Abstract: Online discussion among civilian is important and essential for next-generation democracy. Providing good support is critical for establishing and maintaining coherent discussions. Large-scale online discussion platforms are receiving great attention as potential next-generation methods for smart democratic citizen platforms. Such platforms require support functions that can efficiently achieve a consensus, reasonably integrate ideas, and discourage flaming. Researchers are developing several crowd-scale discussion platforms and conducting social experiments with a local government. One of these studies employed human facilitators in order to achieve good discussion. However, they clarified the critical problem faced by human facilitators caused by the difficulty of facilitating large-scale online discussions. In this work, we propose an automated facilitation agent to manage crowd-scale online discussions. An automated facilitator agent extracts the discussion structure from the texts posted in discussions by people. We conducted a large-scale social experiment with Nagoya City’s local government.

Abstract: The talk briefly addresses the role of data usage control for data sovereignty, that is a main issue in current technologies. It introduces the notions of data centric policies, policy refinement and policy enforcement in several scenarios, including cyber threat intelligence management.

The dramatic spread of smartphones equipped with high performance sensors has enabled filming and recording of other people's faces, voices, and fingerprints. These can also be shared in cyberspace, not only violating privacy but also illustrating the risk of breaches of biometric authentication. Moreover, with the evolution of AI technology and the high-quality biometric data, malicious actors can create high-quality fake media such as deepfakes, which may negatively impact people's ability to make decisions. This talk will outline these threats and introduce technology for users to control distribution of their own biometric data in cyberspace and technology for detecting fake media.

Complex and social systems consist of many interacting nodes that exchange information and can compete and/or collaborate with each other. For these nodes to coordinate effectively, they need to engage in joint decision making through the process of negotiation. In this talk, I will present my ongoing research on intelligent and autonomous systems that can negotiate in a wide variety of upcoming domains, such as smart energy trading, the Internet of Things, and digital privacy.

With the rapid development of deep learning and machine learning, the accuracy of recognition of images, voice, etc., has improved to the extent that it exceeds human recognition ability. In the future, machine learning is expected to make decisions that affect humans' life, death, and well-being. The decision-making process of deep learning is hidden in a complex black box and is not necessarily reliable. In this talk, I will give an overview of the issues related to the reliability of AI, especially on security and explainability of AI, and discuss the recent progress in research toward achieving highly reliable AI.

This work stemmed from a collaboration in the medical domain where our partner was only allowed to release the marginals of the true data, and the marginals of the error made by a candidate hypothesis on these data.The proposed Extremely Private Supervised Learning (ExPriL) approach proceeds by generating synthetic data fitting the true marginals, iteratively learning a hypothesis and getting the marginals of the error as supervisory signal to revise the hypothesis. The epsilon-differential privacy of the mechanism is provably enforced under the assumption that the number of iterations is known in advance. Joint work: Armand Lacombe, Saumya Jetley.

This talk addresses communication-efficient distributed machine learning toward AI-driven data processing in IoT networks. Federated Learning (FL) is an emerging machine learning technique that can enable many intelligent IoT applications, by allowing for AI training at distributed IoT devices without the need for data sharing. However, in the IoT networks, the conventional FL is suffered from the narrowband and lossy communications and non-IID data. To overcome these issues, based on the idea of leveraging an unlabeled open dataset, we propose a distillation-based semi-supervised FL (DS-FL) algorithm enabling communication-efficient model training with non-IID data.

Abstract : The actual cloud shift mega trend is not in line with the demand of reduction of carbon footprint of ICT infrastructures by 2030 to achieve climate neutrality. Edge computing is the evolution of the cloud model, avoiding the transfer of data, supporting resilience, real time operations, security & privacy while at the same time reducing energy consumption. The presentation highlights the background and a research agenda for addressing these challenges such as photonics and 6G as well as new AI ICT architectures.

Artificial intelligence is a medium that processes a variety of data to convey information to us. And if a medium itself is a "message," as Marshall McLuhan once argued, then artificial intelligence itself must also be a message of some sort. Then, what kind of message is artificial intelligence? In this presentation, I will attempt to elucidate the ethical issues facing artificial intelligence by deciphering the message. In particular, I will argue that artificial intelligence is an innovative tool for conducting risk analysis involving humans as its subjects, based on huge data collected through various devices, and that it can foster an attitude that interprets others as the source of risk.

In 1998 David Chalmers and Andy Clark introduced the very influential Extended Mind Hypothesis in philosophy: »Where does the mind end and the rest of the world begin?«. Chalmers and Clark argue that we are not just using tools as a consequence of our thoughts, but our mind is actually extended increasingly into our environment and artefacts.

In 1998 there was no smartphone and no wide spread application of machine learning (»artificial intelligence«). Both are becoming ubiquitous now, and this has the potential — picking up where Clark and Chalmers left off — so not only affect our digital processes but literally our mind, thinking and deliberation.

What are ethical, social consequences and responsibilities?