ABSTRACT
Suppose a student S was asked to solve the equation “2x + 3 = 9”. After 2 seconds, he gave the answer “x = 3”. Both cognitive scientists A and B were interested in understanding how S did it. Scientist A recorded S’s detailed verbal protocol, based on which, and other relevant behavioral measures, A hypothesized the possible knowledge structure underlying S’s problem solving and developed a symbolic computational model that simulated the process. On the other hand, scientist B adopted sophisticated brain imaging techniques including electroencephalograph (EEG) and functional Magnetic Resonance Imaging (fMRI) and acquired a high-resolution recording of S’s brain dynamics during problem solving. Based on some well-supported neural computing principles, B then developed a biologically realistic connectionist model to simulate the brain activities underlying S’s performance. Though both models fitted the data well, the two models are clearly different. While the symbolic model offers a description of the process with psychological plausibility and high behavioral relevance, the connectionist model emphasizes the process’ biological realism and brain foundations. One question is, do we, cognitive scientists who endeavor to discover unified theories of cognition, have justifiable reasons to prefer one to another?
