NEC uses AI to analyse information trustworthiness

This is being carried out as part of a project under Japan’s Ministry of Internal Affairs and Communications to develop and demonstrate technologies to counter false and misleading information on the Internet.

The effectiveness of the technology will be evaluated by fact-checking organisations such as the Japan Fact-check Centre and mass media, including broadcasters, with the aim of improving its trustworthiness. The period of this demonstration project is from August 2024 to March 2025.

Framework of the new technology

The large volume of false and misleading information on the Internet has become a growing problem, and countermeasure technologies are attracting greater attention. Fact-checking organisations and media outlets that are responsible for disseminating highly reliable information are burdened by the large amount of labour that is required to determine the trustworthiness of information. NEC is developing a technology to support fact-checking by utilising state-of-the-art AI, such Large Language Models (LLM) to analyse and judge the authenticity of content.

Outline of the technology

The technology uses AI to analyse whether content consisting of multiple types of data (text, images, video, and audio) is false or misleading. First, the technology detects whether images and other data have been processed, AI then recognises multiple types of data and converts them into text. Next, LLM that specialise in false information analysis evaluate the text to determine whether the content is correct, whether the information has a reliable source, and whether there are any inconsistencies among the data (e.g., discrepancies between the text and video content).

In addition, the technology creates reports in a format similar to reports and articles prepared by experts at fact-checking organisations. The technology also helps to improve fact-checking operations by enabling adjustments according to a user's instructions, such as deleting unreliable information or adding new information.

Fact-checking organisations such as the Japan Fact-check Centre and mass media, including broadcasting stations, will be evaluating the performance of the technology, aiming to improve its trustworthiness and for it to be available for practical application in FY2025~2026.