Year: 2025
Type: article
Abstract: Large language models and machine learning methods have been used to detect pseudoscience and scientific misinformation online. The goal of the project is to develop tools to combat the rise of misinformation that is rapidly drowning out legitimate sources of science information. In the initial phase, undergraduate science majors gathered 800 short, non-technical articles from the Internet on a wide range of physical and life science topics, along with a similar sized set of articles on pseudoscience topics, such creationism, and climate change denial. A special focus is on the archetypal astronomy-related topics of astrology and UFOs, which both have enormous visibility on the Internet. Multiple students classified each article as “real” or “fake” and inter-rate reliability was high. In a testbed on evolution and climate change, machine learning flagged misinformation in each category with over 90\% reliability. The second phase used a state-of-the-art, open-source, language model. Tens of thousands of articles were gathered, and scientific misinformation was successfully flagged across all ten topics, including astrology and UFOs, with over 95\% reliability. This technology will be deployed in a web site and a web browser extension that will allow users to insert a URL and see returned a Bayesian color-coded probability that the purported science content is legitimate. The implications of the new capabilities of large language models like ChatGPT for the evolving battle against misinformation are considered.
Source: Astronomy Education Journal
Institution: University of Arizona
Language: English
Subfield: Sociology and Political Science
Field: Social Sciences
Domain: Social Sciences
Open Access status: diamond
OpenAlex ID: W4416740409
