Subtle emotions are a challenge to recognize, sometimes even for humans. MIT researchers originally aimed to develop a artificial intelligence (AI) system capable of detecting racist posts on Twitter. They soon realized that the meaning of many messages couldn’t be interpreted without guiding the system to understand sarcasm. But how can you teach an algorithm to identify sarcasm? The researchers used deep learning to help the algorithm recognize subtle patterns using a large amount of data. Fortunately, many tweets already use emoji as a kind of labeling system for emotional content. Adding the emoji to help the system read tweets for emotions in general, propelled the researchers’ efforts forward toward teaching it to recognize sarcasm. Now the system can analyze tweets to detect sarcasm, and emotional subtext in general, better than most people! The impact will be huge. Bravo MIT.