Abstract Robots as Mirrors of Human Cognition

Computational modeling of cognitive development has the potential to uncover the underlying mechanism of human cognition. We have been investigating what principle accounts for human and robot intelligence and what computational framework embodies such a principle. My talk introduces a neuroscientific theory called predictive processing and shows how robots as well as humans acquire cognitive abilities based on predictive processing neural networks (Nagai, Phil Trans B 2019). A key idea is that the brain perceives and acts on the environment to minimize prediction errors. Our robot experiments demonstrate that the process of minimizing prediction errors leads to the development social cognition such as imitation and altruistic behavior. The internal models acquired through sensorimotor experiences replicate the function of mirror neuron systems and thus enable robots to infer the internal states of others. Further experiments show that altered predictive processing leads to developmental disorders such as autism spectrum disorder. Aberrant predictive processing results in difficulties in learning and adaptation. I will discuss how our computational approach contributes to a better understanding of human intelligence as well as to a design of human-like intelligence in robots.