REFLEX

Disinformation has evolved from simple factual errors into a sophisticated socio-cognitive phenomenon that exploits human vulnerabilities. In an era where GenAI can generate persuasive content at scale, the urgent challenge for society is not just verifying if information is true, but preserving cognitive sovereignty. The REFLEX project (REsilience via Formal Linguistic EXplainability) addresses this by shifting the focus from automated censorship to intellectual self-defense, empowering individuals to understand how they are being manipulated rather than just being told what is fake.

To achieve this, we are designing a Computational Persuasion Estimator. Unlike standard tools, this engine operationalizes discursive modalization, analyzing specific linguistic structures, such as authority framing or emotional intensity, to quantify the persuasive intentionality behind a message and that manufacture credibility.

This ambition relies on a strategic collaboration between the Laboratoire d’Informatique et Systèmes (LIS), the Laboratoire Parole et Langage (LPL), and the University of Windsor. This partnership drives a reciprocal breakthrough for both fields. Linguistics empowers the project by providing the qualitative rules of influence. This contribution advances the field of Artificial Intelligence by solving its "black box" problem: instead of giving an opaque score, the model becomes capable of explaining why a text is manipulative. Conversely, Artificial Intelligence empowers the project with the processing power to analyze millions of texts. This contribution advances the field of Linguistics by offering a new computational lens, allowing theoretical models to be validated at scale and transforming them into predictive tools.

Rooted in civic action, REFLEX partners with the citizen media organization Anonymal to deploy these models directly into educational programs for critical thinking. Finally, the project leverages the University of Windsor's high-performance computing resources to simulate the spread of these persuasive narratives, ensuring our defense systems evolve as fast as the threats.

This project is designed to train a new generation of transdisciplinary experts. The candidate will develop a dual competence, mastering the bidirectional translation between qualitative concepts and rigorous computational models. This training positions them for strategic roles such as AI Safety Architect, capable of aligning algorithmic performance with societal standards and leading the transition towards ethical, human-centric AI.