The Data Science and AI Academy and the Laboratory for Analytic Sciences will co-host an AI Hot Topics session.

In many domains, including writing software and translating text, AI can now produce outputs on par with or better than those of a competent human performer. But like a skilled human, AI sometimes makes mistakes — and for some critical systems, a mistake can have disastrous consequences.

While we have decades of experience incorporating unreliable machine components into robust critical systems (a sensor failure typically doesn't lead to an airplane falling out of the sky), and even more experience crafting processes based on unreliable human input (a single mistake by a healthcare provider is often caught and corrected before causing patient harm), generative AI outputs are something new: While law firms have always hired newly minted lawyers, only recently has AI become capable of writing a legal brief like a law school grad. Do review and assurance processes developed for fallible humans map unchanged to fallible AI?

Jascha Swisher, technical staff at the Laboratory for Analytic Sciences, will moderate the discussion. The panelists will be:

• Mike Green, an AI engineering researcher at the Laboratory for Analytic Sciences
• Michelle Rodenburg, a law and technology librarian at the UNC School of Law and the UNC Law Library
• Chris Provan, an AI evaluation and monitoring lead at Vega Health