Hisashi Kashima

Human Computation for Human-AI Collaboration

Research Director

Hisashi Kashima

Collaborator

Hiromi Arai	Unit Leader Center for Advanced Intelligence Project RIKEN
Satoshi Oyama	Associate Professor Faculty of Information Science and Technology Hokkaido University
Junichiro Mori	Associate Professor Graduate School of Information Science and Technology The University of Tokyo

Outline

We aim to establish the foundation of human computation for designing trustworthy human-AI collaborative systems by (i) human-in-the-loop machine learning for human-AI collaborative data analytics, (ii) defining and optimizing reliability and trust for human computation systems, (iii) addressing ethical issues for social acceptance of human computation, and (iv) supporting human intellectual and creative activities and developing human capabilities through human computation.

Shinya Takamaeda

D3-AI: Distributed AI for Dynamic and Diverse Environments

Research Director

Shinya Takamaeda

Collaborator

Masaaki Imaizumi	Associate Professor Graduate School of Arts and Sciences The University of Tokyo
Tomoya Kitani	Associate Professor Academic Institute Shizuoka University
Hideki Takase	Associate Professor Graduate School of Information Science and Technology The University of Tokyo
Kentaro Yoshioka	Assistant Professor Faculty of Science and Technology Keio University

Outline

We develop fundamental technologies for energy-efficient distributed AI systems based on federated learning for respecting the diversity of users and data and the spatial and temporal environmental variations. We define D3-AI as the distributed AI with four reliability capabilities: privacy, fairness, dynamic adaptability, and energy efficiency. We pursue researches on the fundamental technologies and applications of D3-AI in collaboration with machine learning theory, computer architecture, IoT platform, and data processing.

Ichiro Takeuchi

Assessing Reliability of AI-driven Hypotheses in Static and Dynamic Environments and Its Application to Medical Science

Research Director

Ichiro Takeuchi

Collaborator

Koichi Ohshima	Professor Department of Medicine Kurume University
Atsushi Kawaguchi	Professor Department of Medicine Saga University
Jun Sakuma	Professor Faculty of Engineering, Information and Systems University of Tsukuba
Shigeyuki Matsui	Professor Department of Medicine Nagoya University

Outline

We develop a mathematical and computational framework for assessing the reliability of AI-driven hypotheses for AI-driven science and technology by extending the traditional statistical hypothesis testing framework. Our main focus is to develop new theories, algorithms, and software for statistical hypothesis testing that is valid for adaptively constructed AI-driven hypotheses. We study AI-driven hypotheses in static and dynamic environments and pursue how to evaluate their reliability from both theoretical and practical perspectives. We develop a protocol for evaluating the reliability of AI medical technology based on the mathematical and computational framework and demonstrate the protocol through the development of an AI-based pathological diagnosis system for malignant lymphoma.

Seiji Yamada

Trust Interaction Design for Convincing Human-AI Cooperative Decision Making and its Social Penetration

Research Director

Seiji Yamada

Collaborator

Tetsuo Ono	Professor Faculty of Information Science and Technology Hokkaido University
Hirokasu Kumazaki	Professor Institute of Biomedical Sciences Nagasaki University
Kazunori Terada	Associate Professor Faculty of Engineering Gifu University
Takashi Hara	Professor Faculty of Engineering Gifu University

Outline

Our purpose is to build human-AI cooperative decision-making systems for trustworthy AI. In this project, we develop trust calibration AI (TCAI) which can detect over/under-trust by monitoring humans’ selection behaviors in human-AI collaboration and adaptively facilitate humans’ trust calibration by themselves. The TCAI uses calibration cues as stimuli that facilitate humans’ voluntary trust calibration. Finally, we apply the TCAI to human-AI cooperative imaging diagnosis and medical examination.