AI Colloquium: Phil Pavlik

Applying Cognitive Models to Improve Education and Human Performance

Cognitive modeling is often used to drive exploration and
experimentation about how to improve human learning and task
performance. While there are many ways to apply cognitive models to
these goals, this talk will describe how researchers can apply formal
economic principles to cognitive models to determine optimal ways for
computers to interact with humans. Specifically, in this talk I describe
a research program in which we have used a quantitative model of memory
performance to predict the optimal way to schedule practice of memory
items during "flashcard" type practice of vocabulary items. This
algorithm works in a computerized "flashcard" system to track a human
learner with a constantly updated model of their expected memories for
the items being learned. Before each drill practice (a test of a card
with review for failure), the model is queried by the system to compute
the most efficient item to give the student next, given a goal of
maximal learning after some pre-specified interval. This work is further
explained and developed with examples from other tasks to establish the
generality of the method.