Researchers have made substantial progress towards developing Artificial Intelligence (AI) systems capable of a human-level performance at a narrow range of tasks. However, there do not yet exist machines capable of the kinds of flexible and efficient learning that humans exhibit. To overcome this gap, I present research exploring how data from intelligent tutoring systems can be leveraged to reverse engineer human capabilities and build more human-like models of learning. Towards this goal, I introduce my Apprentice Learner Architecture—a computational framework for generating and testing alternative models of learning—and demonstrate its use for modeling how K-12 students learn interactively from examples of correct behavior (inverse reinforcement learning) and feedback on problem solving (reinforcement learning) within intelligent tutoring systems.
I showcase two novel applications of these models for supporting the design of effective educational technologies. First, instructional designers can use these human-like models as learner "crash test dummies" to simulate students interacting with a tutor. I show evidence that my models can correctly predict which of two fractions tutor designs will yield better human performance. Second, I explore the use of these learning models for supporting non-programmers in authoring tutoring systems. Like humans, apprentice learner models can be taught by domain experts through worked examples and feedback. I present results showing that the time needed to author an Algebra tutor by interactively training an apprentice learner model is less than half the time needed to author a tutor using another state-of-the-art authoring-by-demonstration approach. These promising findings suggest that apprentice learning models can jointly support the goals of improving our understanding of human learning as well as building effective educational technologies at scale.
Chris MacLellan (https://chrismaclellan.com) recently received his PhD from the Human-Computer Interaction Institute at Carnegie Mellon University, where he explored the development of computational models of how people learn from the examples and feedback provided by educational technologies, such as intelligent tutoring systems and educational games. His work explores how these models can support the development of effective learning technologies—at scale—and how data collected from these educational technology can in turn be leveraged to drive development of better cognitive models and learning theories. Currently, Chris is a research scientist at Soar Technology, Inc., where he is exploring the broader application of these learning models both within and outside of educational contexts. His most recent work investigates how best to design interactive learning systems that are natural and efficient for people to teach. The products of his work have immediate implications for supporting educational technology development (e.g., teachers can build tutors through teaching rather than programming) as well as many broader implications (e.g., personal assistant technologies—such as Alexa or Siri—that everyday users can augment through natural and efficient teaching interactions).