Daiju Ueda

Elucidating LLM Biases in Healthcare and Developing a Foundational Framework

Grant No.：JPMJPR2521

Researcher

Daiju Ueda

Outline

As LLM utilization expands in healthcare amid workforce shortages, concerns arise that sociodemographic biases based on race, age, gender, and socioeconomic status embedded in existing models may impede equitable medical care. This research will empirically quantify biases in multiple LLMs, including Japanese models, across medical tasks, and develop a framework integrating fairness evaluation and risk management for practical implementation in clinical settings. Through multinational collaborative research, we will verify generalizability and contribute to medical AI guideline development.

Hirotaka Osawa

The Augmented Society Powered by Human-Dual-Agent Model

Grant No.：JPMJPR2522

Researcher

Hirotaka Osawa

Outline

The project explores the “dual-agent” relationship involving humans and two social agents across three dimensions: theory and implementation, design, and ELSI verification. It offers a comprehensive classification of models based on internal states, roles, and discourse structures; designs, implements, and evaluates applications derived from these models; and presents future scenarios and societal risks through SF prototyping. The goal is to promote metacognition, foster an augmented society, advance safe and comfortable human–AI coexistence and collaboration, and establish guidelines for the societal implementation of emerging technologies.

Kuniaki Saito

Multimodal Foundation Models for Adaptation to Domain-Specific Knowledge

Grant No.：JPMJPR2523

Researcher

Kuniaki Saito

Outline

The goal of this project is to overcome the limitations of existing AI models, which are biased toward “general knowledge” derived from large-scale web data, by developing a multimodal foundation model capable of adapting to “specialized knowledge” that is highly domain-specific, context-dependent, and difficult to collect or annotate. Specifically, we seek to build a model that can autonomously adapt across diverse domains and knowledge systems, while handling a wide range of tasks such as visual question answering from input images.

Anna Suzuki

Framing Misalignments for Co-Creation: Building Dialogue Support Infrastructure

Grant No.：JPMJPR2525

Researcher

Anna Suzuki

Outline

This research aims to develop a dialogue support infrastructure that facilitates co-creation among diverse actors with differing values and emotions. Recognizing that judgments are shaped by a triadic structure comprising logic, ethics, and aesthetics, the project builds a theoretical and mathematical model to visualize and simulate “framing misalignments.” These misalignments, often hidden beneath surface-level disagreements, are key to understanding the roots of conflict and unlocking pathways to mutual understanding. By integrating dynamic simulations and visualization tools, the proposed infrastructure enables intuitive recognition of value gaps and supports inclusive, creative decision-making. The system will be implemented across multiple fields and its applicability to human-AI collaborative dialogue design will also be explored. Ultimately, this research offers a foundation for empathetic, value-aware communication in complex societal contexts.

Hiroki Nishikawa

Explainability of Causal Responsibility on Mixed Traffic with Autonomous- and Human-Driving Systems

Grant No.：JPMJPR2526

Researcher

Hiroki Nishikawa

Outline

This study proposes a framework for explainable responsibility allocation in mixed traffic, where human-driven vehicles and AI-driven autonomous vehicles interact. The framework is built on the quantification of causal responsibility before and after accidents. This framework expects to realize traffic order that proactively suppresses accidents. In additoin, even if an accident occurs, the framework enables a quantitative explanation of how responsibility is attributed to each vehicle.

Yuuki Nishiyama

Personalized AI Development and Evaluation Platform Using Structure-Mapped Synthetic Time-Series Behavioral Data

Grant No.：JPMJPR2527

Researcher

Yuuki Nishiyama

Outline

This research aims to construct a development environment for safe and flexible personalized AI in an era of human-AI symbiosis and collaboration. We research and develop a platform for generating structure-mapped synthetic time-series behavioral data that effectively replicates the structure of multimodal real time-series behavioral data obtained through passive mobile sensing while ensuring privacy protection. Through the development and evaluation of generative AI specialized for time-series behavioral data, we aim to achieve global deployment and standardization of the proposed platform.

Yuto Hasegawa

Development of personalized AI-based social stimulation for mitigating social isolation-included stress in aged mice

Grant No.：JPMJPR2528

Researcher

Yuto Hasegawa

Outline

This study aims to evaluate the effectiveness of real-time, personalized AI-based virtual social stimulation for mitigating mental and physical stress caused by social isolation in rapidly-aging societies. The effectiveness will be examined using aged mouse models, applying cutting-edge neuroscience approaches to assess multiple biological domains, including neural activity, immune responses involving the gut microbiota, and behavioral changes. Through scientific visualization of these biological impacts, the study aims to contribute to the societal implementation of evidence-based AI technologies for stress mitigation.

Yuichi Hiroi

Infrastructure for Human–AI Collaborative Community Building in Metaverse

Grant No.：JPMJPR2529

Researcher

Yuichi Hiroi

Outline

This research project involves developing a human-AI collaborative system that supports community formation by transferring tacit knowledge from the community support staff of Cluster, a commercial metaverse platform, to AI agents. Using comprehensive user behavioral records, we formalize staff expertise through behavioral pattern analysis. Then, we implement embodied AI agents that incorporate this knowledge and conduct large-scale field studies to validate the approach. This project aims to systematize community formation support techniques in metaverse environments and extend their application to the real world.

Yin Long

AI-enabled Healthcare Infrastructure for an Aging Society

Grant No.：JPMJPR252A

Researcher

Yin Long

Outline

This project proposes a next-generation medical support model for aging regions facing difficulties in healthcare access, where AI and local residents co-evolve. Specifically, it analyzes medical demand at the 1-km mesh level based on medical records and behavioral logs, extracts health concern signals from sources such as SNS and newspapers, and develops elderly-oriented agents that integrate generative AI with knowledge graphs. These agents provide flexible medical route suggestions and behavioral support tailored to regional conditions, aiming to establish a new medical and social infrastructure that enhances accessibility and resilience in super-aging communities.