oh great, just what we need; sarcastic robots

Computers that can tell when someone isn't being genuine or mocking them, that's exactly what we wanted...

Well, actually software which can detect sarcasm with impressive precision based on semantic cues if we wanted to be exact. It learned how to tell when someone is being sarcastic by tapping into the source of never ending snark we know as the web, mainly some 66,000 product reviews from Amazon.com in which sarcastic quips were tagged by human reviewers and fed into an algorithm which learned to tell when words were used to imply the opposite of their original intent in random snippets of text, a key indicator of sarcasm. So will it be just a matter of time until we get wisecracking, sarcastic machines helping us around the house? Maybe. It's quite possible that giving machines the ability to understand normal human snark will help robot designers to make their machines work better with their owners and lead to fewer misunderstandings and mistakes in the real world, where household and medical robotic aides will be expected to function in the coming decades.

Now, while the system is uncannily accurate with written text and seems to perform roughly on par with many humans, it would face some interesting challenges should it be enhanced to deal with such things as actual speech and body language. Listening to humans and looking for the variations in intonation would require an almost science-fiction like ability to parse through human speech patterns. Every person sounds different and we're used to these subtle differences so much, we hardly notice them when conversing with someone who's fluent in the same languages as we are. But to a computer even the tiny differences between two people from the same part of the world are very obvious. Likewise, you when it comes to our body language, which actually makes up much of human communication would be even more complicated due to the sheer range of all our possible expressions, postures and gestures. Combining the two is an even steeper uphill climb.

Nevertheless, just being able to teach a machine how to pick up on something as vague as sarcasm is a very interesting and notable accomplishment, and if how today's AI systems work can give us any indication of the project's potential future, this system could become even better at flagging snarky quips with more comments from social media networks, news sites, and blogs from which to learn. And maybe, when it's precise enough and paired with a good enough speech recognition component, the system wouldn't need to learn very much about human tone and body language in order to determine what a human really thinks when he says "great job there buddy" with a good enough accuracy to be useful in everyday environments. But to really understand the subtle ins and outs of human communication, we'd need to turn to the realm of biology and figure out how we're able to do it since birth and improve on it throughout our lifetimes…