and language, pedagogical agents integrate verbal and
nonverbal communication in rich learning interactions. Dialogue with pedagogical agents takes place in
the context of a learning activity, and agents can
exploit this context to evaluate the learner’s responses and provide feedback. They can assess the learner’s
progress toward mastery of the target skills, and share
those assessments with teachers as well as learners.
This assistance helps teachers in blended learning
programs to focus on areas where learners are experiencing difficulties and relieves them of the burden of
having to evaluate learner responses themselves.
Not all learning applications employing natural
language dialogue will feature animated pedagogical
agents; some applications on mobile devices will likely rely on text messaging and social media, just as
their human users do. However, animated interfaces
will continue to be essential for many types of agents,
including role-playing agents and relational agents.
The Future of Relational Agents
Recent new developments are relational agents,
which are virtual agents that engage in relationship-building behaviors with users. Relational agents are
used extensively in healthcare applications to develop rapport with patients, and there have been initial
experiments with pedagogical agents that use rela-tionship-building behaviors (Bickmore, Pfeifer, and
Schulman 2011). We predict that future agents will
combine affective computing technologies with rela-tionship-building behaviors, resulting in empathetic
pedagogical agents that develop and maintain relationships with learners (Walker and Ogan 2016).
Such agents will not only express emotions and react
to learner emotional states, but also infer psychological characteristics such as personality (Robison,
McQuiggan, and Lester 2010), demonstrate empathy,
and be emotionally supportive. Such qualities are
essential for good teachers but tend to be neglected
in educational software.
Relational agents could be useful in promoting
growth mindsets and teaching grit. They could be
particularly useful as a companion for lifelong learners. Lifelong learners are likely to engage in a diverse
set of learning activities over the course of their
careers. Conventional domain-specific learner models may be useful for pedagogical agents in the short
term, but they will be of limited value over time as
learners move between learning experiences. Relational pedagogical agents that develop models of each
learner’s character traits and establish relationships
with them could be more effective in supporting
learners. Such emotional support is currently lacking
in personal assistants for learning such as PERLS
(Freed et al. 2014) that have models of learning paths
but not of the emotional dynamics, the frustrations
and joys, associated with learning journeys.
The Future of Colocated Agents
Our 2000 paper conceived of pedagogical agents as
“cohabiting” learning environments with learners.
We saw virtual environments as making it possible
for learners and agents to share the same space and
interact with each other — even if virtual-reality
technology was not yet ready for widespread use in
educational contexts. Recent advances in virtual reality, augmented reality, and robotics now make it
much more feasible to colocate agents and learners in
the same space, resulting in more engaging and effective learning experiences.
Alelo has experimented with different ways of
colocating learners and agents, using virtual reality,
augmented reality, and robotic agents. In general, we
find such immersive environments to be highly
engaging for learners. They have more control over
where to go and where to focus their attention, compared to desktop displays. Learners can enter complex environments, containing multiple agents, that
challenge their skills. For example, we can seat the
learner at a banquet table with multiple guests (figure
7) to test the learner’s cross-cultural communication
skills. The learner must decide from moment to
moment which agent to talk to and then observe
how the other agents react. Learners have the freedom to make cultural mistakes (for example, addressing an interpreter or lower-ranked guest instead of
the more senior guest). Low-cost head-mounted displays such as Google Cardboard and mobile VR technologies such as WebVR make it possible to distribute
VR-based agent applications widely. We predict widespread use of such applications in the near future.
Robotic agents with expressive faces (figure 8) are
highly engaging for face-to-face communication,
much more so than screen-based agents. People react
in a very visceral way when a robotic agent responds
to what they say. However, virtual agent technology
is more versatile than robotic technology, and
humanoid robots are prone to mechanical failures.
Therefore, for many applications we expect virtual
agents to be used instead of or in addition to robotic
agents. Students who tested the RALL-E language
learning robot shown in figure 8 very much liked the
experience of interacting with the robot, but still
wanted a version that they could run on their smart-phone and take home with them.
Agents in Digital Ecosystems
Emerging interoperability standards make it possible
to integrate agent-enhanced learning environments
into digital ecosystems. 3 This trend is likely to accelerate the adoption and innovation of pedagogical
In the digital ecosystem approach, agent-enhanced learning activities such as role-play simulations and games exist alongside other digital learning materials to provide learners a seamless learning
experience. Agent platforms such as Alelo Enskill can