Building Trustworthy AI Even When We Don’t Understand It

Artificial intelligence (AI) systems are increasingly opaque, including to their creators. For example, when language systems, in particular large language models, generate claims that are easily verified by people to be untrue, trust is irrelevant. What matters is when they generate claims that are not (easily) verified to be true. I will discuss how people use and trust systems in such contexts, and how the systems can be improved to increase their trustworthiness. I'll also discuss various initiatives on campus at UMD related to trustworthy AI, such as the Institute for Trustworthy AI in Law Society (TRAILS) and the new Artificial Intelligence Interdisciplinary Institute at Maryland (AIM).