Trust Large Language Models at Your Own Risk - ExBulletin
2022-11-23
ExBulletin
The information displayed in the AIM should not be reported as representing the official views of the OECD or of its member countries.
Meta launched Galactica, an AI language model designed to generate scientific content, but quickly withdrew it after users found it produced false, biased, and harmful outputs, including racist remarks and dangerous misinformation. The system's inability to distinguish fact from fiction led to ethical criticism and reputational harm.[AI generated]
Why's our monitor labelling this an incident or hazard?
The event involves the use and deployment of an AI system (Galactica, a large language model) whose outputs have directly led to harms such as dissemination of biased, inaccurate, and potentially harmful scientific information. This constitutes harm to communities and a violation of the obligation to provide truthful information, fitting the definition of an AI Incident. The article describes realized harms from the AI system's use, not just potential risks, and thus it is classified as an AI Incident rather than a hazard or complementary information.[AI generated]