Sarcasm, according to Google is “ the use of irony to mock or convey contempt”. Interpreting it could be final frontier for AI.
Understanding Sarcastic line is a complex process. Depends on lots of variables — context, body language, tone modulation and so on. To be fair there is not proper/scientific way to decode sarcasm. To add to it, degree of closeness is (kinda) proportional to complexity of sarcasm.
In one of the scene Deadpool 2 movie, the lead actor was questioned who he is. Responding, he says, “I’m Batman”. The whole theater erupts in laughter. You can understand this if you’re aware of Batman and his punchlines.
However, if you’re using an AI program to understand this situation, how will it decipher the line? It may conclude DP is impersonating Batman or DP is confused and so saying he is Batman. It’s not completely impossible. You can teach about Marvel — DC rivalry, then DP’s awareness on fourth wall. Batman and his punchlines. Then may be AI can read between the lines.
To be fair, DP’s sarcasm is not even closer to Chandler Bing or Sheldon Cooper. To decode Chandler, you should train the AI on relationships between genders, the history of 5 other friends (and their individual characters) with whom Chandler converses. And for Sheldon, its harder still. You have to teach Physics-based concepts, so AI can understand Physics-based jokes and geekyness should be taught to it. Honestly, I don’t get Sheldon in many of the instances.
You can think DP, Chandler & Sheldon are fictional, and so no need to worry on that. But in reality, the sarcasm is much more intense. During lunch my friend questions how many bandages I had for the work session. We neither work in medical field. What he really asks is number of tasks for the day. Even a co-worker who hears the question can’t decode it on first try.
Think about Google Duplex. Say, we ask Duplex to book a table for lunch and it calls the restaurant at 10 AM on February 14th. For its request the person in the desk, who was busy attending the phone calls replies, “Yea, you wish”, as the restaurant tables were reserved days earlier. How would it respond?
If it responds, “Yea, I understand its too late for reservation. But I would appreciate if you get a table”, or something like that to convince the desk, then its time to run away from computers and hide under a rock. Because, understanding mockery/sarcasm could be the final frontier for AI.
PS: This was originally published in medium, on June 3, 2018.
Page source
Page last updated on: 2024-11-06 09:30:05 +0530 +0530Git commit: a98b4d9