Search
Program Calendar
Browse By Day
Browse By Time
Browse By Person
Browse By Room
Browse By Unit
Browse By Session Type
Search Tips
Annual Meeting Housing and Travel
Personal Schedule
Sign In
Pedagogical agents have been studied extensively for twenty years or more (e.g., Johnson & Lester, 2016), and meta-analyses have documented when and how they improve learning. They are well established as components of learning products. For example, Alelo’s learning products make extensive use of animated agents that engage learners in conversations, to help them develop proficiency in foreign languages and cultural awareness (Sagae, Wetzel, Valente & Johnson, 2009). So it might seem surprising that that there is anything new still to learn about pedagogical agents.
However agents are very well positioned to take advantage of emerging developments in digital technologies. This is opening up new possibilities for agents and raising new research questions. For example, speech and language technologies are now available in the cloud for use by software developers. This makes it much easier to create agents that engage in conversations with learners, and are not simply a media enhancement for static slideshow presentations. Such conversational agents are available to learners any time to answer questions or help learners practice critical skills.
As agent technology moves into the cloud, it becomes possible to collect and analyze data from learners, so that the agents themselves continually learn and improve. This can have a profound impact on pedagogical agent technologies. In fact it no longer makes sense to talk about “agent-learner interaction”, since agents can learn too and so can also be considered to be learners. The following are some examples of ways in which machine learning has changed the way we develop agents at Alelo. For our language learning products we used to script the behavior of the agents using authoring tools. Now we collect examples of the dialogue we want our agents to exhibit, and use it to train dialogue models that we incorporate into the agents. Then once we host these agents in our learning platform we continually collect more data as human learners interact with them. We use these data to retrain the dialogue models, as well as develop models of common human learner errors. As learners around the world interact with the system it identifies examples of errors that language learners make. This makes the agents increasingly able to recognize and provide feedback on these errors.
Agents thus serve as data collection tools as well as learning tools. Agents are developed iteratively, informed by what human learners actually do instead of what instructional designers imagine they might do. The ubiquity of data also makes these agents excellent tools for educational research. New immersive technologies such as augmented reality further accelerate this trend, by providing rich streams of interaction data that agents and researchers can learn from.
Artificial intelligence is having an increasing impact on society, making many routine jobs obsolete. But it can also be harnessed to promote human learning, to help people rapidly develop the skills that they need to succeed in this new economy. As agents learn and become more effective instructionally, they can help people to become better learners too.