We can’t really teach robots to love until we can teach them to learn, can we?
While watching Hod Lipson’s TED talk, I couldn’t help but wonder how exactly the learning mechanism in their AI works. The showcased robot was able to test the waters and figure out how it was going to move itself across a surface, but Lipson also threw around concepts such as “natural selection” like they were programmable tools, without any reference to how these concepts were technically implemented.
The point I’m trying to make is that many artificial intelligences seem to learn in different ways.
Take Cleverbot, for example. This website allows you to enter a chat with a supposedly artificial intelligence, which will respond to your questions and even ask questions of its own. But Cleverbot’s learning mechanism is simply regurgitation. It “learns” common responses to certain inputs from the millions of people that use it, and spits them back out when prompted. Sometimes this makes for some interesting changes in direction for the conversation.
My question is this: in what sense does Cleverbot really learn? What constitutes learning, anyway? Are Hod Lipson’s robots any better at learning? And will we ever produce robots that can learn the same way humans do?
That last one I seriously doubt, but feel free to disagree.