Research Blog / Data Repository
Cite this article
AI is supposed to distinguish fact from speculation. But when institutions are accused of wrongdoing, “not proven” can quietly become “not true.” Research on authority bias helps explain why—and shows how better prompting can force AI to separate evidence, inference and genuine uncertainty.
AI detectors are getting better. That does not mean they can tell who actually authored a piece of work. Substack’s Pangram integration exposes a deeper problem: detecting machine-generated prose is not the same thing as measuring human thought, research, creativity, or intellectual contribution.
Cameras increasingly identify people through faces, clothing, vehicles and even movement. But researchers are developing an equally strange counterculture of CV Dazzle makeup, adversarial clothing, infrared camouflage and other ways to become harder for machines to read.