Conference on Cognitive Computational Neuroscience

Computer programming is an increasingly critical skill in modern society. What makes it possible for the human brain to adapt to learn coding? Previous studies suggest that code experts activate fronto-parietal reasoning networks during Python code comprehension. Here we investigated what pre-existing neural representations enable learning to program by scanning programming-naïve students before and after their first programming course. Before the course, participants took part in a functional MRI scan while reading custom-made pseudocode (i.e., plain language descriptions of programming algorithms). After taking the introductory programming course, the same students were scanned again while reading real Python code. SVM classifiers were trained in both sessions to decode FOR and IF algorithms based on the spatial pattern elicited by pseudocode and real code, respectively. We found above-chance decoding accuracy for real Python functions in the logical reasoning network after taking the course. Moreover, we even found above-chance accuracy for pseudocode before participants learned anything about programming. Our findings suggest that learning to program in Python recycles pre-existing representations of logical algorithms in the fronto-parietal network.