I agree with you that the grammar system it is not sufficient to obtain the meaning but it is a necessary filter.
About your opinion on AI, that's where we take separate roads. I do not consider anything else short of what you call Strong AI, AI - but that's just my opinion.
You mentioned Pac Man (great game, played it, enjoyed it) - how can you consider a few procedures (no matter how adaptive they are) AI?
Personally i think we slap a way to big name on some pieces of software. Could be a marketing issue?!
But i do get your point and i will refer from now on to it as Strong AI.

Ok. Again insisting on the memory issue and rephrasing - You cannot have STRONG AI without a good Memory System.
Here's my view on how memory for STRONG AI should be structured.
I will refer only to Semantic Memory (declarative memory) - the memory of meanings and concepts.
I propose to you a multilayered structure (or a multilayered network, 3d network)
Imagine a network (although i find a network to be somewhat rigid) where each node is a word representing a concept (so far absolutely nothing new, in fact i'm quoting from TLC). On this layer we can have pointers to other nodes of the structure based on similarities although i do not imagine having many links on this layer.
On top of this layer we put another layer (or another network)- let's say a layer for a certain set of properties like colour. Now we have a lot more links on this layer. The idea is that the higher you get on these layers the higher the abstraction level becomes. Let me exemplify:
You have "bird", "parrot","plane" and "wheel" on the first layer - we could have a link on this layer between "bird" and "parrot" but it is not really necessary.
Let's say we have a layer for color - a "bird" could have any color so it would be linked to "parrot", "plane" and "wheel" (because they have colors too). The others are not linked at all.
Now let's go higher, somewhere on an upper layer we have flying properties - evidently "bird" , "parrot" and "plane" are interlinked. But what about the "wheel" - planes have wheels, so wheels fly too (so to speak). This is where things become multi-dimensional, we have links between layers.
Now, why do i think a multilayered system is necessary. It would help computers or AI Systems understand allegories or comparisons , it would help AI understand conversations (by knowing the accessed layers mentioned earlier it would be able to determine the abstraction level of the conversation - for example it would sense the difference between a talk about chicken as food or culinary interest or a talk about chicken that is an allegory for escaping prison). Now this links could be infinite and the software would spend eternity in creating them. To solve this issue we could limit the number of links to be created. Some of them would have to be temporary (TTL comes to mind - the allegory mentioned earlier it is quite far fetched, so that's a rather useless information outside that particular conversation).
That's it for today, i'm sorry that i did not gave you the promised game example but this could not be skipped. And I know that you all think that the direction that i'm heading is a CHATBOT, but trust me it is not. Have a nice weekend!

Hi Jackal,

Have you looked at the work of Geoffrey Hinton? He has done some amazing work surrounding what you have described. There is an excellent Google Tech Talk with him over on YouTube. You might be surprised at the current state of the art when it comes to auto associative memory and unsupervised learning.

This forum probably isn't the right one for these sorts of discussions. Game AI is usually restricted to path planning and goal assignment. That said, if you have some screen shots of you work, or some other results to share, I bet a lot of people would be happy to see it.

Seriously, check out Geoffery Hintons work. And if you're not already familiar with it, have a look at using PCA for 'recollection'.

Will

[...] I do not consider anything else short of what you call Strong AI, AI - but that's just my opinion.
You mentioned Pac Man (great game, played it, enjoyed it) - how can you consider a few procedures (no matter how adaptive they are) AI?

You definitely didn't pick the right forum to talk about this. AI in the context of games is precisely about how to program the movements of the ghosts in Pac Man, and not about strong AI. You may think you have great ideas on how to achieve strong AI, but we just don't care.

And that's all folks.
Thank you very much for guidance - some of you gave me some good advice. I guess i should have asked first: Am I in the right place? ( I choose this forum because of high rankings on google - 'Damn you google!' )
Thread can be pretty much closed.