AI That Can Learn Cause-and-Effect: These Neural Networks Know What They’re Doing

     Neural networks can learn to solve all sorts of problems, from identifying cats in photographs to steering a self-driving car. But whether these powerful, pattern-recognizing algorithms actually understand the tasks they are performing remains an open question.
     For example, a neural network tasked with keeping a self-driving car in its lane might learn to do so by watching the bushes at the side of the road, rather than learning to detect the lanes and focus on the road’s horizon.
     Researchers at MIT have now shown that a certain type of neural network is able to learn the true cause-and-effect structure of the navigation task it is being trained to perform. Because these networks can understand the task directly from visual data, they should be more effective than other neural networks when navigating in a complex environment, like a location with dense trees or rapidly changing weather conditions.
     In the future, this work could improve the reliability and trustworthiness of machine learning agents that are performing high-stakes tasks, like driving an autonomous vehicle on a busy highway.

Oct 23th, 2021