that can be linked with the learner mode, activity model, and pedagogical models. In other words, while the IPA design objectives significantly consider pedagogical and context-aware interactions with the learner, the pedagogical agent awareness of the environment and the considered learning objects are necessary. Intuitively, learning objects are designed for a pedagogical goal that is tied to IPA objectives with the learner as well. Gluing different smaller learning objectives from different learning resources becomes important to be form to a learning path, which has an ultimate objective with the learning
requirement. This gluing can be either performed by the intelligent pedagogical agent or by the learning environment. Nevertheless, its goal-oriented attainment is related to the learner ability and to the various pedagogical related included models. The goal directed behavior, with the visited goal models contribute to the goal directed behavior of the intelligent agent in relation to the different goals of the various resources available. The goal-directed implementation approach thus is important for completing the learning path with the learning in an autonomous manner, which will be discussed again in the following Chapters.
The learning environment hence should support the pervasiveness of various learning resources in a way that contributes to the goal attainment by the agent.
In contrary, in absence of learning objects that contribute to pedagogical goals in the environment, those goals are not achievable.
Furthermore, learning evaluation concepts are important for integration with the adopted approach of a pedagogical agent. Within the visited frameworks, and with the authentic learning nature of the 3D environment, the learning situation can generate errors according to learner mistakes or due to environment generated errors. The pedagogical approach for assessment and monitoring can be either an immediate feedback or in a form of delayed assessments upon finishing the activity. Those were experienced with the error model, the risk model, and the environment model in relation to the learning objects. In learning situations, it might be possible to provide the learner with an opportunity to make mistakes or to complete an experiment with errors. However, in risky situations such as in simulations or training for hazardous scenarios, providing an immediate feedback or blocking the attempted action might be preferred or even mandatory. Thus striking a balance between delayed against immediate feedback is sought. The IMA model (Wesiak, Al-Smadi, and Guetl, 2012) provides a formalized and integrated approach for assessment, with complex learning resources consideration with a learner-centered focus leading to learning efficiency and effectiveness that can also relate to learning in virtual worlds with pedagogical agents. The learning by doing approach that suits learning in virtual world also mandates integrating assessment with the pedagogical agent interaction and behavior with the learner. And with the increased interactive abilities of an IPA, immediate feedback and reporting become both feasible and desirable. It is viewed that the assessment has a strong relation to goal attainment. In intelligent agent systems, a goal directed behavior is necessary for autonomous actions. The goals, as will be discussed later, are subject to reasoning that decides which goals to pursue and which to drop. Determination of goal success or failure should definitely be subject to learner progress in the learning situation and hence should be tied to the intelligent agent behavior.
Considering the cognitive layer of the pedagogical agent in the visited architectures, it can be either centralized focusing on one agent or adopting a decentralized intelligence approach. The decentralized approach is preferred as it glues intelligences from different sources that are multiple pedagogical agents,
learning objects, multiple learners, or other environmental aspects. That is in addition to reasoning with interaction with the discussed models. In relation to the learning object and the environment, the agents and artifacts model (Ricci, Viroli, and Omicini, 2007) as an example suggests the relevance of the learning object to the reasoning function. In relation to the learner aspect, Section 5.5 discussed the relevance of the intelligent agent model to collaborative learning support with abilities to provide cohesion for cooperative work. In relation to multiple pedagogical agents, Figure 50 of Section 8.3 shows a view of agents supporting multiple IPAs. The electronic institution concept (Bogdanovych et al., 2008), for example depicts a view of multiple agents rather than one. Definitely resembling the concept for a virtual university in a virtual word seeks coordination, awareness, and organization among the multiple pedagogical agents inhabiting the environment. The distributed model thus brings different aspects together to give potential of context-awareness and suit the continuous changes that occur in the dynamic virtual learning environment. Utilizing a distributed model of intelligence in a pedagogical agent thus permits extensibility in this regard allowing this cohesion achievement.
With the above mentioned factors in relation to the virtual world as a learning environment, and in order to provide intelligent pedagogical services, further architectural elements are needed. Those elements can form an immersive learning layer that adds to the known learning affordance provisioning of the virtual world transforming it to become pedagogically aware and intelligent in relation to the scope of IPA-based learning. As the virtual world by itself does not provide pedagogical guidance and follow-up, care is sought for executing and managing the pedagogical approach of learning delivery. The IPA along with the intelligent immersive learning layer, both support the delivery of pedagogical methods.
Supporting Models
Based on the discussion above, the objective is to have the pedagogical agent provide learning support to learners who are immersed in a virtual world that is rich in learning resources. The IPA, the VW, the learner community, and pedagogy form important model elements. Those elements are interrelated to achieve the effective and efficient learning. In the focus of implementing IPA in virtual worlds, the IPA becomes more responsible for instruction delivery and hence it has the main need for the models. But also, the IPA provides input as well. As the pedagogical agent provides different functions such as demonstrating, monitoring, and assessment, knowledge about relevant aspects are required to support its functions.
Models of consideration relevant to the IPA not only provide services to the learner but also require understanding from the environment, the learner, and the pedagogical oriented aspects. The IPA requires knowledge about the learner and to be able to also provide further knowledge since the IPA is the major relevant models. While the IPA is the central interaction point with the learner, it is responsible to know about the learner needs and provide the required information through its interface.
The intelligence component for both the cognitive layer of the IPA as well as supporting the intelligent immersive layer mandate several relevant models. For example, in visited research in a multi-agent based implementation of traditional learning environment, distribution of roles among agents is important for agents to act. Also, the goal directed and autonomous behavior of an agent show the need for goals assigned to agents. The role model and the goal model, both found in research, are important to clarify those aspects in the intelligent agent provided support.
The learning object understanding for its behavior is required for all elements, for the environment to integrate it with other components, for the IPA to be able to discover and explain to the learner. Supporting the learning activity mandates understanding about the learning object and the task through an object model and a task model.
It is central for the virtual learning environment to have a pedagogical model which facilitates supporting functions central to learning and learning activities. If there is no learner model used, the environment will definitely not be learner centered. Therefore, the learner model is needed given its potential benefits to
understanding the learner and the consequent adaptation and personalization. A special use case of the learner model is in open learner models (Mabott & Bull, 2004; Kerly, Hall, & Bull, 2006) that demonstrate the special importance of opening the model to learners, for them to be able to know and track their progress and status. It is the view that the learner model is essential for the IPA
understanding the learner and the consequent adaptation and personalization. A special use case of the learner model is in open learner models (Mabott & Bull, 2004; Kerly, Hall, & Bull, 2006) that demonstrate the special importance of opening the model to learners, for them to be able to know and track their progress and status. It is the view that the learner model is essential for the IPA