AI NewsEthics in AI

Uber and Google Building AI Systems That Actually Doubt Themselves

Researchers at Google and Uber are working on modifications to deep-learning frameworks that will allow their algorithms to handle probability. This will provide a way for smart AI systems to measure their own confidence in a prediction to understand whether they should doubt their response or not.

Deep learning processes through neural networks have been a huge success in the past few years, providing machines with a new level of intelligence that allows them to carry out tasks from translation to image recognition. The downside to the approach is that it requires a lot of computing power and large training datasets.

Discussing the latest innovation, Dustin Tran – a Google researcher working on this new AI approach – explains that benefits and its impact on self-driving cars:

“You would like a system that gives you a measure of how certain it is. If a self-driving car doesn’t know its level of uncertainty, it can make a fatal error, and that can be catastrophic.”

The research reflects a growing realisation that uncertainty and measuring probability is a key aspect of human reasoning. Adding this capability to an AI system could make them smarter and less prone to errors – which is going to be even more crucial when AI is used in ever more critical scenarios.

Uber develops new programming language

AI researcher and chief scientist at Uber, Zoubin Ghahramani, explains that ‘we want to have a rock-solid framework for deep learning, but make it easier for people to represent uncertainty.’ The framework Ghahramani is referring to is Pyro, a new programming language released by Uber that merges probalistic programming with deep learning.

While regular deep learning systems learn based on data that is being piped in, Pyro can also be used to build a system that is preprogrammed with knowledge. The impact of this is dramatic and could cut back costs associated with AI dramatically.

Uber aren’t alone though, Edward, developed at Columbia University, is another programming language that embraces uncertainty. Both Pyro and Edward are still at early stages of development, but it isn’t hard to see why the likes of Google and Uber are getting excited by these new advancements.

Uber uses machine learning and AI techniques in a number of business processes including surge pricing, driver routing and of course their self-driving cars. The company has ploughed millions into research and have hired some of brightest minds in the industry to help them come up with solutions. Google on the other hand has rebuilt their entire business around deep learning and AI as of late.


Richard Young

Richard has been interested in the AI space for some years. Questions of ethics can raise some serious problems. What if AI can learn who is pre-disposed to cancer and then not give them health insurance?

Related Articles