Artificial Intelligence in the Mitigation of Disinformation in the Governance of the Democratic System

Abstract

This research presents the development of an evaluation and protection framework based on AI algorithmic models to mitigate disinformation in the democratic governance of Ecuador; to this end, an applied mixed‑methods approach was used, incorporating a literature review, simulations of diffusion on social networks, and news classification, contextualized with surveys of relevant actors in the media ecosystem. According to the findings obtained, the social‑network simulation showed the formation of echo chambers and the amplification of like‑minded narratives, a phenomenon intensified by recommendation algorithms that prioritize user engagement; to mitigate diffusion in our simulation, automated models for detecting fake news achieved 91.67% accuracy; however, a key finding was that the models’ decisions rely on statistical artifacts and stylistic biases of the dataset rather than on a semantic understanding of veracity, revealing fragility and limited generalization. In this sense, an evaluation and protection framework was developed, aimed at ensuring transparency, minimizing biases, and strengthening public trust to bolster democratic governance; concrete guidelines were proposed, such as the requirement of mandatory interpretability and a machine‑learning operations lifecycle to combat model obsolescence, ethically prioritizing the mitigation of amplification over censorship.