Iizuka and Ikegami have done a great job of presenting their interesting research results. The authors created some virtual robots to participate in interaction simulations, in which the robots performed turn-taking behaviors, imitation, and role-playing games, learning through an artificial recurrent neural network trained by a genetic algorithm.

The authors’ study focused on the emergent diversity of behaviors and their relation to human cognitive experiments, making some comparisons to, and making some claims about similarities with, experiments involving children. They describe the dynamic repertoires of their virtual creatures, and a dynamic evolves from the interaction that seems to be not only intelligent, but also cognitive. The authors perform experiments with their artificial agents on things like predictability power and adaptability to a noisy environment.

The analogies between the experiments performed by the authors and human psychological experiments are very interesting and rich. The authors claim that even when a human behavior is very complex, they can characterize some facet of human communication. They state that psychological phenomena can be studied as dynamical systems, in which graphical phase states can be analyzed, exposing patterns and attractors, and classifying behaviors.