Artificial intelligence will become smarter by doubting itself 11.01.2018
Artificial intelligence will become smarter by doubting itself

Experts from Uber and Google are developing modifications of the most popular deep learning platforms. The main goal is for AI to be able to assess the degree of uncertainty about its own decision.

To date, deep learning has one significant drawback – the need for large amounts of data and computing power. In addition, the decision-making process itself is rather ambiguous: for example, AI makes all of them without any doubt about their correctness.

To improve the process, it is necessary to make artificial intelligence doubt when making a decision. This approach will be useful in complex situations involving autonomous machines.

Google developer Dustin Tran believes that users will like the system reporting the level of uncertainty. He says that if a pilotless vehicle does not understand how confident it is in its actions, it can make a serious mistake. AI having uncertainty assessment will make neural networks much smarter.

To integrate uncertainty, Uber develops a programming language Pyro, and Columbia University – Edward. They represent two competing AI development schools: one specializes in neural networks, another – in theory of probability.


Subscribe to the news of Artificial Intelligence Conference in Facebook and VK!

AI-industry news and cases for you