But let's say you want to analyze a current state, and you give your coder 0.7344, how will it maintain knowlege of the parts that make up that?

That's where big data would come into play.

This way you know .023433 is equal to the string "John was here,for a while;,But,He left,shortly,after Judy arrived.txt" as well as see'ing what other parts can be added to that statement. You want to build a collection of information, so as part of that, you'd had data for each word in that sentence, so looking through items related to John, you might have his height/weight/gender/how he looks/what he's wearing etc. Same thing with "here" "a while" etc.

I'm 90% certain you're looking for a big data/data learning implementation.

You define ".023433" as "John was here,for a while;,But,He left,shortly,after Judy arrived."
then use the ".023433" in other sentences, which means expanding the sentence with the value of the definition.

You can stack such definitions, eg define "0.7344" as "I said .023433.txt"
(I think?)


Of course it is possible to do this, but why do you want to do that? What do you gain here?

Yeah, the latest long post started well (explained what everything means, clearer sentences).

But toward the end there was less clear explanations, and again some things were not explained at all, so it was hard to fully understand.

Just spend more time explaining things and thinking how to write clear sentences like there in the beginning of that post and then its easier to understand.

I didnt understand what the "columns" are, or what you mean by the "active column". Also what values you use to calculate the total/gain/loss values, and what the "stat (state?)/weight is for (I understand that the 'code' is arbitrary, but is the weight arbitrary too?).

My understanding of the concept thus far:

-Information can be encoded as an 'address'/'code' (like hashing)

-Information can either be 'raw' information, OR a combination of raw information AND encoded information (codes) like "My [1234] is [3414]" (1234="name", 3414="bob the destroyer")

-Somehow, given limited or ambiguous information, you can 'fill in' what is missing by subtracting the code of known information (or the context of the info) from the 'total'?

(I didnt understand that, because in order to do it, you must ALREADY know the value of 'total'?)

You can upload images here (I think?) or insert image from some image upload service using a URL (I tend to use imgur.com) so it shows here. So if you want to use images you can just draw one, upload it somewhere and then put it in the post using the 'Image' button.

EDIT:

Also, we have the tools and knowledge to do things like this already. The problem is that a useful AI is not simple. It has many components. They must all interact. Each must be carefully adjusted AND taught with tons of information or it wont work well. Even with optimal usage of resources (good at separating relevant information from not relevant), it will still take huge amount of computing power and memory.

you know .023433 is "John was here,for a while;,But,He left,shortly,after Judy arrived

The problem with that is that the computer will have to map the number to the string, or there is no way it can make that relation.

Unluckily, there exists no function to derive a message like "John was here [another 30-40 characters]" from a 6-digit number alone.

You know, I really wish this was possible because it would mean I would get the Nobel Prize for proving the pigeonhole principle wrong and writing the first software that can compress a file of any size down to one byte. Imagine an internet where bandwidth doesn't matter any more, and imagine how upset hard disk manufacturers would get.

(In fact, I have already written the compression program, it's the decompression part where I'm having trouble!)

Another problem is that If you hash "John was here" and "Johnny was here" as well as "John has been here", and you expect to get the same or a similar hash value, you will find that this is generally not the case. Not even remotely.

(Not unless you dive into context triggered piecewise hashes, which is an entirely different beast)

But if I see several bread receipts with only flour as common ingredient.
I can reason to call them "flour bread receipts"

You can, but you have a lot of world knowledge.
For example, you know flour is for food, and receipts are for food.

A computer does not know what food is. It does not know what a receipt is, or what it is used for (or even that it has a use!).

It may select "writing" (all receipts are written), or "blue" (eg they all use blue ink), "letters", or "words" or "bake" or "paper" (they are all written on paper), or thickness of paper, or size of paper.

If one receipt write "fluor" (typo, but a valid word), the receipts don't have flour in common.



For a computer it's like I give you 3 texts in a language you don't know. I ask you to find common words. You can do that, just compare the characters carefully.
You find 3 words after a search. Does that mean you understand the texts? Do you understand the purpose of the texts?

No, I would say, you just have 3 words without understanding what they mean beyond that they are common in the texts I gave you.


We understand "table" not as comparing thousands of tables and noticing they all have a flat surface and some legs.
We understand table as a thing you can put a plate on, sit on a chair with legs under the table, and have dinner.
Such function and understanding cannot be derived from just physical form "flat surface and some legs" of the table.

I.ll get back to all soon.My wife just had an operation on her arm,so I'm tied up at the moment

But the idea is to use this to get to a point,to teach the machine to learn to teach and learn by itself.

Yes, a big part of the world wants that.

But as far as I can see, you are only storing known solutions into the computer, and then have it reproduce them (which works, computers are very good at remembering).

It seems to me you want http://en.wikipedia.org/wiki/Pattern_recognition