| Reverse engineering is the process of discovering and reconstructing the working mechanism of a given functional system. This thesis aims to understand its underlying psychological mechanisms, and investigated how individuals reverse engineer Boolean systems. It proposes a theory of how individuals reverse engineer Boolean systems: they should do so either by focusing on each instance in which the system generates a positive outcome one at a time, or by focusing on the effects of each input component on the outcome one at a time. This account predicts that three factors should determine the difficulty of reverse engineering Boolean systems: the number of input components in the system, the number of instances in which there is a positive outcome (henceforth, number of positive instances), and the degree to which the input components depend on each other in determining the outcome of the system (henceforth, dependence). Experiment 1 tested the theory's predictions of difficulty with three two-switch Boolean electric circuit problems, and found corroboratory results both in accuracy and latency. Experiment 2 replicated the results of Experiment 1 and generalized them to another domain, and the participants were indeed likely to use the strategies predicted by the theory, as shown by their think-aloud protocols. Experiment 3 investigated the effects of the number of positive instances and dependence orthogonally with a more complex set of problems, and found that only dependence mattered in determining a problem's difficulty in those problems; the explanation is that participants were able to decompose an independent problem and build it up part by part to account for the positive instances regardless of the number of positive instances, but they were not able to do so for a dependent problem. Experiment 4 investigated whether the switches in the system had to be congruent, i.e., lined up in the same orientation, affected the difficulty in the three problems in Experiments 1 and 2. Results showed that although incongruent problems were indeed harder to solve, the predicted trend of difficulty across the three problems appeared in both congruent and incongruent problems. Applications of the experimental findings were discussed. |
