Benchmarking exercise and identification of AI tool to detect false information on food

Abstract

The study has the objectives to conduct desk research to identify available AI-based solutions for detecting false information and to benchmark the identified solutions against a set of features to take into account for an implementation at the European Food Safety Authority (EFSA). The desk research identified 58 potential commercial platform candidates, 39 other but less applicable candidates and 8 related solutions from the intelligence sector. It further reviewed 5 EU tools, 19 EU funded projects, 17 academic projects and 3 other resources. Based on desk research, an initial benchmarking was performed using refined criteria, resulting in the identification of 10 top candidates for deeper analysis. These candidates were assessed through in-depth research, including direct provider interactions. Questionnaires were sent to lower-scoring platforms to further validate the desk research. A final benchmarking analysis, accompanied by a SWOT analysis, was produced for the top candidates. A framework for evaluation, incorporating various levels of actions, actors, tools, and knowledge, was developed and linked to a maturity model to assess the platforms’ suitability at different maturity levels. The study identifies solutions that are suitable for EFSA's current and future needs and could provide insights for other agencies and organizations with similar objectives on misinformation monitoring at scale.