[Ultron]

Good thinking Patrick! I didn't think of using theater plays. Maybe I could even use movie scripts. Thanks for the idea!

And yes voice recognition is (as I said) terrible at the moment.

[Freddy]

I'm glad you found ideas, but please now let this get back to discussion of Ivan's project fellas.

Thanks

[ivan.moony]

I realized that when verbally interacting with intelligent being, there are only three types of a verbal feed-back:
1. questioning about a system in the Universe
2. describing a system in the Universe (embraces short answers too)
3. describing a procedure to get specific value of some state in the Universe (building algorithms)

Systems in the Universe could be described with Synth programming language. There should be an unsupervised learner from input by genetic algorithm which constructs Synth systems that respond to input.

If we know how to realize each of three behaviors (I think I know this one) and If we know when to respond which way (can be learned from experience as a function from past inputs to the three behavior types) we would be very rich in a sense of knowing how to make intelligence copy machine. But probably there are a lot of low-level traps in the way, so I'll stay on the ground this time.

[Ultron]

What about questioning about a procedure in the Universe? I like to believe there are only two types - questioning and describing a system. A procedure can be the members interacting with each other within the system - thus the process can be described as part of the system.

Unless I didn't quite understand your point.

[ivan.moony]

Nice bug catch Ultron7

I'd put it like this:

• question
  
  • about describing a system
  • about a procedure over a system
• answer
  
  • describing a system
  • a procedure over a system

I think that description of a system has to be differed from a procedure in a system. One system can have numerous states of its values. A procedure then would be getting a system to desired state. Different procedures get a system to different states. It is true that any procedure is implicitly wired within a description of a system, but it is a part of intelligent prediction to get a system from the current state to desired state.

Take a look at i.e. math problems. Tasks describe a system and fires up a question about the system. To solve a task we need to setup the task in math language (system description), and the other part is transforming the system to desired form to get the answer.

Likewise any system can be transformed to reach desired state. Sometimes desired state is known in advance and we question up just a procedure to get the state, but sometimes (i.e. in math) desired state is detected by some general rules (like one rule in a math sequence: put all knowns at the left side and all unknowns at the right side).

[ranch vermin]

so,  when are you going to get the programming out and do it!   The ai concepts are simple but devious when you try to implement them,  ive got no idea why people havent done this already,  idiot robot sweatshops should be as commonplace as an abbotoir.

About the very first flowchart,  and it fits into my dummy (idiot worker) system perfectly->

https://scontent-b-sin.xx.fbcdn.net/hphotos-xfp1/v/t1.0-9/s480x480/10968338_10204866016225209_8330364442040761391_n.jpg?oh=f3aa9eaad9e86fecb11c400de1f2caa2&oe=55963A1E


I could go on for years about it,  but i best keep it to myself,  because if you ask me I can tell you,  but I think your having fun just thinking about it yourself.   
The difference is,  im actually making this thing, so im not looking at it in any general overview,  its all down low and technical.

[ivan.moony]

so,  when are you going to get the programming out and do it!

Yet several times I was near to the point to sit down and put the thing into computer and every time I get stuck with a part of the implementation and I'm being returned back to pen and paper.

This time it is conflict resolving algorithm that I can't map to the intelligence copying algorithm. I figured out how to connect a pattern with a response, but what if the same pattern gets several different responses in the past? Like (how to get a bread) -> (steal it) or (buy it). I know the answer: (if you have money buy it. Otherwise if you have ingredients, bake it), but how to actually implement this? How to firstly recognize conflicting responses? And next, how to blend all answers into one answer?

I'd do it if I'd know how, but I don't, sorry. There are unanswered questions that I need to answer first, hence this thread is growing up. Sometimes it seems to me that I have all clued out, but every time something new hops out from the darkness and puts me back to the ground.

[ranch vermin]

thats an overloading problem,  theres many outputs with the same input.     The basic way to get over that problem is to have a bigger input space, it takes more into account before it makes its decision.

But your system is goal oriented, ive never done that before,  thats like a state searching system with motivation,  ive noticed that your system has motivation from the beginning,  wheras I didnt handle motivation till way later in thinking, because I knew I could get a simulation running thats equally motivated in all directions to do anything.
Your trying to make in my terms, some situational diagnostics system.   good going,  Im making just a simulation at first, I will be getting up to it later tho.


But if you dont bother with it directing its own behaviour,  this is an example of extending input on a chain based simulation->

An easy way to do it in a basic time based filter is to just extend the amount of time taken into account to decide apon an output.

If you do that with so much prior time taken into account,   the problem is the more time you give it,  the more unique it makes the outputs, as well as specifying them,  so it never gets an opportunity to use the output, only if it was the exact same circumstance again,  and the less time you give it, the more it gets to use its outputs but it overloads them more, and uses them too much.

Doing it in a time based filter, you see its never the real reason why to do the action, its always a correlation not a cause,  it has no logic.


I hope you get somewhere with it,  keep thinking,  youll get there, its not that hard to do kindergarten logic.

[Ultron]

Your program needs to aim to do good and ultimately become better. Over time it would learn that stealing a bread means it will get punished therefore it won't become better. So it will discard that option. I think this is a simple concept and shouldn't be too hard to implement in a program however it may pose a bigger challenge to adapt an already existing concept and algorithm.

[Art]

Found this here...

http://www.sciencedaily.com/releases/2015/02/150205141151.htm?utm_source=feedburner&utm_medium=feed&utm_campaign=Feed%3A+sciencedaily%2Fcomputers_math%2Frobotics+%28Robotics+News+--+ScienceDaily%29

[Ultron]

Yeah I read it... Send my regards to the elderly of the future (oh wait, that includes me... damnit).