Transferring Primitive Elements of Skill Within and Between Tasks - eScholarship

Transferring Primitive Elements of Skill Within and Between Tasks Logan J. Gittelson (gittel@rpi.edu) Cognitive Science Department, 110 8 th Street Troy, NY 12180 USA Niels A. Taatgen (n.a.taatgen@rug.nl) Department of Artificial Intelligence, Nijenborgh 9 Groningen 9747 AG Netherlands Abstract We chose these two tasks because they both produce distinct behaviors depending on the developmental stage of the subject. Specifically, the way in which multiple dimensions are handled changes depending on which stage the subject is currently in. If task-specific skills are guided by task-general strategies we should see some of the general strategies being acquired in one task transfer to another. We explore this hypothesis by training on either the BBT and TTB task to see whether prior training on one seemingly unrelated task improves performance on the other. We can analyze both the transfer between tasks, but also the transfer between stages within a task. At present, our work is free of empirical data and based solely on the concepts and parameters of ACTransfer. Hence, what we present here are theory-based predictions free of parameter tuning to empirical data. The primitive elements of skill theory proposes a set of approximately 2000 primitive information processing elements (PRIMs) (Taatgen, 2013) that compose all cognitive acts by combining and recombining to produce learning and transfer. By this theory, learning is transfer and transfer results from learning as the primitive elements combine to form new elements based on task demands and these more complex elements are reused later in learning (thereby producing increase in skill) and repurposed by different tasks (thereby producing transfer). We illustrate PRIMs in this paper by producing two models of the Balance Beam Task (BBT) and of the Take the Best heuristic (TTB). Although BBT and TTB do not, on the surface, possess much in common, when run in a transfer paradigm (TTB-to-BBT or BBT-to-TTB) each model harvests PRIMs created by its predecessor, thereby demonstrating positive transfer. Keywords: Balance Beam; Take the Best; ACTransfer; primitive information processing elements (PRIMs) Balance Beam Task The balance beam task has been a focus of study for many years. This task was originally developed by Piaget (1958) and extensively studied by Siegler (1976, 1981), among many others. The task involves a subject (typically a child around age 11) being shown a balance beam with a certain number of equally massive weights on each side at certain specific distances from the fulcrum, with the beam being held from tipping in either direction. The subject is then asked which direction the balance beam will fall (or if it will stay balanced) upon release. After the subject submits their answer the beam is then allowed to tilt, showing the correct answer. The methods used by children performing this task were said by Inhelder and Piaget to reflect developmental stages, and not experience-based changes in strategies. However, our immediate interest in this task is not with its history in developmental psychology but as a demonstration of general transfer with the ACTransfer theory. The 4 stages are: Introduction For decades cognitive scientists have been studying the amazing phenomenon of learning. Many researchers have developed theories and models of skill acquisition, but none have fully captured the mechanism behind learning. This illusive concept not only covers the development and refinement of a single isolated skill, but also the influence of previous experience and its influence on future learning. Ever since Thorndike and Woodworth's (1901) formal introduction of the concept, transfer has been a topic touched on by numerous researchers. Some work in the field includes Singley (1989) and, most recently, Taatgen (2013). Besides transfer scientists have also been interested in peoples' capacity to learn and reason at different ages. Inhelder and Piaget (1958) proposed a multi-stage model which has been supported by the results of many decision making tasks (Siegler 1976, 1981). We decided to use these stages in modeling Goldstein and Gigerenzer's (1996) Take the Best heuristic in the form of a decision making task. Both Piaget's Balance Beam Task (BBT) and Gigerenzer's Take The Best heuristic (TTB) have been a focus of cognitive modeling (Van Rijn 2003, Nellen 2003). By using Taatgen's ACTransfer modeling framework (Taatgen, 2013) we can create models for both BBT and TTB and then analyze the transfer between them. Stage 1 Subject only considers a single dimension (in this case usually the number of weights on each side), the subject then makes a decision based on only that factor.