Google and others building “self-doubting” AI systems

Jan 13, 2018, 5:57 AM EST
(Source: Many Wonderful Artists/flickr)
(Source: Many Wonderful Artists/flickr)

Uncertainty, a key aspect of human reasoning and intelligence, is set to be a part of two of the most popular deep-learning frameworks as researchers from Google and Uber train their AI programs to handle probability and measure their self-doubt.

“Self-doubting” AI systems, for example in autonomous vehicles, make great sense since they impart a machine the power to realize its limitations rather than just answer critical situations with just a yes or no, writes MIT Technology Review.

Recently, Uber also unveiled its new programming language, named Pyro, which combines deep learning and probabilistic programming, an approach that can make AI smarter.