Sharon Li, 32, is a pioneering AI security expert at the University of Wisconsin, Madison. Her innovative work in out-of-distribution (OOD) detection has prompted AI gurus to rethink their approach to education. She created one of the first algorithms to help AI models recognize unknown data in the real world so they don’t fail in surprising and catastrophic ways, as has happened.
Last year, Li was selected as an outstanding paper by NeurIPS, one of the most prestigious AI conferences, for her analytical elaboration of out-of-distribution detection. In addition, she was also named one of the winners of MIT Technology Review’s Innovators Under 35 competition. As part of the first generation of AI capable of collecting and analyzing massive amounts of data very quickly, she is a critical player when it comes to building trust in AI.
Li’s approach takes advantage of uncertainty by training AI to recognize irregular data it might encounter. For example, an AI autopilot might avoid an accident if it detects something coming up on the road that its training data didn’t know about and needs to be avoided. Or a chatbot could honestly “not answer” a question it hasn’t been trained to answer. Thanks to LI’s work, we can dream of a better future for AI systems that are safer, more reliable, and can help us solve real-world problems.
Unfortunately, most of today’s AI models are designed to recognize only certain things and often fail when they encounter unknown or unpredictable situations. The inability of AI to reliably understand what it “knows” and what it does not “know” is a serious weakness and is the cause of many AI disasters.
The AI community therefore needs to rethink its approach to training to enable more safety-oriented machine learning. Artificial intelligence should be enabled to recognize the unknown to avoid AI systems failing in surprising and catastrophic ways. Sharon Li has once again taken the initiative to challenge the AI community to develop safe AI models.
As Innovator of the Year, Li is working to create scenarios where AI can help us solve problems in the world instead of causing mischief. Her dedication and achievements in this field speak for themselves. We have borne witness to how AI systems translate into the real world and how we are prepared if they fail. But with Sharon Li at our side, we can transition to a safer future for AI technology.
