Carnegie Mellon University
February 27, 2018

Datta stresses internal processes of AI

Anupam Datta was featured in a story in The Economist discussing the push to understand why artificial intelligence (AI) agents make the decisions they do. Once deep learning neural networks are trained, it’s difficult to understand exactly how they do what they do. The fear, the article states, isn’t that AI won’t do what it’s told, but that it will do it in a way that’s incomprehensible. While a number of researchers are attempting to crack the “black box” of internal AI processes, Datta is focusing on stress-testing the outputs of trained AI systems. He feeds the systems input data and then examines output data for undesirable outcomes. According to the article, Datta’s approach “lets those who make and operate AI ensure they are basing decisions on the right inputs, and not harmful spurious correlations.” Read the article.