The Widcombe Wrobotiers understand AI (robot memory & emotions)

Just a quick blog before I forget something cool that happened this week.  I was approached and then abandoned by a group of 11 year old roboticists a few weeks ago.  (The top picture is from last year's group -- I didn't take any pictures of their faces myself in case I needed special permission for that.) They decided they'd rather visit the lab of my colleague Pejman Iravani than have me visit their school, which is fair enough.  But once the surprisingly-large number of people they'd initially contacted worked out where they'd gone, a number of us turned up to hear Pejman's presentation and see the kids.

Pejman happened to ask at one point "can robots learn?" And the children answered a resounding "YES!"  They clearly knew the answer to this one from somewhere, and were proud to know it.  One of the other mechanical engineering faculty seemed perturbed by this, and asked "Can all robots just learn -- just like that?  Or is there something you have to do?"  And the children answered "No, you have to… um…" they clearly had the idea though they had trouble expressing it, but pretty soon one said "You have to put the place in the robot where the learning is going to go."  This satisfied the mechanical engineers, and I absolutely loved it as an answer.  (Technically, in AI we call this place where learning goes "variable state", that's the word the students were missing.)

Next,  one of the PhD students in my group, Jekaterina Nabikova, talked about her current research.  This is on whether implementing emotions in a robot makes it easier for humans to work with the robot if they don't know what it's programmed to do.  The students (and mechanical engineers) didn't seem to be sure what to think, and someone (maybe their teacher?) asked whether robots could have emotions.  They seemed sceptical at first, which pleased the rest of the adults. But then I said, "Wait a minute! You gave a great answer about robots and learning a little while ago.  You knew that you had to put something inside them for them to learn.  That's what Jekaterina is calling 'state', but for learning, people often call it 'memory'. You know about that, right?" [nods.]  "So do you think its possible to put something in a robot, some kind of state, that would be as much like emotions as what the memory you put in is like learning?"  [Thoughtful faces followed by wide grins...] bursting into conversation.  Including one of the wrobotiers suggesting (and then many discussing) whether the battery's state could be an emotion, since it varies between charged and uncharged. They hadn't thought that battery state might be a simple kind of "hungry," but they liked that idea too when I suggested it.  Awesome!

---

Here's a list of my papers and software for building AI emotions.