Unraveling the Mysteries of Human Intelligence through Robots

While research in artificial intelligence and intelligent robotics aimed at helping people is advancing, our laboratory is engaged in research in a slightly different field called "cognitive robotics." In cognitive robotics, we aim to understand the computational mechanisms that realize human cognitive functions such as perception, action, and learning by mobilizing knowledge from various research fields, including cognitive neuroscience, robotics, and machine learning.

For example, consider the cognitive function of imitation, where a baby mimics the movements of a caregiver or others. What methods can be used to understand this? The most straightforward approach that comes to mind is to actually observe the imitation and measure the brain's neural activity. However, such "analytical methods" alone cannot reveal the underlying mechanisms. Therefore, a "constructive approach"—formulating a hypothesis about the mechanism and then building and running a system to test it—is considered effective. In a baby's development, it is not simply an isolated brain that learns; the interaction between the brain, body, and environment is essential. To account for these interactions, cognitive robotics emphasizes not only building a computational model of the brain but also implementing the model on a robot with a physical body and verifying it in a real-world environment.

The mechanisms of human cognitive functions may seem incredibly complex. We aim to explain these seemingly complex subjects as simply as possible. As a computational principle for this, we are focusing on the idea of prediction error minimization. This is expected to provide a unified explanation for various cognitive functions. For example, perception can be seen as changing predictions, action as changing sensations, and learning as changing synaptic connections in the brain. In this way, each cognitive function can be reinterpreted as a different information processing process to achieve the common goal of prediction error minimization.

The realization of intelligent robots that can coexist and be active with humans in daily life environments is anticipated. However, unlike simple repetitive tasks in factories, this requires the generation of adaptive and flexible behavior that is context-dependent. To achieve this, it is crucial to understand the mechanisms by which humans actually accomplish these things. Furthermore, in the field of psychiatry, there has been a growing emphasis in recent years on how to seamlessly connect the neural, cognitive, and behavioral levels to understand mental disorders. Cognitive robotics, which aims to unravel the mysteries of human intelligence through robots, plays an important role in contributing to the development of other fields such as robot learning and computational psychiatry, which conduct this research.