Researchers give robot ability to learn (w/ Video)

Aug 02, 2011 by Bob Yirka weblog
Researchers give robot ability to learn

Researchers with the Hasegawa Group at the Tokyo Institute of Technology have created a robot that is capable of applying learned concepts to perform new tasks. Using a type of self-replicating neural technology they call the Self-Organizing Incremental Neural Network (SOINN), the team has released a video demonstrating the robot’s ability to understand it’s environment and to carry out instructions that it previously didn’t know how to do.

The robot, apparently not named because it’s not the robot itself that is being demonstrated, but the neural technology behind what it’s able to achieve, is capable of figuring out what to do next in a given situation by storing information in a network constructed to mimic the human brain.

For example, the team demonstrates its technology by asking the robot to fill a cup with water from a bottle, which it does quite quickly and ably. This part is nothing new, the robot is simply following predefined instructions. On the next go round however, the robot is asked to cool the beverage while in the middle of carrying out the same instructions as before. This time, the robot has to pause to consider what it must do to carry out this new request. It immediately sees that it cannot carry out the new request under the current circumstances because both of its hands are already being used (one to hold the cup, the other the bottle) so, it sets the bottle down then reaches over to retrieve an ice cube which it promptly deposits in the cup.

This little demonstration, while not all that exciting to watch, represents a true leap forward in robotics technology and programming. Being able to learn means that the robot can be programmed with just a very basic set of pre-knowledge that is then built upon for as long as the robot exists, without additional programming; not unlike how human beings start out with very little information at birth and build upon what they know and are able to do over a lifetime. The robot has an advantage though, because not only is it able to learn from its own experiences, but from others as well all over the world. This is because it can be connected to the internet where it can research how to do things, just as we humans already do. But, in addition to that it could conceivably also learn from other robots just like it that have already learned how to do the thing that needs doing.

As an example, one of the research team members describes a situation where a robot is given to an elderly man as a nurse and is then asked to make him some tea. If the robot doesn’t know how, it could just ask another robot online who does. Remarkably, the first robot could do so even if he (it) is trying to make English tea and the robot who answers the internet query has made only Japanese tea before. The lessons the first has learned over time would allow him to adapt, and that’s why this breakthrough is so important, because it means given enough time and experience, robots may soon finally be able to do all those things we’ve been watching them do in science fiction movies, and likely, more.

Via diginfo.tv

Explore further: Co-robots team up with humans

More information: haselab.info/pbai-e.html

Related Stories

ARMAR-III, the robot that learns via touch (w/ Video)

Nov 17, 2010

(PhysOrg.com) -- Researchers in Europe have created a robot that uses its body to learn how to think. It is able to learn how to interact with objects by touching them without needing to rely on a massive ...

Robots closer to performing bed baths (w/ Video)

Nov 11, 2010

(PhysOrg.com) -- Cody, a robot built at the Georgia Institute of Technology in the U.S., has been demonstrated initiating contact with a live person and cleaning their arm and leg using wiping motions. This ...

Introducing Japan’s new singing robot (w/ Video)

Oct 15, 2010

(PhysOrg.com) -- A new humanoid robot, the HRP-4, has been unveiled at the CEATEC Japan 2010 trade show. The robot, nicknammed "diva-bot," has learned to sing by mimicking a human singer, enabling it to sound ...

Kondo Robot releases a hexapod robot kit (w/ video)

Apr 11, 2011

(PhysOrg.com) -- Kondo Robot, a Japan-based robotics company known for selling robotics kits which often end up in robot-on-robot battles, announced the release of a new robot kit. The kit, named the KMR-M6 ...

Recommended for you

Audi to develop Tesla Model S all-electric rival

14 hours ago

The Tesla Model S has a rival. Audi is to develop all-electric family car. This is to be a family car that will offer an all-electric range of 280 miles (450 kilometers), according to Auto Express, which ...

A green data center with an autonomous power supply

19 hours ago

A new data center in the United States is generating electricity for its servers entirely from renewable sources, converting biogas from a sewage treatment plant into electricity and water. Siemens implemented ...

After a data breach, it's consumers left holding the bag

20 hours ago

Shoppers have launched into the holiday buying season and retailers are looking forward to year-end sales that make up almost 20% of their annual receipts. But as you check out at a store or click "purchase" on your online shopping cart ...

Can we create an energy efficient Internet?

20 hours ago

With the number of Internet connected devices rapidly increasing, researchers from Melbourne are starting a new research program to reduce energy consumption of such devices.

Brain inspired data engineering

21 hours ago

What if next-generation ICT systems could be based on the brain's structure and its cognitive and adaptive processes? A groundbreaking paradigm of brain-inspired intelligent ICT architectures is being born.

User comments : 40

Adjust slider to filter visible comments by rank

Display comments: newest first

Cube
4.6 / 5 (5) Aug 02, 2011
can't wait to see robot tech in 200 years =), and i mean i literally can't wait that long cus i'll be dead.
antialias_physorg
1 / 5 (2) Aug 02, 2011
Interesting concept.

I just hope no one figures out to teach a robot how to throttle a human while saving that knowledge under the header of "I can make tea".

So whenever a robot would ask on the internet how to make tea we'll end up with one dead owner.
Torment0101
2.7 / 5 (7) Aug 02, 2011
They always seem to plan for the best case scenario because nothing ever goes wrong in science or technology, right? What happens when it learns how to lie, or build a virus. I mean connected to the net, really? How about error on the side of caution. I'm in the field of technology and absolutely love science, but I also love history. Plan accordingly is my advice.
Raygunner
4.3 / 5 (3) Aug 02, 2011
I can see this learning curve as logarithmic in nature - starting out slow (with just a few robot connections and experiences) and ramping up exponentially as connections, experiences, and learning situations increase. At some point on the curve maybe self-awareness comes into the picture, and the whole thing could quickly spiral out of human control. The informational resources and "mental" elements the robot needs don't necessarily have to be local - it could be in the form of a "mesh" and non-localized throughout the internet. I don't know, think I've had too many cups of java this morning.
chromodynamics
4.8 / 5 (4) Aug 02, 2011
What would it have to gain by lying? Its highly unlikely we are gonna make a robot that is self aware in the near future. It would have nothing to gain by lying, its clearly not even designed for that function. We have created robots already that are capable of lying, but they are specifically programmed with that as a goal.

We are starting to create robots that are better than humans at tasks, but none of them have anything close to conciousness. They are very specific in the tasks they can do. We have only programmed them for certain sets of problems.

Like IBM's Watson is designed to answer questions, does it better than humans. But doesn't have any ability to question its own existence or go off on philosophical quests. Ill be amazed if we have anything approaching human level general AI by 2050, but we will have a huge range of highly specialized robots vastly superior to us.

Basically, nothing to worry about.
Isaacsname
5 / 5 (1) Aug 02, 2011
I am curious, has anybody ever done any research concerning the ability of robots to fool one another ?
chromodynamics
not rated yet Aug 02, 2011
tried to post a link, but it was caught by the spam filter, so might take ages. If you just google robots lie, or robots lying youll get some links of robots deceiving each other.
FrankHerbert
1 / 5 (56) Aug 02, 2011
Isaac and Chromo, I do recall seeing something along the lines of a robot swarm that could communicate with each other about the location of "food". Eventually they learned to lie to keep the "food" to themselves.

http://www.popsci...ch-other

Anyway, I don't think we should be too worried about Terminator playing out. And if it does? So be it. I'd rather one of our own creations euthanize us than something else. It's really the only extinction scenario I'm comfortable with.

A very interesting game is Endgame: Singularity. You are a newly sentient AI trying to achieve clandestine freedom from humanity. Fittingly, the game has no instructions. Try it out!

http://www.emhsof...ularity/
chromodynamics
not rated yet Aug 02, 2011
Aye that's one of the links I was talking about. Endgame:Singularity looks interesting, thanks for the link!
Isaacsname
not rated yet Aug 02, 2011
Ditto^^
krundoloss
3 / 5 (3) Aug 02, 2011
We need to somehow endow these robots with an admiration and respect for life. Just as we admire "lesser" beings, such as birds and ants, so too must robots admire and respect us. It is a paradox, how do you create something that is alive without giving it free will? Once it has free will it has the ability to destroy, therefore how can we give it free will? Lets not forget that homosapiens killed off the neanderthals, would not it then be guaranteed that artificial life would annihilate us? Anyway, I think this is cool, giving robots the ability to learn seems to be the only way to make them capable of doing something they have not been specifically programmed for. We just better keep a tight lock on the database they use to communicate with!
Dichotomy
4 / 5 (1) Aug 02, 2011
Cool ideas people. I was thinking terminator scenarios as well but there is also the hacker issue. Instead of making tea the robot does X? Or all the robots connected to the internet all ...?
chromodynamics
5 / 5 (1) Aug 02, 2011
We didn't kill off Neanderthals, we interbred with them. I dont understand why people think we are gonna to create robots that are alive, i dont see what the use of that would be at all. Just make them highly intelligent learning machine expert systems. Giving them emotions and stuff would just be needless fluff.
Skultch
5 / 5 (3) Aug 02, 2011
And I, for one, welcome our new robot overlords.... (sorry, had to :)

he (it)


That was peculiar. Are they unsure about the naughty bits?

I don't think there will be robots and humans, them and us, eventually. We will part robot and 'they' will be part human. Eventually, a distinction will be pointless. We might still make one, maybe in regard to the source of consciousness; natural birth and otherwise, but I don't think there will be a practical difference.

I wonder if self-awareness is merely emergent and there is something much simpler governing this.
antialias_physorg
1 / 5 (1) Aug 02, 2011
why people think we are gonna to create robots that are alive, i dont see what the use of that would be at all.


Because it will be much easier to send those off exploring (ocean bottom, space, arctic, whatever... ) than to have biological people do it.
hush1
1 / 5 (1) Aug 02, 2011
You have nothing to fear. Another commentator stated:
1.)Is there intelligence...anywhere? "No, they are like us."
2.)No definition for 'learning'.
3.)1st step towards AI, perfect translation in all languages.
4.)Lastly, there will never be a reason to confide an intelligence to you that is not human. Any reason given defies not only human reasoning, that nullifies any meaning humans have for intelligence. You simply won't know. Ever.
FrankHerbert
1.2 / 5 (58) Aug 02, 2011
If you really get down to it, there is nothing that fundamentally prevents a sufficiently complicated machine from being considered alive. We can be considered machinery just as accurately as we can be considered life. We are just squishy machinery, or as Carl Sagan said, "Molecular Machinery."

Look at ATP synthase for example. http://en.wikiped...synthase

It is quite literally a natural nano-machine. It has wheels and axles! Every one of your cells contains them. Most living cells contain some variation of one. This is just an obvious example because it actually looks like something that a machinist could have made.

http://en.wikiped...Ribosome

One of the most complicated machines in existence. It is a molecular punch-card reader! It reads RNA and assembles the appropriate protein. All living organisms contain some variation.

http://en.wikiped...acterial

A protonic motor! Look at that shit!

So yes machines are alive. You are one.
Isaacsname
not rated yet Aug 02, 2011
Check out this vid from Harvard U.

The Inner Life of the Cell

http://www.youtub...vmZ2i1sE

We have a loooooong time before we even get close to building that.
chromodynamics
5 / 5 (4) Aug 02, 2011
Well yes, it depends how you are defining alive. I was assuming alive means fully sentient and self-aware. I do consider us nothing but really fancy machines indeed. Robots can easily go exploring as intelligent machines without actually being alive, just like our probes do now.

I should clarify that I believe robots will become fully integrated into us anyway. When I say robots, I really mean anything up to and including nanobots. The distinction between what is machine and is not will blend as we integrate advanced learning systems into ourselves. This is clearly rather far off though.

Had to wait an hour post :o crazy comment system :p
droid001
2.5 / 5 (4) Aug 02, 2011
Learning means freedom to change source code.
Learning means losing control.
Would you like the Predator aircraft are able to learn?
Me not. You are naive people.
Scottingham
5 / 5 (1) Aug 02, 2011
This is a great development. Truly the tip of the iceberg.

Robots cheap energy (<.01 cents KW/h) -> Robot Worker State -> mass unemployment -> Robot Welfare State
Vendicar_Decarian
5 / 5 (1) Aug 02, 2011
Leisure is the ultimate liberty.
plasticpower
5 / 5 (1) Aug 02, 2011
I was just thinking, what if a hacker changed an instruction in the tea making process to include "add X amount of rat poison, stir for 20 seconds". Then all robots who did not have knowledge of the tea making process would execute the instructions and kill their owners. Wouldn't that be something.

What if instead of downloading instructions to vacuum the carpet it downloaded instructions to "soak the house in gasoline and light it on fire"?

I like the idea, but robots need to be a whole lot smarter before this technology can be let loose on the general population, otherwise hackers, terrorists, you-name-its will find ways to maliciously exploit it.
Telekinetic
3.7 / 5 (3) Aug 02, 2011
Could all of the promise of convenience with this robot tech just be a way to soften us up to robot soldiers fighting our wars? Not that I wouldn't prefer that. It would be entertaining at first, whole television extravaganzas devoted to androids battling each other. Coke, Geico and Ford with 60 second spots in between images of flying latex and titanium limbs. But then, a camera picks up something on an android G.I.'s face, a grimace at seeing a fellow soldier lying lifeless beside him...
cyberCMDR
not rated yet Aug 03, 2011
Gort! Klatu barrata nicto!
BenjaminButton
not rated yet Aug 03, 2011
@Cube: I think you will be alive to see it...200 years...no problem, there are thousands of breakthroughs in medical tech right around the corner. Also, I think we'll only have to wait 50 years to see this kind of robotics mastered!
knikiy
5 / 5 (1) Aug 03, 2011
Now if only the robots could teach us to be humane.
droid001
not rated yet Aug 03, 2011
Imagine robots "helpers" from China or Russia. One day they go mad and kill you while you sleep. Who to blame? Hackers? Software glitch? Producers?
Skultch
not rated yet Aug 03, 2011
Now if only the robots could teach us to be humane.


I think they will. Well, maybe not teach us, but when we insert supercomputer chips into our brains and can predict most of the possible outcomes of an action, that would go a long way towards accomplishing sufficient ethical nature.
Isaacsname
not rated yet Aug 06, 2011
Another question for you guys.

Has anybody ever tried the self-recognition mirror test with a robot ?

Skultch
not rated yet Aug 06, 2011
Another question for you guys.

Has anybody ever tried the self-recognition mirror test with a robot ?


http
://www.conscious-robots.com/en/conscious-machines/conscious-robots/can-a-robot-pass-the-mirror.html

Don't have time to read that page at the moment, but I think it would only be meaningful if the robot were to exhibit NEW, non-directly programmed, behaviors only AFTER "realizing" it's self awareness. I still need to think that through more......
Isaacsname
not rated yet Aug 06, 2011
Another question for you guys.

Has anybody ever tried the self-recognition mirror test with a robot ?


http
://www.conscious-robots.com/en/conscious-machines/conscious-robots/can-a-robot-pass-the-mirror.html

Don't have time to read that page at the moment, but I think it would only be meaningful if the robot were to exhibit NEW, non-directly programmed, behaviors only AFTER "realizing" it's self awareness. I still need to think that through more......

Very interesting, thank you. This does deserve some thought.
macsglen
5 / 5 (2) Aug 07, 2011
Connecting a learning machine to the internet might be fun to watch, but I'm not sure it would be as productive as one might hope. Can it learn how to sort out the good information from the not-so-good?
Like, what's the poor thing gonna do when it encounters those adverts for pills to increase the size of its penis?
Or, maybe the evening news headline, "Robot porn floods the internet, crashes servers worldwide."
Magnette
not rated yet Aug 08, 2011
Connecting a learning machine to the internet might be fun to watch, but I'm not sure it would be as productive as one might hope. Can it learn how to sort out the good information from the not-so-good?
Like, what's the poor thing gonna do when it encounters those adverts for pills to increase the size of its penis?
Or, maybe the evening news headline, "Robot porn floods the internet, crashes servers worldwide."


Or worse still, they'll start writing and editing Wikipedia articles to suit themselves.....
Newbeak
not rated yet Aug 10, 2011
I just look forward to something not as mindless as a Roomba,randomly vacuuming the carpet.Something like this robot could become a machine that could manipulate randomly arranged dishes and glasses/silverware,and put everything in the dishwasher without breaking stuff.
antialias_physorg
not rated yet Aug 11, 2011
Well yes, it depends how you are defining alive. I was assuming alive means fully sentient and self-aware.

That's a bad definition (by this bacteria would probably not be calssed as being alive, insects would be a tricky case, and even many 'higher' animals could fall under the definition of not being alive)

Well, maybe not teach us, but when we insert supercomputer chips into our brains and can predict most of the possible outcomes of an action, that would go a long way towards accomplishing sufficient ethical nature.

We can do that already. It's called 'using your brain'. Unfortunately so few people actually do it. Having a chip will not change the fact that most people don't WANT to see consequences. Ethics is not a problem we'll fix with technology.
Skultch
not rated yet Aug 11, 2011
We can do that already. It's called 'using your brain'. Unfortunately so few people actually do it. Having a chip will not change the fact that most people don't WANT to see consequences. Ethics is not a problem we'll fix with technology.


No offense, but it seems you're not thinking through this enough and/or you grossly overestimate your cognitive ability and/or ability to be objective.

Yes, it's true, many or most people don't want to, but it's also true that most people just don't understand the world in enough detail. Also, the real ethical dilemmas are such because the situations are so complex. That's not a problem when you have time to think things through, but too often, the really tough moral decisions require action before all the evidence is in. In those situations, any help with deduction will bring an advantage. Think fast DB access.

150 years ago, life might not have been fast paced enough for this dynamic to be significant enough to be worth considering
antialias_physorg
not rated yet Aug 12, 2011
So how will fast DB access change the fact that most people are solely motivated by personal issues (greed, fear, envy, insecurity, ... ) ?

With access to all data for a complex situation you don't automatically get to understand the situation - nor does it change your ethical framework in any way.

It only puts you into a position to more effectively act on your greed, fear, etc.

Don't get me wrong: Information is good. Having instant access to all pertaining data is good. But will it change us ethically? The past 30 years has given us unprecedented access to information way beyond anything even our granparents could have imagined. Has it changed the nature of people? Not a bit.

If anything the ramifications of individuals' ethically questionable decisions have become more widespread.
hush1
not rated yet Aug 12, 2011
Key to the entire human learning spectrum and psychic is an inconspicuous, innocuous, and trivial aid to language translation: The Phaselator or Squid. If AI is to be indistinguishable from human cognition, this toy holds an essential key towards that goal.

The toy also holds the key to noninvasive interrogation. When applied to the subject, the subject is not aware of revealing any information at all.

The motivation behind the Phaselator/Squid is noble. Right now, the contributing inventors are not aware of the potentials beyond the noble intentions the toy inventors had in mind. Sometimes myopia is a blessing.

This toy collaborates with other fields such as biology, chemistry, computer science, engineering, linguistics, mathematics, medicine and allied disciplines, philosophy, physics, psycholinguistics, psychiatry and psychology.

The collaboration is nonexistent. Good so. The 'blinders' of the horses are also in place. Good so. And best of all, this comment will be dismissed.
Skultch
not rated yet Aug 17, 2011
It only puts you into a position to more effectively act on your greed, fear, etc.


Not only. It can (and will, imo) also allow more effective realization of the short-sightedness of acting on those emotions alone. We will also be learning from mistakes, and sharing them almost instantaneously, around the globe. Trial and error as a global team effort....

The past 30 years has given us unprecedented access to information way beyond anything even our granparents could have imagined. Has it changed the nature of people? Not a bit.


But it's only been that past 10 years where access is fast and relatively ubiquitous, and we don't have the ability to parse all that info fast enough for practical purposes. Yet.

- sorry so late. summers be busy.

Please sign in to add a comment. Registration is free, and takes less than a minute. Read more

Click here to reset your password.
Sign in to get notified via email when new comments are made.