I have been watching Watson playing Jeopardy! It is really interesting. I am impressed by Watson’s performance and ability to analyse questions in natural language (although sometimes it seems like Watson’s performance is at an advantage due to reaction times).
But what really amazes me is the “interaction between” Watson and the audience (literally: the audience’s reaction to Watson). In particular, Watson “made a joke:”
[Watson gets a Daily Double]
Alex: what are you going to wager?
Watson: $6,435
The audience laughs.
Watson got the audience to laugh. Alex then quipped: “I’m not even going to ask.” But the audience laughed before the quip. The audience was laughing at what Watson said, not at Alex’s quip.
I find that really interesting because humour seems like a very human interaction (none of my electronics has every made me laugh, though I have laughed at the human generated content they display. In contrast, I have yelled at my electronics). Now, it seems like part of what is funny here is that Watson gave a very inhuman response. But what does it mean that the audience was shocked by this? What made it funny? Was the audience interpreting the answers as human-like, but then suddenly with this answer Watson fell into the uncanny valley?
It also made me think about my own experience as a human who can more-or-less parse natural language. I have always found humour to be one of the most difficult things to pick-up in a foreign language (I don’t really know why). I wonder whether the next step in the development of AI will involve the ability to interpret and/or produce humour.