Toyota Unveils Violin-Playing Robot

eldavojohn writes "Toyota has unveiled a robot that can play the violin. From the article: 'Toyota said it planned to further advance the robot's dexterity to enable it to use tools and assist with domestic duties and nursing and medical care. The robot has 17 joints in both of its hands and arms now.' It seems there have been small — or maybe even strange, impractical — advances in robotics repeatedly with demonstrations of robots performing a specialized task. Are we merely struggling to hard code each human activity as we strive for an all purpose android? Is there a chance artificial intelligence & robotics will ever become generalized enough to make interaction interesting?"

Comments

Very cool, but

[log1385]

Robots will never be be able to match the musical abilities of some humans. There are too many tonal subtleties involved, especially on the violin.

That is still very impressive, nonetheless.

Re:Very cool, but

[OzRoy]

I think they probably built it to be their new customer complaints manager.

So now when people call up to complain the robot can play a tiny violin in mock sympathy.

Re:Very cool, but

[Mex]

Oh, I'm pretty sure eventually they'll figure it out. It's all just chemical reactions, man. And there's already enough music theory out there. So the rules are already (sort of) written. There's been experiments with music-making robots since the 50's. Not huge progress yet, but it will happen.

I do believe, eventually, "creativity" will be programmable.

Re:Very cool, but

[QuantumG]

As I write this there are 3 replies to your comment modded up but none of them seem to mention that obvious (I hope) fact that programming this robot to play the violin has absolutely nothing to do with matching the musical abilities of humans. Toyota are not demonstrating a product here. They're not saying "in stores next fall: robots that can play violins". They don't think there's a huge market for violin playing robots out there that is just waiting to be tapped.

The point of this demonstration is to show that their robot research has reached a point where they have built a robot with joints that have sufficient degrees of freedom and controllable accuracy that they can do this kind of stunt. You're supposed to look at a robot playing the violin and say "well, if it can play the violin then it can hold a power drill or other tools!"

I'd suggest that maybe they should program the robot to put together some of the crappy furniture you get from Ikea but then people will claim it wasn't cost effective to use a billion dollar robot to do the work of a home handy man or something.

Re:Very cool, but

[eh2o]

Not true; there are many "expression" marks put into the score, the interpretation of which requires the player/conductor to be familiar with appropriate idiomatic interpretations of the time period. In fact this has always existed, but hasn't always been explicitly written. e.g. baroque era pieces don't have any expressive indicators at all, but *did* have specific interpretations as they typically went with specific dances requiring particular tempos and so on.

Furthermore much of the work that a musician does involves subtle modifications to the rendition that enhance the clarity of the structure. e.g. microtiming deviations in the melody and subtle tonal inflections are a major part of what makes multiple voices traceable to your perception in polyphonic music -- with strict timing it is *much* more difficult to hear out any polyphonic structure. These effects can be measured quantitatively, by the way, but are far too complex to notate in any score intended for humans to read, and for the most part are too complex for an experience musician to be fully conscious of. (It is possible, however, to program a computer to reproduce them using machine learning).

Finally there is in fact an emotional aspect of music that is actually a consequence of some neural structure or other brain process. This is an active research topic. Likely it is a type of synaesthesia (e.g. mint flavor -> cool sensation). In other words the emotional response to music isn't just the performer "making it up", it's sort of short-circuit in the brain of the listener. Since this is basically a universal effect among humans it would be silly to think that the composer wasn't aware of it as well.

Robotic vs. Human ability

[AugustZephyr]

I suppose this raises another question regarding the increasingly human attributes of robots. Is something that is "handmade" or "handplayed" by a robot any more or less valuable than its human equivalent? For instance: it may be very impressive that a robot can play pomp and circumstance, but once this becomes more commonplace (as strange it may seem now), does it have more/less value than a human being able to reproduce the same sequence of notes?

Re:Robotic vs. Human ability

[DoubleRing]

Well, it isn't exactly the same thing, but MIDI with good synths has been used to "perfectly" play a composition. Especially for instruments with simple timbres like drums and piano, a synth can sound very close to the real thing. Even with these ways to create a "perfect" performance, people still find a human performance impressive. It's kind of like meeting a person who can compute logarithms in their head, or find all the factors of a number without a calculator. Sure, a computer can do it--it may even do it faster, but I don't think that the fact that a machine can do something makes the feat less impressive when accomplished by a human. I'm still impressed when someone can run a marathon, even if a machine could do it just as easily, if not better. I don't see why there is so much concern that robotic performances will cheapen the value of human performances. Besides, if a programmer is able to write a program that is able to take a piece of music and interpret it beautifully, then it is still a human achievement in that a programmer was brilliant enough to decipher not only decipher the subtle psychology of what makes one performance sound emotional and powerful and the another sound mechanical, but also codify an algorithm that would imitate that interpretation. Kind of like designing a conversation bot to beat the Turing test. It's still a human accomplishment. Remember, the machine is us/ing us.

Pomp and circumstance

[Cristofori42]

For the love of all that is holy why did they teach it to play that song?! I've been spending all of my years since high school band trying to erase that song from memory after playing it over and over and over for hours on end.

general purpose

[phantomfive]

Are we merely struggling to hard code each human activity as we strive for an all purpose android? Yes, playing the violin is a hard coded activity, but the important advance here is the new dexterity of this robot. It isn't so much an advance in artificial intelligence as it is an advance in mechanics.

Is there a chance artificial intelligence & robotics will ever become generalized enough to make interaction interesting? Absolutely. :) I'll do it myself if no one else gets to it first.

It's over man; it's over!

[explosivejared]

They can stand back up after being kicked and now can play the violin. Anyone sane could obviously see that this completes their skill set. They'll use the sweet sounding music to lull us all to sleep, and then with their new found balance and agility put the kibosh on us all. I can feel their cold, icy hands around my throat just now! It's over man; it's over!

Re:Robot's sense of time..

[ubergamer1337]

Considering that I'm just wrapping up a semester of violin methods for a music ed degree, I find this achievement more impressive then building a robot to play any other instrument that I can think of because the violin requires extremely precise movements and pressure. The strings take a fair amount to force to depress, but the instrument itself is rather fragile. Also, to get an even sound out of it, the bow pressure has to constantly and smoothly changed while moving.

yeah, but...

[veganboyjosh]

will it stay standing if i kick it?

Re:yeah, but...

[jamesh]

A soft drink machine was invented that would complain if people kicked it or tried to tilt it. So it got kicked and tilted more than any other.

Some cars, when parked, ask people to move away if they get too close, so people deliberately get close and try and taunt it.

A new digital media format is released, with a claim to being uncrackable, so it gets cracked very quickly.

So logically, what happens when a robot gets invented that's sole claim to fame is that it won't fall over, even if kicked?

And now we find that even a robot who's sole purpose is to play the violin is going to get kicked too, just to see what happens...

I think i'll invent a line of robots who's sole purpose is to whack you over the head with a cardboard tube if you kick them or other robots over, or just generally abuse technology for your own amusement. Then i'll release version 2 which features a crowbar instead of a cardboard tube. I'll make a fortune selling them as guards for kick-overable robots, vending machines, cars, and DVD's.

"small" advances

[loonicks]

It seems there have been small -- or maybe even strange, impractical -- advances in robotics

Welcome to the world of research. It takes a lot of work to make small advances like this one. The point of research is to solve specific, difficult problems. I'm willing to bet there were other reasons for this project.

Bold research on comprehensive AI has halted

[presidenteloco]

it seems, with perhaps Marvin Minsky as an exception, but we need a new guard.

Everything is understanding the nth degree of optimizing Bayesian network inference,
usually applied to a very specific toy problem.

Nothing wrong with that research. Not really knocking it.

But where is the research on how a generally intelligent system could choose what to
focus its inference-engine attention on. Where is the meta-logic about prioritization
and pruning of "trains of thought" depending on success of search and progress
and urgency of need to know compared to other concurrent topics.
Where are the systems that can posit and explore multiple incrementally variant theories
of some aspect of the world, and figure out which theory-variant is a better model of
past and present observations. Where is the system that can take in lots of different
peoples' writings or sayings about things and synthesize an ontology and figure out
whose beliefs are the most promising (truthwise) and relevant.
Where is the episodic memory?
Where is the emotion-tagging of experiences and important generalizations,
and the emotion-guided prioritized recall?
Where are the short-term memory blackboards?
Where is the "utterance" theory and theories for how to inform and motivate
other intelligent agents into execution of a cooperative plan.
Where is the AI just for the sheer wonder of trying to put several techniques all
together and see what emerges?