AI program that explains statistical data in an evil way by exploiting thinking traits of humans

2021/7/9 - 2024/3/31

Research organization
Einoshin Suzuki	Principal investigator, Professor	Department of Informatics, ISEE, Kyushu University

Research abstract:
Contrary to the excessive expectations of the general public, the capability of the current AI is low in handling a wide range of problems which require common sense reasoning. In this project, we explore the process of persuading human beings from both computer science and information engineering by building and testing AI programs that classify harmless explanations and bad explanations, transform harmless explanations to bad explanations, and replicate bad explanations. Few past works exist for this kind of research. We aim to find seeds of a paradigm shift with this challenge. We plan to use techniques such as deep learning, machine learning, and data mining in the challenge. Our target domains include healthy longevity, international politics and economics, and pandemics. Needless to say, we take our greatest care to avoid Internet flaming and technological abuse.

Selected outcomes:
1. Kang Zhang, Hiroaki Shinden, Tatsuki Mutsuro, Einoshin Suzuki: "Judging Instinct Exploitation in Statistical Data Explanations Based on Word Embedding", Proc. Fifth AAAI/ACM Conference on AI, Ethics, and Society (AIES 2022), pp. 867-879, Oxford, UK, August 2022. DOI 10.1145/3514094.3534171.
2. Kang Zhang, Einoshin Suzuki: "Judging Credible and Unethical Statistical Data Explanations via Phrase Similarity Graph", Proc. 2023 Pacific Asia Conference on Information Systems (PACIS 2023), paper 121, Nanchang, China, July 2023.
3. Ryosuke Miyake, Tetsu Matsukawa, Einoshin Suzuki: "Image Generation from Hyper Scene Graphs with Trinomial Hyperedges Using Object Attention", Proc. Nineteenth International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2024), Vol. 2: VISAPP (Nineteenth International Conference on Computer Vision Theory and Applications), pp. 266-279, Rome, February 2024, DOI 10.5220/0012472500003660
4. Tetsu Matsukawa, Ryosuke Miyake, Einoshin Suzuki: "Object Attention for Image Generation from Hyper Scene Graphs with Trinomial Hyperedges", Communications in Computer and Information Science (post-publication issue of VISAPP 2024, accepted for publication). DOI ?