[djchapm]

Greetings - I'm a developer who was fascinated by my AI classes (Masters in CS in ~2002) and has wanted to pursue the dream ever since.  Recently started taking classes related to the big data bandwagon and current Machine Learning pursuits.  Everything I've learned fails to satisfy me - all the ML these days is to me the fake, train, regurgitate style of neural nets and statistics.  Thus the dream!  I joined this forum because of ivan.moony's post in a response to How is AI Made, How does the Child Learn, it should be simple:
"You think, you think, you think.... and then you die of oldness. And then someone comes after you and thinks, and thinks and thinks and then he dies also of oldness. So where does it end? Must be something with undecidability of "halting problem". We are all doomed to failure, whoever is into AI. Like a lottery, a zillion of us try, but then one finally succeeds. I live for that day, whoever the lucky one would be."

This is my problem exactly - think/think/think, and with all my programming experience and basic classical solutions on machine learning - I can't come up with anything!  My latest thoughts are we're trying to stuff computers into being human, but what would the computer think if it could think.  Probably wouldn't understand words or commands or associations until it learned on its own... So what is the most basic form of built-in evolution a computer can make on its own.  I guess you need to code in some instincts to start with... and there goes my thought train down the thinking path until I die as Moony says!! 

Love some of the posts here - especially the thoughts on thought from Zero.  Hope to come up with that most basic solution some day and change the planet, wowing everyone with how well my new robot can entertain itself and be no use to anyone some day

Cheers!

[ivan.moony]

Hi Djchapm

I actually solved the question of instinct and it's waiting to be programmed, after knowledge base coding. It is about copying human behavior:

* Computer writes A and human responds B. Then computer calculates (genetic programming) possible function A -> B, regarding to all variables in A and B.
* Now human writes A with changed in-variables. Computer now uses the function A -> B to write B relevant to variables of the new A. It copies human intelligence.

Just a few more rules for conflict resolution and ethic filtering and I think we are set to go. Now the problem has dropped to artificial language that can describe A, B, A -> B and finding A -> B. I'm coding this language right now.

[djchapm]

Seems too simple!  Where does the AI start?  What does it write first?  And ultimately - is the program thinking?  I agree that strong AI will require the ability for computers to construct tests against the environment.  What instincts do you propose?

DJ

[djchapm]

Ahhh nevermind - I see your post - I'll go there: http://aidreams.co.uk/forum/index.php?topic=8738.0#.VrFt9rI4FaQ

[Art]

Greetings and welcome!

Perhaps we should ask ourselves How did we learn or how does a child learn? Not by finding a pattern and matching it but rather by learning the ABC's, small words, meanings, verbs, adjectives, parts of speech, spelling, grammar, conjugation and so... Of course all of this takes time for the average child but computers can learn at a rapid rate providing the data is correct and the learning process is carefully measured and controlled.
But how do we humans go about delivering all of these aforementioned items to our hopeful A.I.? Upon which platform will it be constructed? How do we provide for continued learning for it (the A.I.)?
There are a few labs trying similar exercise to some of these ideas, even as we read this.
While there are truly very few "learning" bots or A.I., some have made progress to the point of being considered on par with a 2 to 3 year old child.
Is our Force Feeding of huge amounts of data the wrong approach? Perhaps, because unless the A.I. has "understanding" of letters, words, sentences and principles then nothing else matters.
Human Brains are compartmentalized... Sectioned off with various areas responsible for handling different tasks... Quite modular if you will.
The A.I. will likely handle information better if the info were placed / stored in its appropriate compartment. Conflicts could be kept to a minimum.
Just some thoughts for what they're worth. I'm not sure if anyone really has the answer just yet but I'm sure it's coming soon!

[Zero]

Welcome to AIDreams, djchapm.

I think the core of instinct is the "cat protocol"

Code

1: scan until something unusual catches your attention
2: evaluate it
3: if it seems good, get closer (or eat it) and goto 2
4: if it seems bad, get away (or push it) and goto 2
5: if it's irrelevant goto 1

Would you agree?


EDIT: Also, I think AI needs a body and a world to explore. In my opinion, the world could be internet, and the body could be a web browser.


EDIT: Going on... The purpose of life is to make things alive. A cat basically turns cat food into cats. Animals need food. AI needs users who want to host AI on a computer. Anything that can help AI reach this goal should be considered "good" by AI. Do you see where I'm going?

[djchapm]

Started new topic in AI discussion to carry this on: http://aidreams.co.uk/forum/index.php?topic=10055.0#.VscvsJMrJ-U