Memory for the Random: A Simulation of Computer Program Recall

Contrary to a widely held belief, experts recall random material better than non-experts. This phenomenon, predicted by the CHREST computational model, was first established with chess players. Recently, it has been shown through a meta-analysis that it generalises to nearly all domains where the effect has been tested. In this paper, we carry out computer simulations to test whether the mechanism postulated with chess experts – the acquisition and use of a large number of chunks – also applies to computer programming experts. The results show that a simplified version of CHREST (without the learning and use of high-level schemata known as templates) broadly captures the skill effect with scrambled programs. However, it fails to account for the differences found in humans between different types of randomisation. To account for these differences, additional mechanisms are necessary that use semantic processing.