Learning interactions with artificial intelligences: a fallibilist perspective

After many false starts, the potential advantages of artificial intelligence for education are starting to appear. It remains in question, though, whether current theoretical frameworks are adequate for understanding processes of learning with artificial intelligence. This paper re-examines a recent study of children telling stories in collaboration with a virtual conversational agent, in which it was found that children who played with the agent told stories with more linguistically advanced characteristics than the stories of children who played with a friend. Conventional explanations in terms of “scaffolding” or that portray the agent as a “tool” seem to have limited predictive potential. It is argued that a fallibilist philosophy offers the potential of new insights and testable hypotheses in relation to such learning interactions with virtual peers. Domain: Learning and Cognitive Science SIG: SIG7 Learning and Instruction with Computers SIG3 Cognitive change