Singing, dancing robots range from the charming to the inept to the creepy to the inexplicable. They’re all fun to watch. But, I’ve always wondered: what’s the fascination? Why do humans want robots to dance? And what are we trying to learn from building them?
Drexel University Engineering professor (and MIT Media Lab alum) Youngmoo Kim and graduate students Alyssa Batula and David Grunberg have been working on a dancing, piano-playing robot of their own. I called up Dr. Kim to ask him why. “It’s fun to do,” he said, “there’s a very tangible result when you get a robot to play a piano piece or to dance. But my long-term research is to look at these questions of creative expression that we find difficult to analyze.”
Dr. Kim’s “Hubo” robot does something that few other robots can do — listen to the music and coordinate moves based on the waveforms it hears. (Most dancing robots are pre-programmed and synced up with a music track.) Here’s “Hubo” adjusting to changes in tempo:
Dr. Kim hopes that this kind of research will eventually lead to a robot capable of performing in an ensemble, alongside humans – reacting to changes in the music and adapting as the piece unfolds.
But I wondered, do we lose something when we try to study and systematize ephemeral art forms like dance or music? When a robot creates art, does that art lose some of its meaning? Dr. Kim says no: “there’s so much we don’t understand about music. I don’t ascribe to the notion that understanding more about it is somehow going to dehumanize it.”
Dr. Kim graciously answered many such Luddite questions about robots and art a few weeks ago. The abbreviated transcript of our phone conversation follows.
Why make a humanoid robot performer? If all you want the robot to do is play the piano, doesn’t having a human form make it harder?
Sure. It absolutely does make it harder. There are specialized robots that play the piano very well, in fact. But humanoids actually teach us a lot about humans. Our broad interest is using robots to try to study and understand things we have difficulty quantifying such as musical expression. Like, what goes into a piano performance that’s beautiful and lyrical and expressive – something you would see on stage at a great performance venue – versus something that’s mechanical? The same notes are being played, but one sounds boring and the other one is beautiful and moving. We like to use humanoids to study that, because we’re relating these to human performers.
With a humanoid robot, you can ask the robot to repeat things exactly. You can do experiments. Things that are difficult to do with a human performer. Even with the best performers in the world, you can’t say, “OK play that exactly the same way.” You can’t ask the performer to play that again exactly, only use .2 newtons more vertical force. With a robot you can. Having a robot that’s humanoid in shape and teaching that to mimic or perform gestures as a human does, I think we start to understand what’s actually happening when a human performs. How does a human go from just playing the right notes at the right time, to something moving and beautiful?
Have you ever seen a robot give a performance that was moving in any way?
Not yet.
Do you think that’s even possible?
Absolutely. Are we there yet? No. We’re quite far from that. We’re just at the tip of the iceberg. Because we’re at the stage of robots where we were with computers 50 years ago. Big mainframes and only a few people have access to them. As robots become more available, as the costs come down, they’ll do more and more of these tasks that we think are solely in the realm of humanity such as arts and performance.
Even if a robot got really good at playing music that was technically beautiful, how could it be moving? How could it ever communicate an emotion?
That’s the beauty of the experiment. When we see a performance, are we actually feeling what the performer is feeling? This is a way of removing one variable from that: the performer. In this case, the robot’s not feeling anything. It’s just being programmed to perform in a particular way. If that results in having a moving performance to the audience, then we’ve learned something about what goes into music.
If a robot ever got really awesome at producing any kind of art, wouldn’t it take away some of the mystique?
I think about it from a very different perspective. I think about it in terms of expanding the creative palette. If we know something more about how performances are generated, how certain creative ideas or concepts take form, I think that actually improves our ability to create more expressive forms of art. Then the robot becomes a tool for creation. I don’t ascribe to this dichotomy that once the robot can do human tasks, somehow we’ve dehumanized those tasks. I certainly don’t agree with that. I think in understanding those tasks – particularly if they’re artistic tasks, or things we find difficult to understand right now – then we’ve only opened up the possibilities.
I’m a musician. Do I love music? Yes. Do I think my most spiritual experiences have come through music? Probably. When you’re performing or experiencing a fantastic performance, the feeling you get is like nothing else. But to say we shouldn’t even try to understand that, I can’t go along with that.
You say that through these robots we’re going to learn about great performance or art. But what specifically are we expecting to learn?
It’s hard to talk in specifics at this stage because we’re just scratching the surface. I think in terms of physical gestures of performance, we’ll learn something there. You see great performers perform, and a lot of them gesticulate and gyrate all over the place. They’re doing all this extraneous motion which has no impact on the sound whatsoever. But do those gestures impact the audience and the perception of that sound? That’s something we’d like to look at. Another great interest of mine is understanding emotion. Can we develop better models for how people perceive emotion? How much is it impacted by performers’ gestures and the visuals and how much is it just the audio?
Will a robot ever be a great performer, or will it always be a novelty?
The analogy I’ll draw is with IBM’s Watson. The best Jeopardy! player in the world could be a robot. I guess it was a sacred realm until 1997 when Deep Blue beat Garry Kasparov. Will the best performer, pianist, violinist, or singer in the world ever be a robot? I guess I should say never say never. For a long time we thought chess required human intuition. No, it doesn’t. We thought beating Jeopardy! required uniquely human skills of being able to draw analogies and understand puns. No, it doesn’t. So I would be naïve to say it will never happen.
Check out more of Dr. Kim’s work at Drexel University’s Music and Entertainment Technology Lab here.
While he works on getting robots to dance like people, we’ve still got a lot of work to do getting people to dance like robots. Get to practicing:
