[Eric]

Hello AI folks,
My name is Eric and I'm a student in Engineering physics in Lund, Sweden. I am very interested in machine learning and I've just started learning about how these things work. I’ve come to some realizations, so far I've understood that when you want to teach a coded program something either you have
1. A data set with input data and result which you run through your program to train it. For example the MNIST Data http://yann.lecun.com/exdb/mnist/base
2. You generate your own data and result by having a simulation environment. Just like the examples we've seen when a program learns how to play a video game.

After realising these two types of ways to train for example a neural network I started to wonder if you could do it differently. I started thinking about a couple of things.
1. Bringing the learning process into the real world, could there be factors that help the learning or would it just be, like in many other experiments, a too complex enviromen with too much data to handle for the learning program.
2. Could we compensate this environment with supervised learning? Could this affect the quality of the input data?

My idea:
Have a robot like the the Robotis OP2 http://en.robotis.com/index/product.php?cate_code=111310
as an agent acting in my home. The inputs will be all the sensoric data it can detect and all it’s movements in form of walking, falling, getting up and so on. As far as I’ve understood I would need something like a cost-function, giving it a rating of how well it has achieved.
I’m thinking of trying to model its neural network to get as close as possible to an infants mind, being entirely impulsdriven (like babies are in their early years). That would mean that it’s objective would change all the time and therefore also the costfunction(?) which would decide the weights of each neuron, right?

I understand this sounds like an extremely difficult task but do you have any inputs with your knowledge of how machine learning actually works?
I’m trying to aim for a more flexible but initially simple system that can modify itself to a greater extent.

Thanks for all the help!

[LOCKSUIT]

Hey there!

So you're looking at this robot: http://www.trossenrobotics.com/shared/images/PImages/Darwin_OP_1.jpg

My suggestion is to go 3D simulation. Darwin cost 12K and my simulated realistic baby already has fingers and toes and a tongue! Infinite battery. Speed up simulation. And many many more big advantages.

I would teach you for free every rope to know, fast, to be able to pick up my software and start working.

[keghn]

  404 solution?: 

      http://yann.lecun.com/exdb/mnist/index.html

[keghn]

 i like unsupervised learning.
 The CNN of lately are supervised detectors.
 Atari DQN are cool and are more in the direction of unsupervised learning and have little temporal memory going for them.

 For making a neural network AI you are going need a temporal memory recording of when the CNN activated. And where it is. GPS is
not really need.

[Eric]

Thanks for your reply LOCKSUIT.

Hey there!

So you're looking at this robot: http://www.trossenrobotics.com/shared/images/PImages/Darwin_OP_1.jpg

Yes, that's what I was thinking. Any thoughts?

My suggestion is to go 3D simulation. Darwin cost 12K and my simulated realistic baby already has fingers and toes and a tongue! Infinite battery. Speed up simulation. And many many more big advantages.

I would teach you for free every rope to know, fast, to be able to pick up my software and start working.

That simulation thing sounds very interesting and impressive if you've managed to simulate a human being in it. Could you please tell me more and if there is a possibility to integrate machine learning in it.

I would love to see my experiments in real life and the developing behaviour of the robot and its neural network. I'm thinking that with the guidance of a real human it might show interesting results.
Thanks again!

[LOCKSUIT]

Yes I am going to and will be able to put advanced AI in my simulated human baby, and interact with it in real-time. What specifically would you like to know?

Here you can watch one of my recorded videos showing it:

[Eric]

Yes I am going to and will be able to put advanced AI in my simulated human baby, and interact with it in real-time. What specifically would you like to know?

Here you can watch one of my recorded videos showing it:

Impressive, it seems like you're more into the 3d modeling than developing the machine learning to simulate behaviour though.
Thanks for showing me and I wish you good luck in the future.

[LOCKSUIT]

I am very into creating the AI code, and have lots of instructions for my employee. I am so into it that I push the greatest lengths to create such a real baby human body and get Blender right on par with UE4. Blender is indeed not just a modeling tool after all. It is giving me multiple advantages over UE4 too.

So, what are you planning on doing? Are you going to get a real robot, Darwin OP2?

[Freddy]

Hi Eric and welcome to the site 

[Zero]

1. Bringing the learning process into the real world, could there be factors that help the learning or would it just be, like in many other experiments, a too complex enviromen with too much data to handle for the learning program.

I think you should train your program against the very environment it will have to deal with, directly. The difficulty lies in the complexity of the real world (there's way too much information to decode the matrix ), and if you use a virtual sandbox, since you simply can't emulate this complexity, your program wouldn't be prepared for the real world, it would just be prepared for the sandbox. So no, real world is not too complex, in my opinion.
The only way is through!

Welcome to AIDream!

[Eric]

I am very into creating the AI code, and have lots of instructions for my employee. I am so into it that I push the greatest lengths to create such a real baby human body and get Blender right on par with UE4. Blender is indeed not just a modeling tool after all. It is giving me multiple advantages over UE4 too.

So, what are you planning on doing? Are you going to get a real robot, Darwin OP2?

I see. It sounds like an interesting approach then.
My main goal is of course the greatest one in machine learning , consciousness. So I'm just trying to come up with different approaches that would help on this quest. When I was younger I read of the idea to model a young human being instead of a complex grown ups mind and in the best case scenario the AI will grow by itself.
Right now I realize that I just have to learn about the basics and fundamentals that already exists in machine learning. When I've understood the concepts I can try tweak them. Putting my ideas, programs, robots to the test in the real world seem like a difficult but good way. We'll see.

[Eric]

Hi Eric and welcome to the site 

Thanks Freddy, this seems like a cool place to discuss AI.
I've just ordered my first book on machine learning, Programming collective intelligence. Any advice on other ways to start this journey?
Thanks!

[Eric]

1. Bringing the learning process into the real world, could there be factors that help the learning or would it just be, like in many other experiments, a too complex enviromen with too much data to handle for the learning program.

I think you should train your program against the very environment it will have to deal with, directly. The difficulty lies in the complexity of the real world (there's way too much information to decode the matrix ), and if you use a virtual sandbox, since you simply can't emulate this complexity, your program wouldn't be prepared for the real world, it would just be prepared for the sandbox. So no, real world is not too complex, in my opinion.
The only way is through!

Welcome to AIDream!

Thanks for the encouragement Zero!
It's a big Machine learning world and I feel a little bit confused on where to start. Any advice?

[Korrelan]

Welcome…

Bringing the learning process into the real world, could there be factors that help the learning

There are many factors that help humans to learn our complex environment, not least being a slowly rising sensory resolution over a period of months/ years.  As the sensory resolution rises so does the internal/ mental understanding/ representations of the world.



If you study human infant development you’ll find loads of examples.  Look at these drawings from various stages of infant development as an example. (Some adults still draw like this lol)



This is also used to give a very rough measure of the infant’s mental development/ level though I'm not convinced of this… too many factors/ variables involved.

I also agree that an AGI has to learn/ develop in/ experience the ‘real’ world. 

 

[keghn]

 I have a complete AGI theory. And the AGI does need to go through a infancy phase. 
l