A trustworthy AI for the analysis of the written press – TRUSTEDNEWS

The project TRUSTEDNEWS (Astrid Maturation) is a continuation of the project Astrid AI 2021 HYBRINFOX that dealt with hybrid AI approaches to fake news detection and the measurement of informational vagueness in texts. The goal of TRUSTEDNEWS is to integrate the industrial components of HYBRINFOX in order to transition from a TRL of 3-4 to a TRL of 5-6. Those components involve the collection of press texts and the combination of expert systems with neural models of discriminative AI in order to automate the analysis of speeches along several dimensions: whether as fact-reporting, opinions, as agenda-driven or as displaying specific persuasion techniques. The originality of this approach is to promote an explainable AI approach that is usually absent from data-driven approaches relying exclusively on neural networks. The main goal is to elaborate an integrated tool and to test it in a realist environment, on new use cases, in order to give military users and lay practitioners a clear indication of the trust that they can have in a specific online document or a set thereof. The project is an association with three scientific partners specialized in AI, NLP and human-machine teaming (CROSSING, IRISA, LIP6) and two industrial partners including one with a specialization on signal processing and control (AIRBUS DEFENSE AND SPACE) and one of the major actors of the French press (OUEST-FRANCE). The integration of LIP6 and OUEST-FRANCE is an innovation compared to HYBRINFOX, with an expertise of LIP6 on topic analysis and an expertise of OUEST-FRANCE’s data science team on the exploitation and investigation of of press corpora and of Ouest-France's rich array of publications.