AI in the News — #25
Chatbots: common sense vs. Turing (Techxplore)
The Winograd Schema Challenge is a contest to see who can build the best chatbot. It’s not like the Turing test, which is about fooling a human into mistaking a computer for a fellow human. Instead, it’s about designing a chatbot that can understand ambiguous sentences, i.e. that has some common sense. So far, they don’t much.
Conversations in the Uncanny Valley (Phys.org)
A voice-interfaced digital assistant like Siri would seem to be all about acting human. But it turns out that, if such a voice AI acts too human, and yet not perfectly human (because it can’t do that), it seems creepy. This is the conversational equivalent of the Uncanny Valley whereby the physical appearance of a robot creeps you out if it’s very human but still not human enough. Turns out, if the voice AI acts a little less human, a little less competent, and explains its limitations as it goes along, it becomes more acceptable.
Artificial floorwalker (Phys.org)
The Macy’s department store is now offering an app that tells you what’s in stock in the particular store you’re in.
Tesla tweaks (Phys.org)
As you may have heard, there was a fatal accident with Tesla’s experimental self-driving car, when a semi turned left in front of it and neither the driver nor the autopilot reacted in time. Apparently, the autopilot failed to see the truck, which was light-colored against a light sky. The car had radar, but didn’t pay a lot of attention to it, so to speak, since it would have registered too many things as obstacles. They are now revising the radar system…