Using Artificial Neural Networks to Model Siegler’s Balancing Task

Understanding balance is a necessary part of a child’s learning experience. Siegler found that children often go through four stages in understanding balance and torque. Computer simulated models of Siegler’s experiment have been created using Artificial Neural Networks. Shultz used a neural network to model Siegler’s balance task and found that the network moved through the four stages argued by Siegler. Whereas Shultz had a very formal definition of each stage, we believed that these definitions were arbitrary; instead, we focused on observing the neural network’s incremental learning by avoiding the explicit definitions. Five experiments were performed with three categories of problem type (easy, medium, and hard). Each of these problems represent a subset of the balance scale problems used in Siegler’s experiment. We found we were able to replicate the stage-like development and model the whole developmental process of learning the balance task.