An Experimental Investigation of Adaptive Algorithm Understanding

There have been few studies on a cognitive model for algorithm understanding in a human-computer cooperative situation. In the present study, we conducted an experiment with participants to investigate the cognitive process of higher level abstraction (algorithm understanding) performed in a humancomputer collaboration task. The most recently used (MRU) algorithm, known to be one of the simplest adaptive algorithms, and probabilistic MRU algorithm were used to test the human capability to understand an algorithm. The experimental results showed that inductive reasoning in which participants observed the history of computer action, and they updated a statistical model while restricting their focus on a certain history with deteministic bias and Markov bias played key role to correctly understand the MRU algorithm. The results also showed that deductive reasoning was used to understand algorithms when participants rely on prior knowledge, and that there was a case in which the algorithm, even known to be the simplest one, was never understood.