15:00-17:00, November 20, 2025
Main Campus B250, Okinawa Institue of Science and Technology (OIST), Japan
This session is completed and share slieds.
Introduction
At ICONIP 2025, we are hosting a special session titled “Toward Safe Brain-Inspired AI” aimed at enhancing the safety, interpretability, and ethical aspects of AI inspired by the human brain. In this session, submitted papers will be published both as part of the Special Session. We warmly welcome innovative and cutting-edge research that bridges computational neuroscience and machine learning in the pursuit of interpretable, controllable, and safe AI solutions.
Special Session Overview <Toward Safe Brain-Inspired AI>
Modern deep learning systems have achieved remarkable successes but face critical challenges in interpretability, trustworthiness, and alignment with human values. Research at the interface of computational neuroscience and machine learning suggests that incorporating key principles from the human brain—such as sparse, modular network architectures—can enable the development of more interpretable and controllable AI models. At the same time, ensuring these systems are genuinely beneficial requires robust frameworks for AI alignment. These frameworks draw on ethics, policy, and advanced neurotechnology to detect latent internal states and guide system behavior toward socially acceptable outcomes.
This special session will highlight recent progress in brain-inspired mechanistic interpretability, ethical alignment strategies, and neurotechnology-driven cognitive understanding of AI. We will explore how biologically motivated network structures enable clearer insight into internal representations, how alignment mechanisms can be integrated without sacrificing performance, and how leveraging cognitive neuroscience tools can expand our understanding of human and artificial minds.
Through invited talks (spanning academic and industrial perspectives), contributed presentations, and an interdisciplinary panel discussion, attendees will gain a holistic overview of:
- Brain-Inspired AI: Sparse and modular architectures, biologically plausible learning rules, and reverse-engineering neural computations.
- AI Alignment: Balancing advanced capabilities with safety, ethical standards, and user trust.
- Neurotechnology Applications: Using neuroimaging and cognitive modeling to interpret AI systems and refine human–machine interaction.
By uniting experts from neuroscience, AI, and ethics, this session aims to advance the development of human-centric, transparent, and responsible AI—and ultimately to foster new collaborations that propel the field forward.
Program
- Introduction (15 minutes)
- Hiroshi Yamakawa, Persistent Value of Human Brain-morphic Artificial General Intelligence (HB-AGI) [slide]
- Invited Talk(20 minutes): Neuro-aligned AI for machine neuroscience [slide]
- Hiro Taiyo Hamada (ARAYA)
- Abstract: Recent advances show that generative AI models can accurately predict neural responses and are increasingly serving as functional models in neuroscience. This talk introduces the concept of “Neuro-aligned AI,” which aims to align AI systems with the brain’s representations, dynamics, and human behaviors. We will discuss how neuroscientific tools—such as causal analysis, representational geometry, and circuit perturbation—can be applied to dissect and interpret modern AI systems. By treating AI as brain models, we propose for a framework of machine neuroscience that bridges advanced AI systems with human neural systems.
- Contributed Talks (60 minutes total)
- Nakashima T, Taniguchi A, Taniguchi T, Yamakawa H. Interpretable Anomaly Detection in a Hippocampal Formation-inspired Spatial Cognition Model. PS3-1, 890. [slide]
- Hiroaki Hamada, Hiro Taiyo Hamada, Yuto Harada, Alignment with Psychological Concept Network, PS3-2, 1025. [slide]
- Yusuke Hayashi, Decentralized Belief Propagation in LLM Agents: A Brain-Inspired Approach to AI Safety Analysis PS3-3 1084.[slide]
- Panel Discussion (20 minutes)
- A moderated session featuring the speakers and session organizers.
- Topics: Bridging neuroscience and AI; designing safe, transparent systems; ethical and societal implications of next-gen AI models.
- Closing Remarks (5 minutes)
- Akira Taniguchi
- Organizers summarize the main insights, propose next steps, and thank participants, speakers, and sponsors.
Invited Speaker
- Hiro Taiyo Hamada
- He received a Ph.D. from the Graduate School of Science and Technology at the Okinawa Institute of Science and Technology (OIST) in 2019 on systems neuroscience. Since 2022, serving as Principal Investigator in the Cabinet Office–led Moonshot R&D Program (Goal 9: “Realization of a society where people can live positively even in adversity,” Yamada PM Group). Research activities focus on computational modeling of cognitive and affective processes, as well as quantitative analysis of gait dynamics.His current research interests include whole-brain dynamical systems, and generative-AI–based digital twins
Conference Dates & VenueNeuro-aligned AI for machine neuroscience
- Conference Dates: November 20-24
This special session will be held on the afternoon of November 20th, the first day of the conference. - Venue: OIST Conference Center, Okinawa, Japan
Program Committee:
Organizers:
- Hiroshi Yamakawa (The University of Tokyo / The Whole Brain Architecture Initiative):
- Hiroshi Yamakawa is the Chairperson of The Whole Brain Architecture Initiative (WBAI), a non-profit organization, a Visiting Professor of the University of Electro-Communications and a Visiting Professor of the Kindai University. He received an MS in physics and PhD in engineering from the University of Tokyo in 1989 and 1992 respectively. His research interest is brain-inspired artificial general intelligence and concept formation.
- Akira Taniguchi (Ritsumeikan University):
- Dr. Akira Taniguchi is a Lecturer in the Department of Information Science and Engineering at Ritsumeikan University, specializing in symbol emergence in robotics, cognitive developmental robotics, and brain-inspired AI. He earned his Ph.D. in Engineering from Ritsumeikan University in 2018. He has been recognized with several awards, including the Research Encouragement Award from the Robotics Society of Japan and the Best Paper Award (Franklin V. Taylor Memorial Award) from the IEEE International Conference on Systems, Man, and Cybernetics (SMC). His research focuses on developing probabilistic generative models for spatial concept acquisition and the Brain Reference Architecture inspired by hippocampal formation.
- Takeshi Nakashima (Ritsumeikan University)::
- Takeshi Nakashima is a doctoral student at the Graduate School of Information Science and Engineering, Ritsumeikan University. He received a MS in engineering from Nagoya University in 2008. His research interest is the mechanism of adaptive behavior.
Japanese