We projected that the following pedagogical agent
interaction capabilities would promote learning:
Expressing emotions. Agents can better engage and
motivate learners by expressing emotional reactions
through facial expressions and body movement.
Nonverbal feedback. Agents can give feedback nonverbally through shaking of the head, facial expressions,
etc. Agents can use these cues to provide feedback continually, supporting learners without interrupting
their activities. For example, when STEVE coached
trainees in power-plant operations, he would watch
and nod or shake his head as the learner performed
each step of the procedure.
Gaze and gesture as attentional guides. Similarly, agents
can use nonverbal cues such as gaze and gesture to
direct focus of attention. For example, in figure 1
(right) STEVE is pointing at a button that the learner
must press to start the power plant. Attentional guides
such as these can make it easier for learners to follow
complex procedures, and raise learners’ awareness of
events in the environment without interrupting
Conversational signals. Agents can use nonverbal signals such as backchannel head nods to regulate conversations with the learner, especially in the context of
spoken dialogue. Nonverbal signals like these allow
the agent to have a conversation with the learner
while the learner is performing other tasks.
Adaptive pedagogical interaction. Combining the above
capabilities, agents can respond adaptively to interruptions, conversational turn-taking, and learner
actions in the course of an instructional scenario. For
example, Herman the Bug provided visual problem-solving advice that was adaptively tailored to learners’
individual needs as they designed virtual plants.
The capabilities of pedagogical agents make possible new types of interactions between learners and
learning environments. Back in 2000, we identified
two examples: interactive demonstrations and navigation guidance.
Interactive demonstrations. Agents such as STEVE can
demonstrate tasks while at the same time explaining
what they are doing, and why, directing the learner’s
attention to important features in the environment,
and answering learner questions.
Navigational guidance. Agents can lead learners around
a complex virtual environment, such as a power plant,
and prevent them from getting lost.
We also foresaw that pedagogical agents could
assume new roles in learning environments besides
that of intelligent tutors. For example, they could act
as virtual teammates and collaborate with learners in
How Did the Future Pan Out?
The fundamental hypothesis motivating our proposed research agenda was that pedagogical agents
would improve learning. Early empirical work on
pedagogical agents found the persona effect, which is
that introducing pedagogical agents into learning
environments can have a significant positive effect
on learners’ perception of their learning experience
(Lester, Converse, Kahler, et al. 1997). However, the
effect of pedagogical agents on learning outcomes
had not been established. Educational psychologists
greeted pedagogical agents with mixed reactions.
Some were concerned that that they might prove to
be “seductive details” (Garner, Gillingham, and
White 1989) — interesting but irrelevant and distracting to learning. Others such as Clark (2001)
argued that they were expensive to produce and that
equivalent learning effects could be achieved
through less costly means.
Since the publication of the Johnson, Rickl, and
Lester (2000) article, a broad array of studies have formally investigated the effect of pedagogical agents on
learning. These investigations have involved many
Figure 1. Example Turn-of-the-Century Pedagogical Agents.