CREATIVITY AS A RANDOM WALK SEARCH ON A SEMANTIC NETWORK

This thesis studies creativity in the context of network science. It investigates whether a random walk is a good model for the problem-solving process in humans during the Remote Associates Test, a multiply-constrained semantic creativity task. The work introduces two search models, which consist of a number of random walkers starting from each task cue and traversing a semantic network for a given number of time steps. We keep track of coincidences, defined as the nodes where the walkers’ trajectories intersect asynchronously, and we look for the task solutions amongst the earliest or most frequent coincident words. Our simulations show that the model with multiple walkers starting from each cue matches average human performance and correlates with individual example accuracy for a given set of its parameters. The results are validated by running the simulations on randomised versions of the original network. We have also shown that the model is consistent with findings about the structural differences between association networks of high- and low-creative individuals.
These results agree with the analysis of human data which indicates that local undirected search processes might lead to solving convergent creative tasks. We argue that the model is grounded in established theories of creativity in cognitive science and is plausible from a neuroscience perspective. The findings from this study can be the basis for further investigation into the dynamics behind the observed variance in creative ability in humans.