Page 232 - Socially Intelligent Agents Creating Relationships with Computers and Robots

Socially Intelligent Agents in Educational Games 215

2.1 SIAs to Support Game-Based Collaborative Learning

Effective collaborative interaction with peers has proven a successful and
uniquelypowerful learning method[14]. Students learningeffectively ingroups
encourage each other to ask questions, justify their opinions, and reflect upon
their knowledge. However, effective group interaction does not just magically
happen. It depends upon a number of factors, including the group composition,
the task at hand, and the roles that the group members play during the interac-
tion [14]. Some of these factors (such as the composition of the group), need to
be taken into account when creating the groups. Others can be enforced during
the interaction by a human or artificial agent that oversees the collaboration
process and detects when the conditions for effective collaboration are not met.
We are working on creating artificial agents that can provide this mediating
role within multi-player, multi-activity educational games designed to foster
learning through collaboration. As a test-bed for our research we are using
Avalanche, one of the EGEMS prototype games, in which four players work
together through a set of activities to deal with the problems caused by a series
of avalanches in a mountain ski town. Each of the Avalanche activities is de-
signed to foster understanding of a specific set of mathematical and geometrical
skills, including number factorisation as well as measurement and estimate of
area/volume.
Preliminary pilot studies have shown that the collaborative nature of the
game triggers a tremendous level of engagement in the students. However,
they also uncovered several problems. First, students seldom read the available
on line help and the canned instructions provided within each activity. Thus,
students often lose track of the game goals and of the means available to achieve
them. Second, often students succeed in the game by learning heuristics that
do not necessarily help them learn the target instructional knowledge. Third,
the game at times fails to trigger effective collaboration. For instance, students
that are not familiar with the other group members tend to be isolated during
the interaction, while highly competitive students sometime turn an activity
designed to foster collaboration into a competition.

3. A Comprehensive Computational Model of Effective
Collaborative Learning

The above examples show that Avalanche can greatly benefit from the ad-
dition of SIAs that help students find their way through the game, trigger con-
structive learning and reflection, and help mediate and structure the collabora-
tive interaction. To succeed in these tasks the agents need to have:
(i) explicit models of the game activities they are associated with, of the emo-
tional states that can influence learning from these activities and of effective
collaborative interaction;