Pedagogical Agents on the Web

    Lewis Johnson, Erin Shaw, and Rajaram Ganeshan
    Center for Advanced Research in Technology for Education
    USC / Information Sciences Institute
    4676 Admiralty Way, Marina del Rey, CA 90292-6695 USA

    {johnson, shaw, rajaram}


Animated pedagogical agents are lifelike animated characters that facilitate the learning process. This paper describes Adele, a pedagogical agent that is designed to work with Web-based educational simulations. The Adele architecture implements key pedagogical functions: presentation, student monitoring and feedback, probing questions, hints, and explanations. These capabilities are coupled with an animated persona that supports continuous multimodal interaction with a student. The architecture supports client-side execution in a Web browser environment, and is able to interoperate with simulations created by off-the-shelf authoring tools.


Animated pedagogical agents are animated characters that facilitate learning in computer-based learning environments. These agents have animated personas that respond to user actions. In addition, they have enough understanding of the learning context and subject matter that they are able to perform useful roles in learning scenarios.

Although pedagogical agents build upon previous research on intelligent tutoring systems (Wenger 1987), they bring a fresh perspective to the problem of facilitating on-line learning, and address issues that previous intelligent tutoring work has largely ignored. Because pedagogical agents are autonomous agents, they inherit many of the same concerns that autonomous agents in general must address. (Johnson and Hayes-Roth 1998) argue that practical autonomous agents must in general manage complexity. They must exhibit robust behavior in rich, unpredictable environments; they must coordinate their behavior with that of other agents, and must manage their own behavior in a coherent fashion, arbitrating between alternative actions and responding to a multitude of environmental stimuli. In the case of pedagogical agents, their environment includes both the students and the learning environment in which the agents are situated. Student behavior is by nature unpredictable, since students may exhibit a variety of aptitudes, levels of proficiency, and learning styles.

We strive where possible to create agents that have life-like personas, and which are able to interact with students on an ongoing basis. This contrasts with other pedagogical agent work (e.g., [Ritter 1997]) that ignores personas. Animated personas can cause learners to feel that on-line educational material is less difficult (André et al 1998). They can increase student motivation and attention (Lester et al 1997). But most fundamentally, animated pedagogical agents make it possible to more accurately model the kinds of dialogs and interactions that occur during apprenticeship learning and one-on-one tutoring. Factors such as gaze, eye contact, body language, and emotional expression can be modeled and exploited for instructional purposes.

This paper focuses on a particular pedagogical agent developed at USC: Adele (Agent for Distance Education Ė Light Edition). Adele shares many capabilities with our other pedagogical agent, Steve (Johnson et al 1998, Johnson and Rickel 1998, Rickel and Johnson 1998, and Rickel and Johnson 1997). But whereas Steve was originally designed to operate in immersive virtual environments, Adele is designed to operate over the Web. This paper describes Adeleís capabilities and discusses issues relating to hosting such agents in a Web-based learning environment.

Using Adele

Adele is designed to support students working through problem-solving exercises that are integrated into instructional materials delivered over the World Wide Web. In a case-based clinical diagnosis application, students are presented with materials on a particular medical condition and are then given a series of cases that they are expected to work through. Depending upon how Adele is used instructionally, she can highlight interesting aspects of the case, monitor and give feedback as the student works through a case, provide hints or rationales for particular actions, or quiz the student to make sure the student understands the principles behind the case.

Figure 1. Adele explains the importance of palpating the patientís abdomen.

Figure 1 shows a typical case using Adele. The student is presented with a simulated patient in a clinical office visit setting. The student physician is able to perform a variety of actions on the simulated patient, such as asking questions about medical history, performing a physical examination, ordering diagnostic tests, and making diagnoses and referrals. Adele monitors the actions that the student takes and the state of the patient, and provides feedback accordingly.


Adeleís system consists of two main components: the pedagogical agent and the simulation. The pedagogical agent consists further of two sub-components, the reasoning engine and the animated persona. An optional third component, the session manager, is employed when the system is run is multiple-user mode.

The reasoning engine performs all monitoring and decision making. Its decisions are based on a student model, a case task plan, and an initial state, which are downloaded from the server when a case is chosen, and the agentís current mental state, which is updated as a student works through a case. Upon completion, the studentís actions are saved to the server where they will help determine the studentís level of expertise, as well as how Adele will interact with the student in future cases.

The animated persona is simply a Java applet that can be used alone or incorporated into a larger application. Animation frames can be swapped in and out to provide the case author or user with a choice of persona.

The simulation can be authored using the language or authoring tool of oneís choice. For example, the simulation for the clinical diagnosis application was built in Java while that for a critical care application was built using Emultekís RAPID, a rapid-prototyping tool whose simulations can run in a Web browser via a plug-in. All simulations communicate with the agent via a common API that supports event (e.g., the student orders a lab) and state change (e.g., the lab results get updated) notifications as defined by the simulation logic.

The integrated system is downloaded to and run on the clientís side for execution efficiency. This is in contrast to the architecture of most other Web-based Intelligent Tutoring Systems where the intelligent tutor sits on the server side, resulting in increased latency in tutor response to student actions (eg.,[Brusilovsky et. al 1997]). Reducing latency is especially critical when the overhead of animating an agentís response is considered.

Figure 2. Adele instructs a student to answer a quiz after the student elects a urine dipstick test.

Implementation Issues

To facilitate Web-based delivery, Adele is implemented in Java, making it possible to download Adele-enhanced course modules over the Web. This approach offers long-term advantages, although in the near term, incompatibilities between Java virtual machines make portability somewhat difficult. High quality text-to-speech synthesis is platform-dependent, so variants of Adele are provided to take advantage of the text-to-speech synthesis capabilities available on each platform. Adeleís animations are produced from two-dimensional drawings, instead of three-dimensional human figures. This makes it possible to run Adele on a variety of desktops, without relying upon specialized 3D graphics capability. The main drawback of 2D imagery is that it is difficult to compose behaviors, e.g., frown while looking to one side. We are experimenting with VRML browsers as a way of providing articulated human figures on a desktop setting. However, since adding a VRML browser adds complexity to the software installation, there will still be applications where 2D animations are preferable.

Another implementation issue that influenced the design was the need to interface to externally authored simulations. Simulation authoring tools such as VIVIDS and Emultekís RAPID are frequently used both to author simulation behavior and the simulation interface at the same time. When moving to the desktop environment with Adele, it was necessary to make Adele integrate with existing instructional authoring tools, without having to rewrite the interfaces to those tools. For this reason, Adele was designed to run in a separate window, communicating with existing educational simulations using an interprocess communication link. We then developed behaviors for Adele to give the impression that she is integrated with the other displays running on the desktop, even though she runs in a separate window. When the student clicks on a button in the simulation window, Adele turns her head to look toward where the student clicked. She has a pointer that she can use to point toward objects in the other windows, similar to the pointer used by Andréís PPP persona (André et al. 1998). These behaviors partially compensate for the inability to actually manipulate objects directly, as Steve does.

In spite of these design and interface changes, the functionality of Adele is quite similar to that of Steve. This is particularly remarkable given that Steve is designed to support "training", and Adele is designed to support "education." In reality, the type of learning being fostered in each case, skill acquisition, is much the same in both cases. For each medical case, Adele is provided with a plan that describes the actions to be performed in carrying out the case, the relationships between the actions, and the rationales for performing the actions. This allows Adele to give hints about what to do and to explain why, as appropriate, just as Steve does.

Adele has been extended to support some additional persona features and instructional capabilities that Steve presently lacks. On the persona side, Adele has a repertoire of facial expressions and body postures that represent emotions, such as surprise and disappointment. These permit Adele to respond in a more lifelike fashion to student actions. On the instructional side, we have developed ways of building more instructional guidance into Adeleís interactions with the student. Based on a studentís action, Adele may choose to intervene, offering advice about what to do instead, e.g., "before you order a chest X-ray you should examine the condition of the lesion." Alternatively, based on the context or action history, Adele can ask the student a Ďpop quizí question that must be answered before proceeding to make sure that the student understands the implications of the facts that he has gathered about the patient so far.

Current Status

Adele has been used so far to create two course modules: one for clinical diagnosis problem solving (illustrated in Figures 1 and 2) and one for emergency room trauma care training. Each has been iteratively reviewed by physicians for accuracy, and subjected to usability testing. We are currently adding finishing touches to the system, such as improving the look of the graphical interface, and adding support for multiple skill levels, so that Adele can be used both by medical students and by practicing physicians. Once the cases have been tested with different student populations, we will proceed to create additional cases and integrate them into complete on-line course modules.

Meanwhile, we have initiated a new project in collaboration with the USCís School of Gerontology to create additional problem-solving exercises for other life science courses, all centered on health care for aging populations. The first of these courses will be concerned with clinical problem solving in geriatric dentistry.

Lessons Learned

Although the Adele project is still work in progress, there are some preliminary lessons and conclusions to be drawn. These are summarized below.

Feasibility of Client-Side Tutoring

The World Wide Web has attracted attention in the intelligent tutoring system community as a vehicle for delivering intelligent tutoring to a wide audience. The first WWW-enabled tutors placed the intelligent tutoring on the server side, and used the Web as a remote user interface. It is difficult to attain a high degree of responsiveness in such systems, since student actions must be routed to the central server, and since the server becomes a potential processing bottleneck.

Adele demonstrates the feasibility of client-side tutoring. Because Adele runs on the client, she can respond immediately to student actions. At the same time, Web-based delivery allows Adele to rely upon a central server when it is appropriate to do so, for example to maintain a database of student progress and to provide synchronization for collaborative exercises being carried out by multiple students on multiple computers.

Advantages of the Agent-Oriented Approach

Adeleís design was based on an autonomous agent paradigm instead of an intelligent tutoring system paradigm. The design was based heavily upon earlier work on Steve. In the case of Steve, the distinction with conventional tutoring systems is fairly clear. Each Steve agent is able to operate in a dynamic environment incorporating multiple students and multiple other Steve agents. Steve can manipulate objects in the virtual environment, in order to demonstrate how to perform tasks, or in order to collaborate with students. Steve can sense where the student is and what he or she is looking at, and can adapt instructionally. He can use gaze, body position, and gesture to engage the student in multimodal dialog.

Because Adele is confined to a conventional desktop GUI interface, she has fewer options for interacting with students. Nevertheless, the agent-oriented approach offers advantages, even with the more limited interface. Adeleís use of gaze and gesture, and her ability to react to student actions, makes her appear lifelike and aware of the user. Her use of emotional facial expressions can have a motivating influence on the student. Finally, the agent-based approach made it relatively straightforward to extend Adele from a single-user system to a multi-user collaborative system.

Amount of domain knowledge required

Adele and other pedagogical agents like her also have lessons to offer regarding autonomous agent design. Researchers in believable agents argue that the audienceís perception of an agentís competence is more important than the competence itself (Sengers 1998), and that competence is but one of many factors that agent authors should take into account (Reilly 1997). A key question is to what extent these claims are true for pedagogical agents. User feedback from Adele does indeed suggest that agent design must take the studentís perspective into account. Behaviors such as gaze shifting are essential in order to give students the impression that these agents are aware of them and understand them. Presentation details such as body posture, facial expression, and tone of voice have a big impact on studentsí impressions of these agents. However, students differ from the typical "audiences" for believable agents in that they can engage agents in instructional dialogs. Giving students the ability to probe an agentsí knowledge requires a certain depth of knowledge on the part of the agent. The main lessons to be learned are: 1) Pedagogical agents need enough domain knowledge to support the anticipated instructional dialogs; 2) An agentís behavior and appearance enhance the perception of expertise in the agent; 3) Users can react to agents in unexpected ways, so prototyping and experimentation are essential.

Representation of domain knowledge

It is common knowledge in the intelligent tutoring systems community that knowledge representations that support problem solving do not necessarily support tutoring (Clancey 1983). Previous work by the author (Johnson, 1994) showed that machine learning techniques can be used to reorganize an agentís expert knowledge to support explanation. In Adele we abandoned the method of building a capable agent first and then reorganizing its knowledge to support pedagogy. At the same time, building representations directly to support pedagogy was problematic: experts in AI might be able to build them, but typical course authors would probably be unable to. Our focus has been to build representations that support an agentís activities, and where possible support authoring as well. However, knowledge transformation steps (either manual or automatic) appear to be necessary in order to create a rich enough representation for agents to use. Adele utilizes a hierarchical plan representation, including preconditions and effects. Domain experts will enter knowledge via a form-based interface, which agent developers will then code into case plans.

Type of persona required

Adele has an effective persona, which viewers readily accept; we have found no clear advantage of 3D over 2D as far as user acceptance is concerned. However, finding the right level of realism has been difficult. Text-to-speech synthesis quality has been a critical factor for Adele, and we continue to experiment with different speech synthesis systems in order to arrive at an acceptable solution.

Are agents cost-effective?

The Adele work demonstrates that agents can be deployed on low-cost platforms. The Adele engine has been integrated into multiple educational applications. Our generalized approach to authoring makes it relatively easy to create large numbers of cases, making the overall approach cost-effective. The Adele persona responds to a range of high-level gesture commands, making it straightforward to integrate persona gestures into course materials. In addition, new animations can be easily added to Adeleís repertory of gestures.


CARTE students and researchers Andrew Marshall and Jeff Rickel contributed to the work presented here. Our collaborating organizations provided indispensable assistance, Drs. Demetrio Demetriatis, William La, Sidney Ontai, and Beverly Wood at the USC School of Medicine, and Carol Horwitz and Craig Hall at Air Force Research Laboratory. This work was supported by an internal research and development grant from the USC Information Sciences Institute.


André, E., Rist, T., and Müller, J. (1998). Integrating reactive and scripted behaviors in a life-like presentation agents. In K.P. Sycara and M. Wooldridge (Eds.), Proc. of the Second International Conference on Autonomous Agents, pp. 261-268.

Brusilovsky, P., Schwartz, E., and Weber, G. (1996). ELM-ART: An intelligent tutoring system on world wide web. Frasson, C., Gauthier, G. and Lesgold, A. (Eds.), Proc. of the Third Intíl Conference on Intelligent Tutoring Systems, pp. 261-269. Springer Verlag.

Clancey, W. (1983). The epistemology of a rule-based expert system: A framework for explanation. Artificial Intelligence 20(3), pp. 215-251.

Johnson, W.L., 1994. Agents that learn to explain themselves. Proceedings of the Twelfth National Conference on Artificial Intelligence, pp. 1257-1263. AAAI Press, Menlo Park, CA.

Johnson, W.L. and Hayes-Roth, B. (1998). The First Autonomous Agents Conference. The Knowledge Engineering Review 13(2), pp. 1-6.

Johnson, W.L. and Rickel, J. (1998). Steve: An animated pedagogical agent for procedural training in virtual environments. SIGART Bulletin 8, pp. 16-21.

Johnson, W.L., Rickel, J., Stiles, R., and Munro, A. (1998). Integrating pedagogical agents into virtual environments. Presence 7(5), forthcoming.

Lester, J.C., Converse, S.A., Kahler, S.E., Barlow, S.T., Stone, B.A., and Bhogal, R.S. (1997). The persona effect: Affective impact of animated pedagogical agents. In Proceedings of CHI í97, pp. 359-366, ACM Press.

Reilly, W.S.N. (1997). A metholodogy for building believable social agents. In W.L. Johnson and B. Hayes-Roth (Eds.), Proc. of the First Intíl Conference on Autonomous Agents, pp. 114-121, ACM Press.

Rickel, J. and Johnson, W.L. (1998). Animated agents for procedural training in virtual reality: perception, cognition, and motor control. Accepted for publication in Applied Artificial Intelligence Journal.

Rickel, J. and Johnson, W.L. (1997a). Intelligent Tutoring in Virtual Reality: A Preliminary Report, Proc. of the Intl. Conf. on Artificial Intelligence in Education, pp. 294-301.

Ritter, S. (1997). Communication, cooperation, and competition among multiple tutor agents. B. du Boulay and R. Mizoguchi (Eds.), Artificial Intelligence in Education, 31-38.

Sengers, P. (1998). Do the Thing Right: An architecture for action-expression. In K.P. Sycara and M. Wooldridge (Eds.), Proc. Of the Second Intíl Conference on Autonomous Agents, pp. 24-31, ACM Press.

Wenger, E. (1987.) Artificial intelligence and tutoring systems: Computational and cognitive approaches to the communication of knowledge. Los Altos, CA: Morgan Kaufmann Publishers, Inc.