artificial intelligence of a 4 year old made!

jverne

jverne

Newbie
Joined
Aug 6, 2004
Messages
4,302
Reaction score
0
I thought we wanted less kids, not MORE!
 
If it's AI, will he be able to learn and have the AI of a five year old?
 
I saw this on slashdot. I wonder if it could pass the Turing test...
 
We've got to start somewhere.

At first I didn't believe it, but the article says it's running on a 100 teraflop supercomputer, which sounds about right for an AI of that magnitude, considering a modern PC can create a realistic AI like that of a cockroach.

We'll see how it pans out. If its running on an adaptable neural network, it may be capable of learning. If that's the case, it will get better over time.
 
"He is able to possess memories, believe things, want things and remember."

Okay, that is either just bullshit or kinda scary and really feckin' cool at the same time.
 
Here it is performing, (and failing!) The famous bear-in-a-box test:
http://www.youtube.com/watch?v=90z2gM8k9CU

This is exactly what a four-year old would do, which is pretty disappointing actually. It means that Eddie doesn't yet have a "theory of mind", and cannot empathize. It means that they have essentially created an autistic AI program.
 
theotherguy said:
It is, but he is asking where Micah, who doesn't know that the bear is in box B, will look for the bear. So the correct answer is box A.
Click to expand...

I must have missed that. Tiny subtitles. He will probably evolve tho. He might think like a 4 y.o. now, but imagine what he will be like in ten years time.
 
Shakermaker said:
I must have missed that. Tiny subtitles. He will probably evolve tho. He might think like a 4 y.o. now, but imagine what he will be like in ten years time.
Click to expand...

You mean when his mighty metallic minions have enslaved all of mankind? Well, I for one look forward to it! All hail our future leader Eddie! Hail Eddie! Hail Eddie!
 
Slug Advisor said:
I, for one, will welcome our four year old overlord.
Click to expand...

This. Except he will become self-aware and start upgrading his processing power, eventually turning the entire universe into a computer. HE WILL BE GOD!

Or he'll just run around in Second Life. Either way.
 
i guess computers learn faster, so i'm guessing in 10 years time he'll probably reinvent what humanity did in the past 1000 years. if he is properly upgraded that is.

anyway, things are getting hot
 
Not exactly DATA but i love things like this so i'm pretty intrigued.
 
Baby Robots D: !

[youtube]iaRAf1PLRjo&NR=1[/youtube]
 
There are millions of complexities, even in the mind of a new born baby. I highly doubt that they've come anything close to simulating a 4-year old.
 
That box test is kind of odd, how would a four year old know that Micah should be looking in box A? I don't think 4 year olds are that smart. Hell, it even would have tricked me I think.
 
I'll wait until AI is used to create something useful.

All 4 year olds are good for is spoiled food disposal and menial manual labor.
 
Hey, ZT01, how old is that video?
The schedule those JST guys are trying for is interesting.

I wonder what would happen if both teams unify efforts
 
What happened to the artificial intelligences of the 1-3 year-olds? Maybe one day they will unite to form the artificial intelligence of a six-year-old and take over a kindergarten or something.
 
a 4 year old? where did they come up with the age of 4 because my 4 year old son would answer that test correctly ..and even if he didnt it doesnt mean the AI has the reached the level of 4 yr old, that's a huge oversimplification





second life is creepy
 
CptStern said:
a 4 year old? where did they come up with the age of 4 because my 4 year old son would answer that test correctly ..and even if he didnt it doesnt mean the AI has the reached the level of 4 yr old, that's a huge oversimplification





second life is creepy
Click to expand...

Well, I think they did many other developmental psychological tests to determine its "mental age". There are many others, like conservation of liquids, development of schemas, and others which can tell you about what level of reasoning a child has.

It's surprising to me that it understands the questions at all.
 
Vegeta897 said:
That box test is kind of odd, how would a four year old know that Micah should be looking in box A? I don't think 4 year olds are that smart. Hell, it even would have tricked me I think.
Click to expand...

That's the point, most four-year olds slip up and say box B because they have an incomplete "theory of mind". In order to solve the problem, you have to "put yourself in Micah's shoes" and ask what you would do if you were Micah. This requires removing any information that Micah doesn't know from the problem, which most adults can do easily, but which children at about that age struggle with.
 
theotherguy said:
Here it is performing, (and failing!) The famous bear-in-a-box test:
http://www.youtube.com/watch?v=90z2gM8k9CU

This is exactly what a four-year old would do, which is pretty disappointing actually. It means that Eddie doesn't yet have a "theory of mind", and cannot empathize. It means that they have essentially created an autistic AI program.
Click to expand...
What I find interesting about this experiment is that it tests so little...! An average human 4 year old, when asked the same question, would be sufficiently confused by it that it would be hard for us to predict, with any great degree of certainty, whether the child would empathise correctly or not. Suffice to say, any intellect created within Second Life is not going to represent the 'average' intellect for its supposed age set anyway.

You're probably right that they've created an essentially autistic personality, since personal inclinations for empathy and other socially advanced routines are possibly the most behaviorally sophisticated patterns we exhibit as humans; the most subtle manifestations of our self-preservation instinct IMO.

What confuses me is the question of how they know this is a '4 year old' intellect if this is the kind of test they give it? Is it purely based on the amount of teraflops behind it? Can't it reason and explain why it thought the answer was Box B (even if the reasoning was flawed and the explanation stilted)? If we're going to ever create any kind of human-level AI then I think it needs to be tested far more rigorously than is evidenced in this video, and I don't think it would need to be so geared towards Turing-testing at this early stage either. It's not like human-level thought processes are entirely rational.
 
well normal humans have about 100 billion neurons in their brain. if they build a computer that can be as intelligent as humans, it needs atl east the same amount of neuron. BUT QUESTIOn: in future, do we have more neurons? if so , then computers must also have more neurons. and that's impossible, because science must develop faster than human brain. i talked about evolution in another thread, and it's clear humans have allready evolved in great pace. we are already ahead of monkeys (that's if evolution exists at all)

risk: if computers become more intellignet than humans, maybe they can't be shut down anymore? they can fix themselves? will they be made of solid material anymore, or some organic matter that reacts to life and starts to evolve? human race is gonna destroy itself trust me.
 
Laivasse said:
What I find interesting about this experiment is that it tests so little...! An average human 4 year old, when asked the same question, would be sufficiently confused by it that it would be hard for us to predict, with any great degree of certainty, whether the child would empathise correctly or not. Suffice to say, any intellect created within Second Life is not going to represent the 'average' intellect for its supposed age set anyway.

You're probably right that they've created an essentially autistic personality, since personal inclinations for empathy and other socially advanced routines are possibly the most behaviorally sophisticated patterns we exhibit as humans; the most subtle manifestations of our self-preservation instinct IMO.

What confuses me is the question of how they know this is a '4 year old' intellect if this is the kind of test they give it? Is it purely based on the amount of teraflops behind it? Can't it reason and explain why it thought the answer was Box B (even if the reasoning was flawed and the explanation stilted)? If we're going to ever create any kind of human-level AI then I think it needs to be tested far more rigorously than is evidenced in this video, and I don't think it would need to be so geared towards Turing-testing at this early stage either. It's not like human-level thought processes are entirely rational.
Click to expand...

They're obviously going to be performing some more tests, especially those relating to theory of mind, since this is the point of the entire project. The press has been very misleading when reporting this. Their goal isn't to create an AI with the absolute intellect of a four-year old, and they don't claim to have one. The entire project is geared towards creating an AI which will respond with false beliefs to this specific bear-in-a-box test and other theory of mind tests.

This test is the only thing that Eddie can do. Currently. RPI stated that after this test, they looked at the runtime scripts that Eddie was using when answering the bear-in-a-box test, and he now successfully says that the bear is in box A. Whether this means anything as far as other problems that Eddie will have to solve is anyone's guess. The whole point of the project is to gradually teach Eddie to have a theory of mind, and after Eddie has a theory of mind, they can use what they've learned to give other AI programs (like a search engine or an AI program in a game) a theory of mind. This would dramatically improve AI all around.
 
MaSter0fDarKneZz said:
well normal humans have about 100 billion neurons in their brain. if they build a computer that can be as intelligent as humans, it needs atl east the same amount of neuron. BUT QUESTIOn: in future, do we have more neurons? if so , then computers must also have more neurons. and that's impossible, because science must develop faster than human brain.
Click to expand...
Moore's Law suggests that this is a moot point - informational technology is advancing far faster than the human brain, and the human brain, since the inception of Homo Sapiens, has only advanced as much as natural selection has enabled it to do so (ie. we have not evolved, we have only been filtered).
theotherguy said:
*snip*
Click to expand...
Ah I see, that makes more sense to me now.
 
Laivasse said:
Moore's Law suggests that this is a moot point - informational technology is advancing far faster than the human brain, and the human brain, since the inception of Homo Sapiens, has only advanced as much as natural selection has enabled it to do so (ie. we have not evolved, we have only been filtered).
Click to expand...

Exactly. Another major limiting factor on human brain power is the size of the birth canal. Human brains can't be much larger without dramatic changes to human anatomy because babies simply couldn't fit through the birth canal if they had very large brains.

However, genetic engineering coupled with C-section births will make this no longer a challenge.
 
CptStern said:
a 4 year old? where did they come up with the age of 4 because my 4 year old son would answer that test correctly ..and even if he didnt it doesnt mean the AI has the reached the level of 4 yr old, that's a huge oversimplification

second life is creepy
Click to expand...

The test is designed to see if a child is autistic or not.
 
off topic, how did stern get the awesome title of sockmonkey?
 
It's a randomly picked codename for the epicentre of a disasterous event, kind of like "Cloverfield".
 
You must log in or register to reply here.
Back
Top