You are so likely to succeed in creating a superior AI that I wouldn't be at all surprised if a fighter from the future jumped back to the present in order to eliminate you so that you don't succeed in inadvertently causing the fall of humanity.

I, for one, am keeping my distance from you, sir!

I'm going to try to provide constructive criticism here, hard as it might be.

Many fields that we call AI have seen a lot more progress in the last few years than people generally realize. A few examples off the top of my head:

• Board games: Computers are much better at chess than even the best humans, and they are getting good even at games where a few years ago there was little hope or progress (in particular, go).
• Automatic text translation works pretty well these days, at least for well-written text.
• Google is working on a car that drives itself, and their first prototypes work very well. There have been other remarkable success stories in autonomous vehicles, like the DARPA challenge.Of course you may deride those achievements by labeling them as "not real AI", but that's not a very fair criticism: These are real problems that a few years ago machines couldn't solve, and now they can. Once enough such problems have been solved, we'll have machines that can beat us at any game, speak any language, make rational decisions based on whatever information they have, interact with the physical world... our instinctual theory of mind will kick in and nobody will remember what we meant by "real AI".
  
  You seem to be proposing an alternative line of research that takes more of a top-down approach, identifying fuzzy logic and knowledge representation as the most promising tools. I don't think there's anything new in that approach, and the people that followed similar ideas have little to show for them. I agree that knowledge representation will probably be a part of any machine that can understand the world around us in a way that is useful to us humans, and in particular of a machine that can hold conversations. But I don't think it's going to be the magic bullet you think it will be. My opinion of fuzzy logic is much lower: We already have a well-developed theory that governs the assignment of truth values between 0 and 1 to uncertain facts, and it's called Bayesian probability. I really don't think there is much of value to any other way of thinking about it.
  
  If you want to be able to contribute anything to the field, start by identifying a problem where you can quantify progress (playing Starcraft, folding laundry, recognizing faces...). Then work hard at it until you have something to show the world. If your solution is so general that you can apply it to many other problems, more power to you. Good luck!

You realize that there are literally thousands of people around the world who have been working on this day and night for many years? Sometimes decades? Because that's their job? If you truly have this magnificent breakthrough, I suggest contacting them.

In the mean time, if you aren't willing to post your magnificence here and, instead, are only planning on claiming that you have it, then I'm going to have to suggest you keep it to yourself.

You realize that there are literally thousands of people around the world who have been working on this day and night for many years? Sometimes decades? Because that's their job? If you truly have this magnificent breakthrough, I suggest contacting them.

In the mean time, if you aren't willing to post your magnificence here and, instead, are only planning on claiming that you have it, then I'm going to have to suggest you keep it to yourself.

I'm not saying that my concept is better - i'm saying that it "could" be. So, there is no need to be ironic. And I'm not holding back anything. I have to structure my notes to present them in a somewhat clear way, and also i have to translate them. I have a job, various projects that i'm involved in - so it will take a while. And let's be honest - this is a huge topic - you cannot exhaust it with just a few posts, or even a book.

Let me be clear - I'm not doing this for fame and fortune (i will remain anonymous and nothing will be asked in return). I'm just trying to give some tips (if you like) that maybe someone is willing to implement - Because i will treat the subject only at a theoretical level (hence the title of the post - Concept, Theory, Logic). I don't have real programming skills (although i have some experience programming), that's why i'm trying to work with others.

I would really appreciate if you challenge me with problems regarding AI, and challenge my theories because that's the only way we can move forward.

Thank you!

I would really appreciate if you challenge me with problems regarding AI, and challenge my theories because that's the only way we can move forward.

Thank you!

There is another way.
Take the time you use to write here telling us that you have no time and use that to actually write down your ideas.
As mentioned before, the field of AI is vast. There are thousands of problems you could tackle to improve the state of the art AI of today. If you were confronted with problems one by one, this thread would reach 300 pages soon and that would be madness!
If you were confronted with all the existing problems, loose ends etc. at once, I doubt you would post again.

If you would pick anything in particular that you think your "solution" can solve/improve, tell the world (or at least us) how you think it could be improved and then accept our humble replies with help, advice and criticism, maybe then people would listen, interact and participate.

Just my 0.02 €ent

Here is my chat with a chatbot ( http://testing.turinghub.com/ ):

Robitron P3 v.1.3 © 2009 CyberMecha Studios
Fred is now operational.

>hello
Hey.

>who are you?
Do you sometimes think I am ?

>that's irrelevant
Could there really be some connection ?

>you are an idiot
Don't you have anything better to do?

>could you repeat my last statement
Must not repeat. Must not repeat.

>could you repeat my last statement
Does someone else believe could I repeat your last statement ?


>i have a toy in my left hand and a gun in my right hand
Does that suggest anything else which belongs to you ?

>what do i have in my left hand
Is it important to you that your left hand ?

I have pasted this small conversation (in fact it is no conversation at all) - to prove to you that the first and most important thing in AI is Memory - a way to store and parse (analyze and conceptualize) information.
Human memory and reasoning it is highly linked to language. And we have a highly developed tool to use already - it is called GRAMMAR. What i'm trying to say is that if you want a fully developed AI you should have a fully developed GRAMMAR system in place that would act as a filter for Memorizing and Reasoning. And again it is very well structured (GRAMMAR) and if you implement it the right way you shouldn't have (almost) any problems with "vagaries of human language" - given that we too have trouble sometimes understanding each other.
Everything stated so far it is already theorized in Semantic Memory theories and TLC The Teachable Language Comprehender (in fact none of my theories are new, and I believe that we already have all that it takes to create true AI but we don't know how to put it together - it is like we are having all the wheels and gears and motors but we don't know how to put it together to create a car). I am just underlining that these systems are a MUST if you want to create AI.
Maybe we can debate a little on the small conversation that i had with the chatbot. I'm waiting for your feedback.
Next time i would like to debate with you another system that would be needed for Memory and Memorizing, and to catch your attention i will provide an example to be used in a game. The topic will be OBSERVATION, OCCURRENCE AND PATTERN RECOGNITION

[font="arial, verdana, tahoma, sans-serif"]I have to admit that I am in no way an expert (or even apprentice) in the field of artificial intelligence. Ok, got that one out of the way.


Ok, you found out that a chat bot doesn't represent the lates SOTA of AI power. Fine. Do you want to develop a better chat bot? I don't think so. [/font]Are current AI implementations able to "learn" from other beings? I think so (at least to some extend and in a restricted definition of learning).

From the little knowledge that I have on AIs I think that in current games there are AI implementation that learn from the player to react with better strategies. Those AIs are able to recognize patterns in the player behavior, use data gathered from maps etc to apply new strategies and "improve". It may not be the same as real learning because it is still limited to patterns and behaviors loosely defined by their creators but I still don't see anything new in your approach.


I'm still intrigued where this might lead but it looks to me like an armchair AI developer.

Pac-Man's AI works just fine with no memory.

Therefore, your opinion is wrong.



"AI" does not exist as some grand goal of equating a machine with human behaviour. (There is a notion referred to as "strong AI" which is similar, but different in subtle and very important ways.) AI, especially as practiced in games, is about accomplishing some specific goal with whatever technique makes the most sense. Put bluntly, AI is all about pragmatics: what do I have to do to reach my goal? It isn't a field amenable to highly generalized, sweeping idealism like you're demonstrating here. Just having memory does not make a good AI, and plenty of good AI operates with no memory or learning capacity whatsoever.

In short, I think you've got a very distorted idea of how AI actually works in the real world, and as such your idealistic notions are not even wrong. Your misguided attempts at solving problems are ignoring huge swaths of prior art, existing solutions, and well-trodden ground. For instance, automatic translation systems disproved decades ago that you can interpret natural language based solely on grammatical rules. There have been huge amounts of exploration into natural language processing, and the consensus among experts is that semantics go far deeper than grammars can handle. A fascinating treatise on the topic is Stephen Pinker's The Stuff of Thought, which explores how language conveys meaning far deeper than you can divine by just looking at word orders and suffixes.

It would genuinely behoove you to research how real AI work is done before trying to revolutionize a field you clearly don't understand very well.

I'm not saying that my concept is better - i'm saying that it "could" be.

Anyone have 100 monkeys and a bunch of typewriters?

And I'm not holding back anything. I have to structure my notes to present them in a somewhat clear way, and also i have to translate them. I have a job, various projects that i'm involved in - so it will take a while. And let's be honest - this is a huge topic - you cannot exhaust it with just a few posts, or even a book.

Then post them or post nothing. We don't do "coming attractions" around here.

I'm just trying to give some tips (if you like) that maybe someone is willing to implement

No you're not. You haven't given one tip yet.

Because i will treat the subject only at a theoretical level (hence the title of the post - Concept, Theory, Logic).

So you aren't announcing your magnum opus? What then is the function of this thread? Did you have a question? A point? A statement for us to digest?

I don't have real programming skills (although i have some experience programming), that's why i'm trying to work with others.

This is priceless, by the way.

I would really appreciate if you challenge me with problems regarding AI, and challenge my theories because that's the only way we can move forward.

What theories would you like us to challenge? (Now I'm starting to sound like a chatbot.) I'm not just going to make up random problems for you to solve like some college professor.

I have pasted this small conversation (in fact it is no conversation at all) - to prove to you that the first and most important thing in AI is Memory - a way to store and parse (analyze and conceptualize) information.

You haven't proven anything other than you can slap the snot out of a bad Eliza clone. The chatbots that compete in the annual Turing Test would spin your brain around. If you had any information or research worth sharing, you wouldn't have brought that particular bot to the party. Like I said before, read the books I mentioned and then get back to me. There's no programming in them... only concept, theory, logic, and philosophy... right up your alley, I suppose.

Human memory and reasoning it is highly linked to language. And we have a highly developed tool to use already - it is called GRAMMAR.

We've landed somewhere between Wittgenstein and Chomsky. Old news.

What i'm trying to say is that if you want a fully developed AI you should have a fully developed GRAMMAR system in place that would act as a filter for Memorizing and Reasoning. And again it is very well structured (GRAMMAR) and if you implement it the right way you shouldn't have (almost) any problems with "vagaries of human language" - given that we too have trouble sometimes understanding each other.

Whooooo boy, are you in for a surprise!
In the mean time, answer the following question:

"Where is Dave's nose?"

I am just underlining that these systems are a MUST if you want to create AI.

Depends on what you call AI. Around the game biz, not too many people care at all about chatbots. In fact, if you were to ask 100 game AI developers to list the top 5 AI needs, I'm quite sure that none of the 100 would list "chatbots" or anything similar.

Maybe we can debate a little on the small conversation that i had with the chatbot. I'm waiting for your feedback.

What is there to debate? It sucks. You know it. I know it. Likely the author knows it.

Next time i would like to debate with you another system that would be needed for Memory and Memorizing, and to catch your attention i will provide an example to be used in a game. The topic will be OBSERVATION, OCCURRENCE AND PATTERN RECOGNITION

Now THAT might be interesting.