Wonderfest - Can AI Know What It Doesn't Know?

With increasing frequency, AI algorithms are making high-impact decisions: When should a self-driving car slam on the brakes? Can an MRI scan reliably detect a tumor? Will facial recognition software identify your son as a Most Wanted fugitive? AI algorithms need to be aware of their confidence level - to "know what they don't know" - in order to be reliable and safe. Fortunately, time-tested ideas in statistics are providing solutions. How are old and fundamental mathematical concepts blending with recent tech breakthroughs to create safe, uncertainty-aware AI?

Speaker: Stephen Bates, UC Berkeley