The Massachusetts Institute of Technology has taught a computer system to play a text-based game, even if it has no prior knowledge of how knowledge works. It was important for the authors of the study that the system wouldn't base it answers on syntax, which is already a long problem faced by natural-language processing, so they created it in such a way "that just relied on collections of keywords as a guide of action." The researchers also used the deep learning approach, which helped the system organize in layers.
One of the authors of the paper, Karthik Narasimhan, expresses that the system that they have created is "adversarial," and believes that it can contribute greatly in test language understanding. The learnings could also be applied in bigger scales and more complex domains.
Share This Article