I read the Keepon and Shadowplay articles first, and then I read the article about Robovie. And I found, while reading that third article, that I kept thinking about a quote from one of my classmates, talking about his experience interning at the interaction design consulting firm Adaptive Path. Many companies have big plans for all the kinds of things they want their software to do, but not the time or resources to successfully do them for the first version they release. The metaphor they use at Adaptive Path is that of making a cake: the company wants to make a whole cake, but they can’t the first time around. So what most companies do is they make cake batter and give it to the users and say “we’ll bake it for you later.” But what they should be doing is trying to make cupcakes - identifying a smaller-scale version of that final goal that also offers a complete and useful experience. That metaphor really struck me as a great way to explain what a lot of companies get wrong in early versions of software, and I think it is applicable to these robots, too.

Keepon and the shadow play robot seem like cupcakes to me - the designers knew they could not create an entire human-like robot that would successfully interact socially in all the ways humans can, and in the case of Keepon, in fact they explicitly did not want a robot with that much complexity, for therapeutic purposes. So in both articles, the authors identified particular small pieces of interaction - for Keepon, it was eye contact, joint attention, simple emotional expression and later rhythmic motion, for the other it was imitative shadow play. And then they each built robots that could do just enough to successfully elicit those experiences with interaction partners. They created cupcakes.

But Robovie seemed much more complex for much less compelling reasons. The authors talk about how the robot’s behaviors were intended to shape specific kinds of responses from children and provide them with a framework for interpreting the context of its actions, but not why chose to give it such a wide range of behaviors or why, beyond a few cursory examples, the particular behaviors they programmed were chosen. My impression was sort of that they wanted the robot to be able to do lots of stuff because they thought the way to simulate human interaction was to make a robot that can do as many human-like things as possible. Knowing all the ingredients for a whole cake, they put together cake batter even though they didn’t have the capability to bake it.

And I think that shows in how the children interacted with the robots. True, many children liked Robovie and continued to interact with it over a period of several weeks, but even unbaked cake batter is tasty in small amounts. Other children got bored with the repetitive actions of the robot, and didn’t play with it once the novelty value wore off. Keepon, on the other hand, has the capability to do far less than Robovie, but captured children’s attention for far longer periods of time. True, they were younger children, and the ones in the longer study had social deficits that may have made Keepon’s simple interactions more compelling than for the average neurotypical child. But I’m a lot older than all those children, and I’m barely interested in Robovie, whereas I don’t even get bored with watching Keepon’s youtube videos multiple times since 2007. Sure, I lick the spoon when I’m making a cake, but I never just stop once I have the batter. What I really want is that fully baked cake experience, even if its just a cupcake.

And now for your enjoyment, Keepon: the best little dancing robot.