Dynamics of individual learning

Abstract

This work focuses on understanding how individuals learn complex tasks. Most real-world tasks that we learn and need help with, are complex. Yet most of our knowledge about human learning is based on simple tasks. We think the main reason is the lack of appropriate tools of analysis, as learning curves in complex tasks must be studied from an individual-specific perspective due to large individual differences of methods. Therefore, as our first step, we developed a novel, uncertainty-based tool -- the SpotLight -- to model changes in uncertainty of individual performance and highlight the plateaus, dips, and leaps at different levels of complex task performance. We applied the SpotLight to investigate the changes of individuals' methods while learning the complex game of Space Fortress (SF). Our results indicate that, underneath a sea of individual differences of methods, individuals applied a common, simple heuristic to recurrently update methods. While our SpotLight analysis helped us identify similarities across individuals, it also revealed scopes of improving measures of complex task performance to prevent false negatives of training benefits. Application of heuristics to update methods, as observed for our SF players, indicates at boundedly rational behavior to deal with the computational complexity of finding appropriate methods for complex tasks. Therefore, in our final study, we investigated how an individual searches for better methods while learning the complex game of Ms. Pacman. Our results indicate an efficient approach by the individual to search for better methods, by developing and refining a set of elementary, heuristic-based methods. Finally, we tie together the four studies and discuss how we may progress towards a boundedly rational theory of learning complex tasks.