Current Position & Biography
Jun Tani is Professor of Cognitive Neurorobotics at the Okinawa Institute of Science and Technology (OIST). His research uses robots and artificial neural networks to investigate how cognition can develop through bodily interaction with the world. Rather than treating intelligence solely as the processing of abstract symbols, his work examines the dynamics linking perception, action, learning and social interaction.
Tani studied at Waseda University and the University of Michigan and earned his engineering doctorate at Sophia University. His career includes research at Sony Computer Science Laboratories from 1993 to 2001, leadership of RIKEN’s Laboratory for Behavior and Dynamic Cognition from 2001 to 2012, and a professorship at KAIST in South Korea from 2012 to 2017. He became a full professor at OIST in September 2017. OIST’s current biography also lists a visiting professorship at the Technical University of Munich beginning in December 2019.
Notable Contributions
Tani and collaborators have investigated how neural-network dynamics can organize sequences of behavior. Their work on multiple-timescale recurrent neural networks used different rates of internal change within one network to study the emergence of functional hierarchies in sensorimotor learning. Experiments trained humanoid robots on goal-directed actions and examined how simpler movements could become components of longer sequences.
Another strand of his collaborative research developed recurrent networks with parametric biases to learn and recognize multiple behavior patterns. Related embodied-language experiments explored whether links between words and actions could support generalization to unfamiliar sentences. These studies provide computational models of particular learning mechanisms, not a claim that robots reproduce the full complexity of human understanding.
His Oxford University Press book, Exploring Robotic Minds, connects robotics experiments with dynamical-systems theory and questions about actions, symbols and consciousness. It presents an approach to understanding cognition through constructing and testing synthetic systems.
In 2026, Tani was senior author of a Science Advances study with Theodore Tinker and Kenji Doya on curiosity-driven development of action and language. The team tested a virtual robot in a physics-simulated environment, combining a predictive recurrent neural network with external task rewards and an internal curiosity reward. Curiosity-driven models learned specified language tasks more efficiently and showed play-like exploration. The study’s abstract setting and task-specific comparisons are important limits: the result does not establish human-equivalent understanding, consciousness or a general-purpose physical robot.
Why Katharos Is Following Them
Tani offers a research perspective on embodied AI that connects learning algorithms with perception, movement and developmental questions. His work is relevant to evaluating what robot demonstrations reveal about cognition and what remains an interpretation of a controlled model.
Areas to Follow
Embodied language learning, predictive processing, recurrent neural networks, developmental robotics and human–robot interaction.
Reporting Scope
Coverage distinguishes simulation from physical deployment, experimental task performance from general intelligence, and models of cognition from established explanations of the human brain. Collaborative contributions and the limits of comparisons with children remain explicit.
