The Internet and, more recently, social networks deliver continuous content to their users, who are
more connected than ever but overwhelmed and not necessarily better informed. It is often challenging
to distinguish trustworthy from malicious content, especially when it appeals to emotions
and beliefs and comes from familiar sources. In recent years, we have witnessed an increased
volume of false messages in social networks, which tend to spread faster and broader than truthful
information. Not surprisingly, disinformation, meaning incorrect information purposely intended to
harm, has been highlighted as a significant contributor to the polarisation and demeaning of democratic
institutions.
Debunking false information is complicated and time-consuming, requiring expert participation and
manual work. Hence, computational methods based on massive data processing technologies are
envisioned as essential to better understanding and mitigating disinformation. Accordingly, Artificial
Intelligence (AI) has emerged as a suitable toolbox for disinformation detection and fact-checking.
However, the remarkable AI advances in natural language processing, social network analysis or
even synthetic content generation have yet to permeate outside the research labs. The motivation
behind this report is the realization that there is still a considerable gap between AI research labs
and practitioners’ daily challenges on fighting disinformation.
This report provides a concise guide to navigating the recent literature on AI for fighting disinformation,
emphasizing the region covered by the IBERIFIER project. The report covers the fundamentals
of Machine Learning (ML) algorithms, describes their application to several facets of disinformation
analysis, and maps datasets and tools generated in the Iberian research community. The main conclusion
is that there is not only a great need to transfer technologies from research to the industry but
also to redirect research efforts toward human-supported tools rather than fully automated solutions
—which are often biased and very domain-specific.
CITATION
Montoro Montarroso, A., Camacho, D., Martín, A., Torregrosa, J., Rosso, P., Chulvi, B., Rementería, M.J., Calvo Figueras, B., Philippe, O., Molina Solana, M., Cantón Correa, J. & Gómez Romero, J. (2023). Is the ‘Ai toolbox for disinformation’ ready? Pamplona: IBERIFIER.