Label Over Logic? How Source Cues Bias Human Fallacy Judgments More Than LLMs

On September 7, 2026, arXiv cs.AI published a study indicating that when judging logical fallacies, humans are significantly more influenced by source clues (such as author identity and media labels) than large language models. The study found through comparative experiments that when faced with the same content, humans are more likely to make biased judgments due to source information, while the reasoning process of large language models is relatively less affected by such external cues. This result reveals the mechanistic differences between humans and machines in processing logical fallacies, indicating that current AI systems may be superior to humans in terms of objectivity. However, it also reminds humans to be cautious about their excessive reliance on source information.