Fighting fake news using AI is very hard
Fighting fake news using AI techniques is way harder than many people (including Mark Zuckerberg) think. This article reports on some approaches and admits that the results are very poor. The problem seems to me that one aims to search for certain meta-characteristics of fake-news articles, such as (apparently) repetitive occurrences of the same words. But such characteristics do not go to the core of what fake news is. Clearly among the things really needed to detect fake news are being able to test for (1) consistency, (2) coherence, and (3) truthfulness. Of these three, I think (1) should in principle be the easiest for an AI algorithm. Yet, I do not know of any AI-system that can detect inconsistencies in or among written texts.