Fact checkers are overwhelmed by the amount of false content that is produced online every day. To support fact checking, several research e orts have been focusing on automatic veri cation methods to assess claims and experimental results show that such methods enable e ective labeling of textual content. However, while fact checkers start to adopt some of these tools, the misinformation ght is far from being won. In this talk, I cover the opportunities and limitations of computational fact checking and its role in ghting misinformation.
P. Papotti
to estimate the factuality of a claim, for example, by analysing the language used,
who is spreading or repeating a claim, and if it has already been veri ed [
New algorithms keep increasing the accuracy and the scope of the computational veri cation approaches, but several issues prevent their adoption in fact checking organizations. One important problem is that most real claims are complex and subtle and go beyond the reach of the automatic solutions. For example, a claim stating \COVID vaccines have been adopted despite incomplete testing campaign over humans" require fact checkers to collaborate with domain experts, collect and check multiple pieces of evidence, and write a rebuttal that is aware of the context and the framing of the claim. While algorithms can help in this process, the nal veri cation (and explanation of such decision) requires a human understanding beyond current AI solutions.
Finally, I conclude with observations about the next steps that we need to take to improve the coverage and e ectiveness of computational fact checking. We de nitely need to keep pushing the technical research. In this agenda, one of the most important points is to focus on interpretability of the proposed solutions. Algorithms and models should output explainable checking decisions, with considerations about possible bias in the reference information. However, technical advancement is not going to be e ective in reducing the misinformation problem unless a real holistic approach is taken to tackle it. Misinformation is a societal issue that involves actors ranging from politicians to tech companies, with duties that are still to be de ned clearly. As computer science researches, we have the responsibility to design and deploy the best algorithms and models to attack the problem, but also the obligation to make clear that misinformation is not a crisis that will be solved only by developing more advanced AI solutions.