Research overview

It is becoming possible for AI to learn human-derived information such as faces, bodies, speech, and natural language, and to generate synthetic information that is almost indistinguishable from the real thing. Synthetic information is being used in a variety of fields, including communication and entertainment. On the other hand, as a negative aspect of synthetic information, there is an international threat of criminals and attackers using AI to generate and spread fake information such as fake videos, fake audio, and fake documents for the purposes of fraud, thought guidance, and public opinion manipulation.

Research objectives

In this research, we will establish technology to prevent the spread of fake information that can provide detailed information, including not only authenticity judgments such as whether or not it was generated by AI, but also the history of how it was generated, for multimodal media composed of multiple modalities, in addition to unimodal media such as images, videos, audio, and text. We will also establish technology to prevent the spread of fake information, which can prevent the automatic collection of media by third parties without their consent, the modification of media by AI, and the learning of fake information in advance. Furthermore, we will establish a fake information research platform (SYNTHETIQ X) suitable for the practice of fake information generation, defense, and prevention, consisting of a large-scale dataset of real and fake information and the generation models for various fake information, as well as the defense models and the prevention models developed in this research, and a number of defense models related to this research, for unimodal and multimodal media.