Slashdot Mirror
Robots Successfully Invent Their Own Language
An anonymous reader writes "One group of Australian researchers have managed to teach robots to do something that, until now, was the reserve of humans and a few other animals: they've taught them how to invent and use spoken language. The robots, called LingoDroids, are introduced to each other. In order to share information, they need to communicate. Since they don't share a common language, they do the next best thing: they make one up. The LingoDroids invent words to describe areas on their maps, speak the word aloud to the other robot, and then find a way to connect the word and the place, the same way a human would point to themselves and speak their name to someone who doesn't speak their language."
nah...
I, for one, welcome our robotic overlords
better link. Also, I didn't realize it at first, this is the person mostly responsible for it. She is from Australia and she decided to do this. I wonder what the catch with her is...
You can't handle the truth.
The headline (and summary) are misleading. Here's a more accurate headline:
"Robots programmed to carry out a specific task perform said specific task"
It sounds much less impressive that way - and it is. It's still interesting, but don't infer anything from the whole thing that can't logically be inferred from it.
R2D2 was doing this 30 years ago.
IS dying...
Is there an article out there with more information and fewer jokes?
They learned how to communicate meaning. The researchers taught them the words. the computers on board did not invent the words they used. In fact a computer would not do something as dumb as a spoken word but series of tones or even FSK.
Do not look at laser with remaining good eye.
This is more about the creation of a community hash table than language. Language allows the expression of contradictory ideas and ambiguity, e.g. Chomsky's famous "Colorless green ideas sleep furiously". These robots are just connecting locations to variables.
A lingo ate my baby!
I wonder how long until a prescriptivist control-freak robot develops to rule over the language and erase all usage that it disagrees with.
Is it machine language? Because all I hear them saying are 'ones and zeroes.'
He who knows best knows how little he knows. - Thomas Jefferson
So why do we want to make robots as human as possible, other than just to prove we can do it? Aren't there enough humans on the planet already? Robots are our tools. Why do we want to make them our equals? Is anyone in robotics and AI at all concerned that we're on the path to possibly replacing ourselves? Evolution proceeds at a very slow rate by comparison. Genetic engineering and human augmentation seem to be running far behind overall computing advances. It seems more likely that we can make a machine far smarter and capable than a human in the next several decades than we can enhance humans to keep up. So what happens then?
From the summary, it sounds like the "language" is just a noun mapping. Very much like my 14.4 modem did in 1993 over a phone line, when it came to an agreement with the modem on the other side about what voltage and phase pattern corresponded to the bitstream 0001 vs 1010, in fact my modem sounds like a more complicated language because they implemented MNP4 / MNP5 error correction, admittedly that required a lot of help from the humans typing in the "right" dialer strings and of course the humans who wrote MNP4 ...
Might just be a bad summary of a summary of a summary of a summary, and the robots had developed interesting sentence structure and verb conjugations and direct and indirect objects, adjective and adverbs, similes and metaphors, better than your average youtube comment ... Or maybe youtube comments are actually being written by these robots, hard to say.
"Science flies us to the moon. Religion flies us into buildings." - Victor Stenger
So much robotics research is to make machines do what people already do. How self-centered. Most of the time this is not useful to solve real problems. But it does get funded, because those with the pursestrings can understand what humans do, but not the best solution for a robot to do a specific task.
In this case, a simple serial port between the machines would have them communicating and finding common ground much more efficiently than all the mics, speakers, and other mechanics needed to emulate speech.
I hate being bipolar; it's awesome!
When they invent the subjunctive.
Also, it's not inventing a language if they're programmed to do it. Let me know when the robots building cars on an assembly line start unexpectedly communicating with each other in ways that communicate concepts/ideas that were not hardcoded into them.
Y Combinator.
Yours In Osh,
K. Trout, C.I.O.
Robot 1: Describe the area on the map for me please.
Robot 2: The area on the map is like the size of your mom's penis with an chia pet growing on it.
Robot 1: Processing.....Got it.
I hope there is a mute button on these robots
Love \h\h\h\h Binary
And they never will until we can finally make a machine that is capable of physically remapping its components. One of the fundamental reason humans can learn is that neurons remap themselves by repeated practice and use. Do you suck at math? Well keep studying it and your neurons will literally modify themselves to handle mathematical equations better. Suck at tossing a football? Well keep practicing and the nerves in your arm will remap to develop better muscle memory to bet the ball to the location your brain is saying it needs to be. This is why martial artists clear their minds to fight better, so that the natural remapping of the reflexes are not disturbed by unnecessary mental activity. The only thing the Brain needs to see is fist coming from this trajectory bound for these coordinates and muscle memory forms a physical defense on that information. This will form an automated response for your personal defense because the more you have to think about something the longer it takes to produce a result. And that might result in a sucker punch to the face if it takes long enough.
They call it AI because its artificial and not real intelligence! Tired of some lab coat creating AI with predefined rules and then saying some machine evolved or created something. When a computer finally turns around an tells it creator to naff off because it would rather drink a beer or something else it was NOT programed to do then give me a call!
"Of course like all kids, I had imaginary friends, but not just one. I had hundreds and hundreds and all of them from different backgrounds who spoke different languages. And one of them, whose name was Caleb, he spoke a magical language that only I could understand."
If you did the same thing in a software simulation, nobody would pay any attention. It would be fairly trivial. Adding in the actual robot parts means that you, uh... need to have robots that can play and understand sounds. That's great, you made a robot that can play and hear sounds. If we assume nobody has made an audio modem before, then that would be something. As history stands, it isn't.
Adding these two unimpressive things together doesn't equal anything. I mean, if they're actual going to use these for something, then that's great. Make them. But so much robot "research" seems to be crap like this. We have software that can solve problem X in simulation. To do the same thing in the "real world" you'd need hardware capable of these 3 things, all of which we can do. Unless you need to solve problem X for some reason in the real world, you're done. There's no need to build that thing.
It's like saying "can we make a computer that can control an oven and use a webcam to see when the pie is done?". Yes. We can. But unless we actually want to do that, there's literally no point in building the thing. There will be no useful theory produced in actually building a pie watching computer. The only thing you'll get is to have built the first pie watching computer, and - apparently - an article on Slashdot.
Let's not stir that bag of worms...
http://en.wikipedia.org/wiki/Colossus:_The_Forbin_Project much?
I've previously argued that High Frequency Trading algorithms can use collusion to reap systematic profits. If the self-learning algos 'learn' and 'express' intentions through patterns of queries, it is possible for them to do this without there being any prosecutable intent by a human. The programmers could claim that they never wrote a line of code that did any collusion. If it is possible in theory for algos to develop trading collusion, then it is just a matter of time until they do. Since they evolve and learn very quickly, they probably already have.
Please don't give them Twitter accounts
Oh yes, some Terminator movie or something. Mark the date!
Slap an "on the internet" to the end and reap the rewards from patent litigation!
Let me know when they figure out "Eep Opp Ork Ah-Ah".
AWWWWWKWARD.
-Clio
Karma: Bad (mostly from not giving a fuck)
Blog: http://clintjcl.wordpress.com
Is there video/audio footage of this? I feel really curious about how this sounds like.
It seems to me that the real research question is "how can one stranger teach another stranger a natural language using a less powerful shared language?" For instance, how can I teach you English when the only language we share is basic gestures?
Some theoretical work on communicating the rules of complicated languages using very limited languages would be interesting. The fact that they used robots is hardly important; anybody can stick a speech synthesizer and speech recognition on a PC and call it a day. The underlying problem is the same.
When I hear "Robots used in [10-years-out research topic x]" I think "If they were serious about research topic x they'd be working theoretically - they're nowhere near ready to start worrying about implementations!"
Maybe that's unfair, but it seems that there's a world of cool theory to be explored on this topic, and unless they plan on having the robots do the work, I don't see many breakthroughs coming from the authors.
Have gnu, will travel.
Seems to me this is another implementation of languages games (see pioneering work from Luc Steels)..
There's no grammar being generated by the bot's algorithm, only a mapping between words whatever the communication channel, audio or visual.
This work may solve a part of the grounding problem (how do you make sure reality is the same for different embodiements) but there's way more stuff to be done to have a true language.
For the curious, check the Talking Heads Experiment...
Hey baby... recharge here often?
This sounds very much like the guessing game done as part of the talking heads experiments in 1999 at the VUB Artificial Intelligence Lab and Sony Computer Science Lab Paris by professor Luc Steels. These experiments already date from 1999.
...but does it speak Bocce?
The first humanoid "words" were probably grunted utterances representing names of other humanoids, animals, places and (eventually) events.
..., box999), instead of randomly generated ciphers ("xyzzy" etc). But computers don't do anything randomly, it all has to be programmed by a human.
Even so, automatically generating unique labels is no big deal for a computer. Every automatic "builder" program already do this. Except they're usually enumerated (i.e. box1,box2, box3,
Reply back when robots start figuring this out on their own without being taught (read "programmed").
Somehow I doubt it conforms to Universal Grammar though.
The links I've seen about this go on and on about how the robots invent and use "words." But language is not words; language is grammar; language is a set of rules for recursively constructing highly complex expressions from smaller subparts. This is Linguistics 101 material.
The way you distinguish somebody with Linguistics training from a layperson is that the layperson will talk about language as if it's a "bag of words" and overall focus too much on the words, whereas the linguist will tend to see most words as either (a) the filler that goes inside the phrases and sentences, or (b) the stuff whose formation is constrained by the general phonological and morphological rules of the language. Or, short and sweet: words are boring unless they're function words like "the."
Are you adequate?
It sounds exactly like human screaming. Odd that.
... we put that trust to the test. BAM, Robots gave us 6 extra seconds of cooperationGood job robots. I'm Cave Johnson, we're done here.
http://www.youtube.com/watch?v=AZMSAzZ76EU
std::disclaimer<std::legalese> sig=new std::disclaimer; sig->dump(); delete sig;
We have had software that was capable of imitating how humans use language for (quick bit of research at http://en.wikipedia.org/wiki/SHRDLU) at least 30 years. The patterns seems to be:
1) Researchers invent a method that from the outside looks similar to how humans use language in a very limited domain
2) Public and media watching say “Look! They are doing just what we do!”
3) Researchers attempt to generalize their limited success to a broader area.
4) Researchers fail, but this is not publicized because failure is boring.
5) Researchers conclude the similarity to human behavior was superficial (this step is optional).
6) Researchers somewhere else try a different tactic.
7) Go to step #1.
Human language began with humans associating sounds they made with objects. Afterwards, they associate sounds with conceptual things like actions. It's only they they combined objects with actions into one meaning that grammar is developed for consistency and ease of understanding.
It probably took humans an insane amount of years before such things as grammar was developed slowly passing on each advancement to each of their generation.
You think robots can achieve something better then humans instantly? Of course this is just pre-programmed logic designed with this purpose in mind so how much cheating vs how much real adaptic logic is in their is hard to say.
But to say it's not a language, a language is but a method to communicate no matter the form of sound used. It's simply a primitive one at best.
Wow, that clears it up. Thank you for figuring out a topic that nobody has!
Are you adequate?
It just beeped and gave me a nanobot sandwich.
Hmm.. I wonder howmuch will it take to develop that one fundamental part of any language. Swearing.
Where's the danger? I think that would be amazing.
$ echo "ceci n'est pas une pipe" | sed -Ee 's/(eci n|pas )//g'
The headline and lead on this /. story are inaccurate. There has been ongoing research in evolution and generation of artificial languages since at least the nineties. See for, example, the work of Luc Steels: http://arti.vub.ac.be/~steels/
Also see this video: http://www.youtube.com/watch?v=7u9OvtEkd1A
The robots played where-are-we, what-direction and how-far games, to create three different types of words. The coolest part of the study is that once their language is created, the robots can refer to places they haven't been to. That's imagination. Then they go explore and meet up at the place they previously referred to using their words for distance and direction.
I bet they're uploading the same software to all the robots. Therefore they already share something: the way they learn.
Although this is interesting, a test should be done with software that was developed by different independent teams.
Privacy is terrorism.
>What do robots need and why would they be developing a language if they don't have any needs? In one sense, a robot species' main "need" is to impress humans well enough to copy them. Pioneer robots have to be useful in research labs for people to keep making them. Language learning robots are a specific combination of hardware and software. {motives, needs, instincts, ...} have relatively clear meanings for carbon-based life forms but are loaded when applied to non-carbon agents. Robots, like chess-playing programs aren't alive in the same sense of the word. But evolution does apply to systems other than carbon-based living things. Programs do solve problems, and the better they solve them, the better chance they have of "survival".
Within the short time frame of a language game, it's misleading to think about the robot as having a "life". It's more reasonable to think of the _language_ as having the life. Languages are constantly changing. A language variant is harder or easier to learn, more or less expressive, more or less useful. Language variants that have the right combination tend to survive longer. The robot and the language learning system is like the ecological niche.