Nan Li, William W. Cohen, Kenneth R. Koedinger
Building an intelligent agent that simulates human learning of math and science could potentially benefit both education, by contributing to the understanding of human learning, and artificial intelligence, by advancing the goal of creating human-level intelligence. However, constructing such a learning agent currently requires significant manual encoding of prior domain knowledge; in addition to being a poor model of human acquisition of prior knowledge, manual knowledge-encoding is both time-consuming and error-prone. Recently, we proposed an efficient algorithm that automatically acquires domain-specific prior knowledge in the form of deep features. We integrate this deep feature learner into a machine-learning agent, SimStudent. To evaluate the generality of the proposed approach and the effect of integration on prior knowledge, we carried out a controlled simulation study in three domains, fraction addition, equation solving, and stoichiometry, using problems solved by human students. The results show that the integration reduces SimStudent’s dependence over domain-specific prior knowledge, while maintains SimStudent’s performance.
The final publication is available at Springer via https://doi.org/10.1007/978-3-642-30950-2_63.