Researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) have devised a way to train neural networks so that they provide not only predictions and classifications but rationales for their decisions.

Illustration: Christine Daniloff/MIT