Rubbish AI

http://news.bbc.co.uk/1/hi/technology/7544099.stm

Make sure to watch the video - the man trying to get up is especially comical.

Are we seriously supposed to be impressed by this? It obviously doesn't work very well.

It is reasonably impressive I guess, considering it starts with practically no information. It reminds me of

http://www.fun-motion.com/physics-games/sumotori-dreams/

It reminds me of stupid babies... they can't seem to walk when they are born either. What's with that?

I would say its pretty impressive too, considering the machine is learning from almost nothing. There is no preset AI, just a set of pre-given rules like gravity, bones, muscles etc. That dog just 'learned' how to get over the wall.

I have seen something similar to this before using very simple unprogrammed spider robots that had only one function, to walk to a target as fast as possible. It took them a while but eventually they were able to walk efficiently to the target. However, if you pulled off a leg, they would spaz around for a bit then adapt and walk differently towards the target. Even with just one leg it learned which moves it needed to struggle forward with no programming needed. All it needs to know is what direction it must travel.

Each time progress was made towards its target it 'learned' using just a simple logic matrix that every action performed is either a positive or negative step.

Lol @brucey. Yeah human babies are physically pretty pathetic compared to the animal kingdom :D

break dancing at its best ;)

I'm not impressed, just some random moving joints.
There is still no real learning part in this one.

If you would like to pick up an object you won't randomly move your all limbs until the object is in your hand.

The human in the video reminds me of the game Toribash :-)

That reminds me of this fellow: http://www.ted.com/index.php/talks/hod_lipson_builds_self_aware_robots.html

Impressive to me. I'm sure the program has an objective to accomplish.

Have you ever seen a baby trying to learn how to stand up by doing a backflip and landing on its face?

Me either.

Cool, but it still looks like it needs some work...i don't see babies trying to learn how to walk like that haha.

I also think they should have put more work into a more important 'learning' activity. I don't see how it would be useful to use something like that in something based on a character who got such hardcore amnesia they forgot how to walk....

I would be much more impressed if you could add/remove limbs, and it would learn how to use them efficiently.

It's kind of like a baby on steroids and LSD.

But they never successfully learned to do anything. They just moved their joints around and sort of oriented themselves upright.

Can't say I was impressed. I would expect it to evolve its movements in a more intelligent manner and show more progression to something that can stay balanced and move believably.

I think if the technology worked it could be pretty interesting. Might even teach athletes better ways to move their bodies and those extra little bits of efficiency and performance. Looks like they have a long way to go :)

Sounds like mankind to me! :)

The video and the text don't quite go hand in hand. Either the text is seriously exaggerating its abilities or the video doesn't do justice to the intelligence.

plus parts of it were played in fast forwards
under normal speed he might not be that good a break dancer ;)

It all was. I guess they knew people weren't going to sit and watch it flailing about hopelessly for 20 minutes waiting for its backflip/faceplant finale.

I'm a bit disappointed - they didn't seem to 'learn' to jump over the fence or do a back-flip to me, they just seemed to do it by accident.

Slightly impressed. Would be a bit more interesting to see how they moved when they reached a working solution. The dog was stuck half way and I would like to see it climb the wall a second time. And was the humanoid able to stand up? He was close, but would be fun to see him stand up and keeping the balance.

I thought it was quite impressive. The only curious thing is that I'm sure I recognize the second one from a friday night out in Cardiff a couple of months back.

As i said above, this is a hybrid "BEAM Robot" simulator (Biology, Electronics, Aesthetics, and Mechanics). If you watched it in slow motion it would probably become clearer what its actually doing is completely random and then forming a pattern.
A BEAM robot's "intelligence" comes from trial and error from very simple logic with no CPU involved, just simple oscilating logic from lots of sensors. In reality it has no intelligence. However, the Neural Net part just looks and learns (in a manner similar to the way a real brain works).

The COOL part is that the data from the robots mistakes in a certain situation can be used to build a neural network, so the next time the robot encounters a similar situation it could take over full control and try to do it itself. If it fails, rather than just standing there totally confused, it allows the BEAM to take over to figure the rest of the task out. (and so expanding its own intelligence and logic paths all the time)

The impressive thing about the article is that's exactly what they are doing:. A VR model is easier and faster to build and experiment with than a real one, simulations can be run and learned from in a fraction of the time. (Cheaper too :)

The bbc reporter probably completely missed the point and just filmed some flopping about robots for 5 minutes.

I personally find this more interesting.
http://www.ted.com/index.php/talks/jeff_hawkins_on_how_brain_science_will_change_computing.html

So do I. We need more people like that in this world, who maybe have more power to make encouraging changes. Thanks for sharing that Andy.

Although I don't understand how one could apply those ideas into AI and 'just' have a pattern system. Would you not need to input a real 'Goal'? Telling an ultra intelligent computer to understand won't do it lol.

It's almost as if people think if you give a computer the ability to learn, they'll do everything automatically, but without a programmed concious that contains goals, that doesn't seem possible. It's like giving a sub-concious without a concious.

..to me, this guy talking trash..more/less trash..it seems that everybody are wrong and hes is right..i dont like people with such attitude..at least leave some small window for other possibilities..but hell..he knows everything..

?

Yes every evolving 'organism' needs a "boot up" program containing a basic set of rules on what we are supposed to learn in order to adapt to the surroundings, to survive and to exist. (eg. the desire/impulse/drive to move as efficiently as possible, rest, eat, run away or fight if threatened etc.) I assume in bio organisms these "initialisation" instructions are stored somewhere in the DNA.

If the situation changes (ie loose an arm) it takes time for the brain's synaptic pathways to re-program to realise the limb is no longer there. If you threw a ball at someone who has recently lost an arm they will instinctively (and unsucessfully) try and catch it with the missing limb until the brain (or neural network) adapts/re-programs itself to function without it. Try and write a conventional AI instruction set for that lot and you would soon see why it needs something more new and neural-like. Basically a computer/program that can constantly re-program itself.

I think it might be possible (in the very distant future) to program a new form of 'frequency' based on the concious, but then the results would vary completely, depending on what patterns are being stored.

Frequency > Thought (Memory Access based on Frequency) > Comparison based on Memory > Action based on comparison > Alter next frequency based on Access, comparison, and action.

You would still need to give the ai senses to make it mammal-like. What makes us happy is based on society and our surroundings, so the ai would become happy from certain properties based on its own surroundings (data).

I think the main issue, is that when a baby is born, it is cared for, and develops based on what it sees another human (mother) do. So AI would still, even in this case, require some form of instruction, although the instructions wouldn't need to be pre-made, although that would probobly take the A out in AI.

It's an interesting topic, but unlikely anything extremely significant will come of it in our lifetime. The more I think of it, the more I feel like everything people do is based on reactions, not 'real' decisions.

.NVM, lost the link :(.

Well that was pretty naff and I saw nothing resembling intelligence and certainly not worthy of a BBC news item.

It looks like a newbie exported animations wrong and plonked them in Blitz3D.

If you've ever watch those funky nature programmes on TV then nothing in real life acts like that when trying / learning how to walk.

If the effect was for them to appear like they've got some crazy neuro disease causing total body spasms then it's brilliant!

>..to me, this guy talking trash..more/less trash..it seems that everybody
>are wrong and hes is right..i dont like people with such attitude..at least
>leave some small window for other possibilities..but hell..he knows
>everything..

The thing is that he IS right. And the people doing the research like he is, all agree on it. The neo cortex is an awesome patternmatching machine. That is not even being disputed. That's why artificial neural networks are so useful for patternmatching.

>Yes every evolving 'organism' needs a "boot up" program containing a basic
>set of rules on what we are supposed to learn in order to adapt to the
>surroundings, to survive and to exist. (eg. the desire/impulse/drive to
>move as efficiently as possible, rest, eat, run away or fight if

All of this is handled by the reptilian brain(the brainstem)

>threatened etc.) I assume in bio organisms these "initialisation"
>instructions are stored somewhere in the DNA.

You would be right. I read an article on how when we wake up, we are actually running some form of program that boots the brain.

Look at this...
http://www.ted.com/index.php/talks/jill_bolte_taylor_s_powerful_stroke_of_insight.html

She talks about the sequential part and the massively parallel part.

The brain is such an interesting topic!

I did not know that. It is 'built and programmed' by DNA though.

DNA is nature's 'start up' program. The cells themselves are like tiny robots commanded by DNA to converge and form something for a given task. It covers practically everything except experience. (Although there is some talk of genetic memory but i don't know if thats a fact)

Yeah the brain & AI has always fascinated me too. (wouldnt know where to begin if i had to program it though :O )

That jill woman had an interesting story (a bit evangelical/carried away towards the end, but she ended quite well)

"Real learning", "real intelligence" in this thread seems to be defined by how similar it is how human brains function. The space of all minds is extremely large, and there is no particular reason to replicate a human brain to accomplish some task, like "general intelligence". Simulating a human brain is of course one option, but this route has many problems.
http://www.singinst.org

"AI is so stupid it is unable to make coffee."
"Meat intelligence is so stupid it is unable to calculate 3412*123123 in 0.1 seconds."

Human/advanced animal intelligence, maybe not needed, no... and perhaps a little bit too scary.
(Although i was kind of hoping for a hot Angelina Jolie 'pleasure-bot' to look after me in my old age. :P)

However, 'general intelligence' in simple synthetic lifeforms like insects or certain marine creatures could be very useful for dealing with a robot's basic "survival instinct" and the ability to move and adapt correctly. When combined with conventional AI systems or human influences to 'steer' it in a more controlled way (remote or not) it really gets useful.
Imagine the usefulness in areas like research in dangerous or impossible to get to areas like off-world, deep-sea or sub-subterranean research etc.. Perhaps even in the further future, tiny 'simple' robots with a synthetic nervous system (with no room for processors/electronics) that "instinctively" knows how to swim in the body to fight diseases.

..Or in symbiotic vehicles. Imagine a giant "spider" ATV, the driver simply has to steer it and make it faster or slower, nothing to do with controlling the legs. However the thing 'knows' how to behave differently on certain terrains to get max speed, climb over stuff, swim and jump gaps without being told, knows that when falling distance it should "brace" itself for the impact etc... (as if the driver was riding a horse).

..well..maybe there are some other things involved here rather than bunch of cells..because if you talking about brain as a computer, then who is the 'programmer' who designed 'software' for it?? Cells itself cant do a thing without coordination, so..is there something more than pure bunch of cells as Jeff Hawkins talking about? In whole his speech, I see brutal ignorance of God..so, is it wise, having in mind that so many things we absolutely dont know, not just about our small world but also whole universe, is it wise to at least keep open one door for possibility of God, rather than brutally ignore it? To me guy is lost and his theories, whatever they are will lead to one big nowhere..

>..well..maybe there are some other things involved here rather than bunch >of cells..because if you talking about brain as a computer, then who is >the 'programmer' who designed 'software' for it??

There doesn't need to be a designer. Replication errors and cell division is all that's needed.

>Cells itself cant do a thing without coordination,

They contain dna, sustain and maintain themselves and divide themselves.

>so..is there something more than pure bunch of cells as Jeff Hawkins
>talking about?

You phrase it as a question.... Unfortunately, the scientific method doesn't deal with superstition, only what can be observed.

>In whole his speech, I see brutal ignorance of God..

He is talking about the neo cortex. Why should there be any reference to 'god'.

>so, is it wise, having in mind that so many things we absolutely dont
>know, not just about our small world but also whole universe, is it wise
>to at least keep open one door for possibility of God, rather than
>brutally ignore it?

Where is he talking about rejecting god! I've seen a few videos with him, and I have never seen him comment on 'god' at all.

As for your idea that we should maintain superstition to explain what science canno explain, well, that just silly.

>To me guy is lost and his theories, whatever they are will lead to one big
>nowhere..

What exactly is so offensive to you about studying the brain, particularly the neo cortex?

To dismiss a guy who studies brains for a living and knows a heckuvalot about how they work, you must have some impressive scientific credentials.

>>Where is he talking about rejecting god! I've seen a few videos with him, and I have never seen him comment on 'god' at all.<<

Watch link posted above again, and you will see that on few places he did mention God and making some jokes with it, as a nonsense...

>>There doesn't need to be a designer. Replication errors and cell division is all that's needed.<<

It is interesting..then how come some mammals who has more brain cells than we do, havent reach even 2% of our intelectual capabilities?? Dolphines for example?

>>They contain dna, sustain and maintain themselves and divide themselves.<<

Really, and what is basics for such extraordinary well designed structure of DNA? Its just popping up from nowhere and compose itself in such well ordered structure in to extreme??

My point is simple..there are many questions way deeper than pure understanding of brain mechanisms..and if you watch carefully this guy keep telling that hes right and noone else had or has vision what to do...microprocessor guy is someone with lack of vision..he just 'invent' processor but its not that good..he has no theory....what a crap..guy claiming that what he think is correct way without leaving door for possibilities that he is mistaken..that means hes in absolutism and such things doesnt exist on this world..its all relative, and im very sure he and his 'theories' will end up nowhere, and he will still continue developing his mobile devices because he will not eat bread out of this 'theories' of his..thats for sure...

>Watch link posted above again, and you will see that on few places he did
>mention God and making some jokes with it, as a nonsense...

So you estimate the value of a persons research based on a few seconds of video of him telling a joke.

>It is interesting..then how come some mammals who has more brain cells
>than we do, havent reach even 2% of our intelectual capabilities??
>Dolphines for example?

Because they evolved in a different direction.

>Really, and what is basics for such extraordinary well designed structure
>of DNA? Its just popping up from nowhere and compose itself in such well
>ordered structure in to extreme??

No, creationists believe in things just popping up fully formed. Evolution is based on improvement upon improvement upon improvement...

>there are many questions way deeper than pure understanding of brain
>mechanisms..

So you are saying that noone should look at this subject, because you think it's unimportant. I disagree... The small stuff is usually the most worthwhile.

>and if you watch carefully this guy keep telling that hes
>right and noone else had or has vision what to do...

He is right about the patternmatching aspects of the neo cortex. As for the approach to understanding the brain, I think he is also right, although he isn't the only one to look for a different way.

>microprocessor guy is someone with lack of vision..

Understanding the brain is lack of vision?

>he just 'invent' processor but its not that good..

He didn't invent a processor, he integrated various off the shelf parts into various products.

>he has no theory....what a crap..guy claiming that what he think is
>correct way without leaving door for possibilities that he is
>mistaken..

I don't know what your problem is. Science is about discovery and verification of discovery, not discovery and remembering to secondguess oneself, incase one is mistaken.

>that means hes in absolutism and such things doesnt exist on this
>world..

Are you saying that gravity isn't real?

>its all relative, and im very sure he and his 'theories' will end up
>nowhere,

So his research fits with the other research and all available science on the neo cortex, and yet his theories(now you suddenly decide he does have theories after all) will end up nowhere. I don't think you understand the scientific method.

>and he will still continue developing his mobile devices because
>he will not eat bread out of this 'theories' of his..thats for sure...

You really need to study the scientific method.

And of course religion has always been open to science and have never made 'nonsense' claims on new discoveries. :/

Well as for who designed the current version of the 'software', I would simply say nature did. It has adapted and re-programmed itself over millions of years of evolution and started by some complex chemical reactions. Likewise I believe the universe is evolving too and we are the twigs on the end of a very big possibly infinite 'tree'. For all -we- know, the entire universe and big bang itself could be a concious thought, a mere electrical flash in the neuron of some higher being's brain.
While i am not really 'religious' in the usual sense, i do believe in a 'conscious creative force' far beyone our comprehension (call it god if you like) that created, designed or caused the initial 'rules' like physics and chemisty on universal scale (and beyond?). I also believe in positive constructive and negative destructive forces that effect absolutely everything in the universe (Call it good and evil if you like). "GOD" on this scale deserves/needs no pretence of understanding or ignorance as it -is- essentially EVERYTHING from humble cosmic dust to complex thought.

Edit: This is simply 'my belief' of course. (sorry)

That's because you don't understand that self-replicating systems are an emergent behavior of physics and chemistry, and life is as natural a phenomenon as gravity.

Thats kind of what i just said.
"Life" and gravity are both natural phenomena, i agree completely.

My point was 'Life' has reprogrammed itself (evolved) from the influence of its local chemical and physical environment -by- using "chemistry and physics" and that it was born from "chemistry and physics" in the first place.

I was more talking about what? (or who?) wrote the "initialization" program (for the universe in general, not just 'life' as we know it)

You don't understand.

..and im sure you are able understand eons of passed time, within few decades of your lifetime..

No? Care to enlighten me then? If you are talking about life comming from chemical reactions then yes, i do agree, but what made the chemicals and designed the 'laws' of physics?

Erm, of course not, thats why i said:
"....deserves/needs no pretence of understanding"
-Meaning we could never possibly have any hope of understanding what "it" is.

Anyway, i was just what saying what -I- believe. Im not trying to say "this is the way it is, join my cult/religion" or anything. I simply believe in the nature of 'absolutely everything' --EVERYTHING-- as a singular 'whole' like a giant 'chaotic' machine for purposes unknown. (Maybe its some kind of giant brain, who knows.)

Its no more far-fetched or provable than anyone elses beliefs. People can pay homage to a potato for all i care.

..anyway im not even going to bother to argue about "Gods", its pointless, so back to AI eh?

That says it all!

The sigs seem to get bigger each day.. (adblock ftw!)

actually plash, yours is about the same number of pixels high, lol :)

I do however agree, 45 pix max is about right. (I have been meaning to make mine slimmer too but not got round 2 it :P)

yeah who created the matter that created the big bang in the first place?

And who created God?

Talking about cells comming together, anyone see that load of balls that stick to each other to form a shape. They have no intelligence of their own other than how to roll where commanded and stick and act a little bit like cells (except they don't divide). This kind of technology could lead to an intelligent polymorphic liquid that can make any shape and individually change color. (So the T1000 may not be so incredibly far fetched after all! :D)

Meanwhile, back on-topic....

I'm sure that on a TV program awhile back ( the one about the dinosaur mummy I think ) that they use this AI technology to figure out how dinosaurs really move and what speeds they could do.

That bloke one flailing around reminded me of myself coming back from the pub.

Did it conclude that Dinos spent most of their lives thrashing around on the ground and then using their tails to perform silly backflips?

No. But it did show what state they would have been in if pubs had existed in that time (you know lots of staggering about and falling over a lot and causing broken teeth). Eventually it did work.

Oh, didn't you get the memo? God's just there, and so are we. Let's be ignorant and say that's just it, and not look into things any deeper, or in any other way =P

Retimer, it is just as ignorant to believe the there just happened to be some random molecules floating around in space and that they just happened to come in contact with each other to form the big bang. Unless Evolutionists get some real evidence about their theory, it will remain just as faith oriented as Christianity or any other religion.

Can you guys start your own God thread please and stop trying to get this one locked?

Ta!

It all began with a thought. Only thoughts can create something from nothing.

to get back to the topic...

What is good AI? And to differentiate between subjects, lets assume we are talking game related...

Good AI is artificial intelligence which appears to think and perform like a human does. Assuming its AI for a human, obviously.

For example, a child learning to walk will go onto all fours first in order to become mobile, and eventually pull itself up onto its feet by grabbing onto furniture and the like.

One thing it absolutely would not do, is assume a 'lying' position, perform a backflip and end up lying on its face.

In a game, artificial intelligence should follow the same train of thought that a player does. I.e.

Can I get shot if I stand here?
Yes: Find somewhere I can't get shot.
No: Identify a target and shoot at it.

[controversial?]Currently there is no good AI (particularly in games). Until we feel some guilt at hitting ESCAPE or turning off the computer because of the evolving intelligence and emotion we've observed on the screen, then AI has failed. Rule based AI (on its own) can never really offer this possibility.[/controversial?]

Although you could argue that this already happens with Tamagotchi (at least for small children). :)

I suspect it will all truly begin with intelligent operating systems that really grow to understand the user over time. This provides both function for humans, and motivation for the machine itself.

Dont forget a small child has a HUGE amount of input. The reason they don't attempt a backflip and land on their face is:

A) They are not physically able. If babies had developed enough muscles they would most probably injure or kill themselves by doing something just like that (as they have no (or very little) control over them). They learn through trial and error and by absorbing everything going on around them.
In the case of my own son, it was obvious the day he discovered he had a hand and could do things with it. He would stare at it fascinated and then experiment with it, by hitting it on things and touching everything, closely followed by the discovery of 'feet' a week or so later. I am positive if his un-controlled kicking legs and waving arms had had some power he would have launched either himself (or me!) into some kind of somersault before long. (ok mabe not a backflip that perfect, but the result would be pretty horrifying).

B)They spend several months developing and discovering their own bodies and gaining strength before they can even turn over let alone hope to stand up. All the while have adults around to copy and learn from. Its also a while before they realize gravity hurts, which is probably another reason that they are born so weak. Chimp babies are stronger and are able to hang on to their mothers, but their extra strength and mobility is defined and controlled by their own instincts and the environment (humans would have overwritten this 'unnecessary' code long ago) .

Good point. And that's just another reason why that entire video seems so wrong. Maybe the ragdolls need to learn to move one step at a time as well, instead of flailing all over the place from the start, without any achievements at all in the end really.

Also, as for the 4 legged, if you have ever seen kittens or puppies come splashing out, they learn to move around extremely quick from a slow pace, without beating up their siblings and mother as their legs are(n't) flying around. I would feel pretty bad for any mammal mother if babies did that inside them.

@Jerome

Not necessarily my beliefs, but I think it is better to ask why, instead of taking one answer to heart. Apply that to nazi's and a million other chaotic moments in history, that the 'why' question was more of a taboo.

@gfk

Sorry. I would prefer a PM system actually, instead of adding topic counts for single comments, but that makes too much sense for here.

Ontopic:

Anyone else have any other similar (or alternate) AI movies they can share since this topic is growing plenty of interest?

One promising artificial general intelligence is called Novamente. http://video.google.com/videoplay?docid=574581743856443641

They will offer an animal avatar to Second Life using their AGI system in the near future, and eventually other games.

@Retimer, i agree, that is why the video -seems- wrong, fetuses will do a lot of their twitching around to sub-consciously discover their limbs inside. Also, don't forget instinct and the multitude of complex "getting started" information hard-coded into a creatures brain, something a robot dummy would not have, at least not at that level.

It is not supposed to be impressive in that way, the fact that it does not stand up and walk is not the point. Most here are forgetting that it is doing this with no pre-written AI or 'instinct' whatsoever. They are simply 'dumb' random state systems that have been designed to help a neural net with (eventually) long term memory to learn. Although it already has a limited memory and very limited learning capacity, as they say in the article, the real thing has not been built yet.