On the first day of the Intelligent Tutoring Systems Conference, there were several workshops that could be attended. As a full time e-learning author, I decided to enter the workshop that concerned intelligent support for collaborative learning. After all, there is a strong emphasis on group learning in the e-learning domain nowadays. At TinQwise, we try to understand the process of collaborative learning by using groupware platforms and various strategies to connect learners to one-another.
From an academic point of view, collaborative learning has informed instructional design and strategies that leverage the didactical benefits of learning in groups. Now, service providers are deploying web-based platforms that allow learners to access relevant materials and share their learning experience with their peers. Educational games accessible on mobile and tablet devices enrich learning interactions by bringing learners together. Which insights were given on process of collaborative learning?
At the first full paper session that I attended, Gregory Dyke from the Carnegie Mellon University at Pittsburg, elaborated on the subject of adaptable conversational agents for online collaborative learning. These agents are used for scaffolding online discussions through an approach called academic productive talk, or APT. This approach allows the learner to follow their own lines of reasoning, which leads to learners explaining, stating agreement and disagreement, reading and reexplaining statements of other learners. This contrast the ‘traditional way’ in which agents work, usually they lead the learner through directed lines of reasoning.
The downside of this can be formulated as follows; the learner is never questioned again and the learner will not receive additional insights from peers. Traditional learning agents are predictable, generic and likely boring to use. The question that was raised, how can creativity, that humans posses, be used in knowledge construction dialogues? Dyke suggested to to add socials behaviour to computer agents. However, how do we get different types of behavior? According to Dyke, this can be done by evaluating academic productive talk. By studying relevant talk that is produced in class, social and task guiding agents can be constructed. In short, agents that have a ‘revoiceÂ behaviour’, which means the agent sums, and revoices the learners’ input, contribute to the learning gains of the learner. Agents that merely provide generic feedback do not contribute to the learning gains of the learner.Â