Artificial Intelligence and Human-Induced Seismicity: Initial Observations of ChatGPT

Abstract

Abstract Freely available artificial intelligence (AI) tools, such as the Chat Generative Pre-trained Transformer (ChatGPT), offer an alternative method to online search engines for scientific results and opinions to be automatically collated into concise summary prose. We applied this approach to human-induced seismicity by asking ChatGPT common, earthquake-related questions before asking questions about natural, induced, and debated earthquakes. ChatGPT provided rudimentary descriptive distinction between natural and induced seismicity, and for clear-cut cases of each type provided a response giving the scientific consensus. For debated cases, a natural origin was implied by responses focusing on fault rupture and tectonic setting. The inclusion of the word “induced” in the question phrasing led to discussion on possible induced origins, but not all response information was consistent with our knowledge of the scientific literature. ChatGPT did not provide an answer to a case more recent than its training data. For most questions, ChatGPT tended to include irrelevant information to increase response length. Online AI tools could become a mainstream technology, particularly for nonspecialists, to obtain concise summaries of published science. However, it is important to recognize the limitations of the current technologies, particularly sensitivity to question wording and inability to correctly reference scientific material, especially where a definitive answer does not yet exist to the question asked.