Detecting computer-generated disinformation

Abstract Modern neural language models can be used by malicious actors to automatically produce textual content looking as it has been written genuine human users. Due progress in the controllability of computer-generated text, there is a risk that state-sponsored may start using such methods for conducting large-scale information operations. Various detection algorithms have suggested research literature identify texts produced model-based generators, but these are often mainly evaluated on test data from same distribution they trained on. We evaluate promising Transformer-based large variety experiments involving both in-distribution and out-of-distribution data, well evaluation more realistic in-the-wild data. It shown generalizability detectors questioned, especially when applied short social media posts. Moreover, best performing (RoBERTa-based) detector non-robust also basic adversarial attacks, illustrating how easy avoid current state-of-the-art algorithms.