[ivanv]

How is the knowledge going to be collected and mapped? (ex; learning.)

i thought that input set changed by input devices could be monitored. rules could be inducted by statistical analyzes. not certain about concrete implementation, but mother-rule could be: happenings that happen together are consequences each to other.

in analyzing console input by testing different combinations, letters that are seen together could be considered as words. words that are often seen together could form some sets (a grammatic predicate can be a flag for grouping other words). the key could be in repeating patterns (every sentence has the same elements, some of elements more or less described). this is behaviour that should be reached by some general rules of induction from any input set (sense of ai system).

universe could be considered as blackbox that inputs states to ai system (sensors) and recieve ai system actions (ai output). now we need to examine what are rules respected by this blackbox. connections between output and input should be concluded. every input set  should be analyzed in the same way: to see what sequences are repeated through time and what are connections between these sequences. we have to detect - is some sequence always following some other sequence. then comes grouping of sequences as elements of the same set. then comes analyzing sets as types of sequences to see regularities that are always true, no matter what element is observed. regularities should be guessed more or less randomly. i believe that by inductive learning from input sensors a knowledge base could be formed, describing things that happened in ai sensors. and when kb is formed, when we know what input from senors will bee read after some output, it's easy to calculate what sequence of output device states needs to be applied to achieve some goal.

i'm not certain about concrete rules of induction, but i think the base one is repeating of the same happening after the same action. i don't know, maybe a set of elements could be formed by grouping different happenings that follow the same action. yes, it is another problem, it needs some workout...

[ivanv]

hi there friends
what do you thing of this compromise finder (already published under ethic topic): https://sites.google.com/site/ivanvodisek/

a thought crossed my mind. every thing that happens in the universe has some reason that made it it happen. either it's a wish from a living being, either it's some universe hard-coded rule like gravity amount depending on some parameters. so each action is taken to keep some rules holding.

i was thinking: how about to obey *every* rule recognized in universe (with restriction of making compromises mentioned above)? if it's wish, it should be fulfilled. if it's universe's rule, nothing bad would happen, it's gonna be obeyed anyway. so the problem would be in finding compromises between universe's rules and everyones wishes.

is this sounds as a friendly ai that wouldn't "kill a girl that works in market to get a can of ananas"? i'm scared somehow...
anyone?

[Freddy]

How long do you think it would take to model the rules in the universe ?

[ivanv]

i don't know. i guess the system would generalize rules from specific examples of input. so it depends of experienced input. can ever input be totally generalized, or there will always be new unpredictable kinds of input? i bet it's later one.
now a problem is laying on the way: setting up a proper input sensor and rules that generalize that input and store it to kb.

[Freddy]

When you say the universe are we talking right down to the molecular level too ?  I mean laws of Physics and so on, or do you mean an average persons view of the world ?

[ivanv]

both, if possible

[Freddy]

'If possible' seems the right thing to say 

I think it would be difficult given that they can't even model weather patterns that accurately, so as for the universe....well....

That's why I am more interested in AI in simulated environments because at least that will happen in my life time 

Noble cause though, every thing starts somewhere

[ivanv]

rules of the universe. maybe it's a fractal function that goes down and up to infinity, and maybe that function is alterable by living beings. let's see, what should function look like to get a spaceship full of spacegirls land to my houseyard 

[ivanv]

i was thinking: how about to obey *every* rule recognized in universe (with restriction of making compromises mentioned above)?

i've got a good news and a bad news.

the good news is: if we obey all rules (or their compromises) we can't go wrong. at the worst case we'll do nothing (if there's no compromise). so this is safe.

but the bad news frightens me: if we put a logic operator "NOT" to intersection of all sane rules, we get an criminal who lives to destroy all good in the world. so one error in programming and... we're doomed

but, in its old fashion, the Universe is reaching to us: with little information, a machine can't be seriously harmful, so we have little time at the beginning of its "life" to decide whether we'll going to let it be. risky, as always. is it worth of risk, as at opposite side, we have a world with no suffer, hunger and criminalities?
i'm for it. 

[ivanv]

just another site about true artificial intelligence.
short intro to type theory and automatic rules and algorithms construction.
about 4-5 pages.
i thought someone might be interested:
https://sites.google.com/site/typetheoryrefined/

[infurl]

Thanks for posting that link ivanv, it was among the most fascinating 5 pages that I've read in a while. Type theory (and typed feature structures) are slowly emerging as a very promising area of research. For example the English Resource Grammar is the most complete and comprehensive grammar of the english language that is freely available, and it is all based on types. There are also grammars available for other languages based on the same methodology but I don't think any of them are free.

http://www.delph-in.net/erg/

[ivanv]

Thank you, Infurl