Without context: If you were typing: “what is her nam”. I’m able to predict to a high decree of accuracy that you meant name. But lets say you mess up and put “what your naeme” You going back and deleting “e” to correctly spell the word is a pattern. This pattern can be mapped into the AI.

Again, this is how google works. It also takes advantage of users correcting themselves to recognize misspelled words in context (ie, with regards to the entire input and other recent input).

Mark is right that without changing a single algorithm, google can learn to correct “Howesedfaboutttyrlnowfrt” and similarly styled input. Call it database retrieval or recognizing how often the concept “Howesedfaboutttyrlnowfrt” is found with “how about now”, but it’s the same behavior in practice.

As far as future word prediction goes. Its based solely on knowing the context of the current writing or recognizing a series of input.

How will your system have a sense of context? Do you mean simply which other words appear (and in what order)? How is that different from recognizing a series of input?

Lets say the AI got inputs of “jumped, wanted, etc” it will recognize the pattern of the added “ed” and map it as a concept. Next time the input occurs, lets say it received input it was not familiar to like “amused”. The concept of “ed” activates.

Careful, without guidance this can run away from you. School children are explicitly taught common prefixes and suffixes. And they have the benefit of hearing words pronounced and receiving non-verbal clues. Left to its own devices, a thinking bot could be left wondering about svelte monarchs.

Not quite, the memory component of the system is not a typical storage database. It won’t store all combinations of letters. It will only store 26 objects, one being each letter of the alphabet. The combinations are links that connect to other objects its related to. So a link can be generated for the object “w” that connects to the objects “o”, “r” and “d”.

Andrew’s right, it sounds like you’ve never dealt with large data storage/retrieval before. (Hey, I haven’t either, but you should heed the warnings of those who have.) How much space will it take to store your “links” (pointers, or however you intend to do this)? How much space to store the number of times a letter occurs before another? After? Three letters before? Three letters before if the letter “b” is inbetween?

How about all those relationships for the “concept” of “wo”?

So on and so forth.

The memory doesn’t deal with chance or statistics, its deals solely on how strong a concept and how close it is to other concepts and object. Chance and statistics is of itself a modeled concept in the brain, which is learned.

If you are interested in how many times concept A is found with B, and you have some reference for what counts as “often”, then you are dealing in probabilities. And although our understanding of statistics is learned, the way that dendrites build connections can still be statistical in nature.

C R Hunt - Oct 13, 2011:

Again, this is how google works. It also takes advantage of users correcting themselves to recognize misspelled words in context (ie, with regards to the entire input and other recent input).

Mark is right that without changing a single algorithm, google can learn to correct “Howesedfaboutttyrlnowfrt” and similarly styled input. Call it database retrieval or recognizing how often the concept “Howesedfaboutttyrlnowfrt” is found with “how about now”, but it’s the same behavior in practice.

There are two ways to accomplish a problem. One way is through generality, the other is through brute force. It may be the same behavior but different capability and scalability. Google consists of different types of algorithms that is geared solely on accomplishing ONE main goal. Google cannot be extended more than its programmed base. Its a form of narrow AI. I on the other hand is looking towards generality, not an AI that is programmed for one specific goal. Its like an AI that can play only one game while another can plays multi-games. They are exhibiting the same behavior outwardly, but the different is. One was built to be general and can extend to varieties of games while one was built for one sole task.

While the behavior may look the same, its quite different. My algorithm is geared towards recognizing all kinds of patterns while google is focused on specific ones. Secondly, my system is looking to understand the patterns of its input, while Google is geared towards using specific pattern for specific task.

The patterns my system finds and maps is intended to be used generally in every possible situation that it sees fit. While google’s is being used for specific tasks like spelling checks, grammar. etc…

I have one algorithm with layers look to find all patterns, google has one algorithm to accomplish a specific goal.

C R Hunt - Oct 13, 2011:

or recognizing how often the concept “Howesedfaboutttyrlnowfrt” is found with “how about now”

Well my system isn’t intended to operate like that. It will accomplish this task the same way you do it. Recognize the legal words in that input and also recognize the gibberish. Seperate the intelligent words from the gibberish (How | esedf | about| ttyrl| now| frt) and tada.

same with “howyxareldyiu?”

C R Hunt - Oct 13, 2011:

How will your system have a sense of context? Do you mean simply which other words appear (and in what order)? How is that different from recognizing a series of input?

“With great power comes great responsibility” is what I meant by series of input.
its different because using “simply which other words appear (and in what order)” may recognize this statement to be related with the above statement “With much power comes much responsibility”

While in the case of a series of input, we are looking for the exact same input in the exact order. When we hear familiar sayings, our brain activate the rest of these series of input.

For example: “Ask Not What Your Country Can…”

Your brain just activated the rest of that data input in your brain.

C R Hunt - Oct 13, 2011:

Careful, without guidance this can run away from you. School children are explicitly taught common prefixes and suffixes. And they have the benefit of hearing words pronounced and receiving non-verbal clues. Left to its own devices, a thinking bot could be left wondering about svelte monarchs.

I’m not really worried about any thing but basic basic grammar when it comes to phase one.

C R Hunt - Oct 13, 2011:

Andrew’s right, it sounds like you’ve never dealt with large data storage/retrieval before. (Hey, I haven’t either, but you should heed the warnings of those who have.) How much space will it take to store your “links” (pointers, or however you intend to do this)? How much space to store the number of times a letter occurs before another? After? Three letters before? Three letters before if the letter “b” is inbetween?

Andrew is far from being right. I’m sorry but if I wanted to take advice from someone, its definitely not someone with the same personality as Mr.Smith. The amount of negativity in his post and the fact he can’t make a post addressed to me without spewing venom is grounds for him to belong in my “ignore list” if the forum even has that capability.

I welcome all constructive criticisms that is actually geared to help me or to further the discussion. None of the posts Andrew has made specific to the topic has any substance.

The fact that I even respond to him is because I’m trying to be polite. But Mr. Smith can continue doing what he’s does, it seems to be his personality, more power to him. But lets not expect that I will continue responding to his low quality posts.

Anyway, yes I have dealt with extremely large data storage.

C R Hunt - Oct 13, 2011:

“How much space to store the number of times a letter occurs before another? After? Three letters before? Three letters before if the letter “b” is inbetween?”

There is a misunderstanding here, I’m not storing any of that.

C R Hunt - Oct 13, 2011:

How about all those relationships for the “concept” of “wo”?
So on and so forth.

I suppose you meant “who” what relationships are you talking about? I never outlined any. I did say that words that occur together will be brought closer together, there is nonadditional needed memory space for that.

C R Hunt - Oct 13, 2011:

If you are interested in how many times concept A is found with B, and you have some reference for what counts as “often”, then you are dealing in probabilities. And although our understanding of statistics is learned, the way that dendrites build connections can still be statistical in nature.

Alright.

I think we hear all the time how “google” or other corporations has a bunch of computer scientists conjuring up new technologies and ideas in their secret classified lab.

But look again. That’s not the case. I think its pretty much firm that 80% of googles products does not belong to them. But were purchases thanks to a large sum of green. I used to respect google till I saw their list of no-novation. No genuine inspiration, no pure creation.

http://en.wikipedia.org/wiki/List_of_acquisitions_by_Google

Here’s a highlight

Oh boy its the famous Google +... wait that belongs in the “Fridge”
Android did someone else’s chores
IGoogle didn’t first belong to Google.
Blogger was blogging somewhere else.
Google Search didn’t have Google’s name on it
Gmail without the G
Adsense: a-couple of digits in-front of a million makes sense to me too.
Google Docs: worked in a different hospital
Youtube: well we knew that one.
Google Talk: Google wasn’t the first one talking.
Google Maps and Google Earth: Earth to who? goo….no.

Want to make a driverless car google? well hire the people who won the DARPA grand challenge.

So the saying “how can you do something when ....has a-couple of computer scientist working on it and haven’t succeeded.” should be out-lawed.

And yes Google improves of ideas. Just like any other company. But its much easier to improve on something than to inspire it. Surely if someone handed me the source code of An AGI on a dvd. It will be very convenient.

Genesis - Oct 12, 2011:

Carl B - Oct 12, 2011:

Genesis, if you do not mind me asking- what is your relationship, if any, with Rhonda Software?

None. I just find their algorithms fascinating. Though they are not doing anything really special. There are variants of these type of algorithms that are openly available.

Bummer- since you linked to those videos by Rhonda Software I had a glimmer of hope for a moment that you actually had some demonstrated tangible skills in programming, image processing, or… anything related to the project you are trying to describe. BTW, DO you have any formal training in programming, engineering, or linguistics, ‘cause it sure would help with your Cred here?

Genesis - Oct 13, 2011:

I think we hear all the time how “google” or other corporations has a bunch of computer scientists conjuring up new technologies and ideas in their secret classified lab.

But look again. That’s not the case. I think its pretty much firm that 80% of googles products does not belong to them. But were purchases thanks to a large sum of green. I used to respect google till I saw their list of no-novation. No genuine inspiration, no pure creation.

...

So the saying “how can you do something when ....has a-couple of computer scientist working on it and haven’t succeeded.” should be out-lawed.

And yes Google improves of ideas. Just like any other company. But its much easier to improve on something than to inspire it. Surely if someone handed me the source code of An AGI on a dvd. It will be very convenient.

You can’t be serious can you? Google acquires the TOP talent (via direct hiring and whole company/technology buyout) it thinks it needs to succeed in it’s goals.  You could learn something from them I believe. As a wise person once said- “Luck favors the prepared.”

Carl B - Oct 13, 2011:

You can’t be serious can you? Google acquires the TOP talent (via direct hiring and whole company/technology buyout) it thinks it needs to succeed in it’s goals.  You could learn something from them I believe. As a wise person once said- “Luck favors the prepared.”

My sole point is that its the regular individuals like you and I that inspire new technologies, not the huge corporations that has their name branded on it. There is no secret group of computer scientists in a classified covert lab in Google or in any other corporation for that matter that is genuinely inspiring new technologies.

Its usually an individual who does create something and his work is bought out for a large sum of money and then the corporation introduces his/her technology into their products and every one bows down “oh great Google, you outdone yourself this time!” (Ala Google plus.) But all it was is a previous social network called “Fridge” that Google bought.

So we should stop looking to corporations for ideas and start looking to ourselves. If there is a new breakthrough, it will come from us not them. Imagine if someone discovers AGI, the amount of money Google and the likes will offer them? Un-imaginable.

Personally if it was me, I would tell them, their money perish with them.

Carl B - Oct 13, 2011:

Bummer- since you linked to those videos by Rhonda Software I had a glimmer of hope for a moment that you actually had some demonstrated tangible skills in programming, image processing, or… anything related to the project you are trying to describe. BTW, DO you have any formal training in programming, engineering, or linguistics, ‘cause it sure would help with your Cred here?

2nd year of CIS. Nothing other than that. I’m not here to establish a cred.

All I’m doing is updating the thread when I make progress and discussing with whomever that would like to discuss the concept. I’m not putting a gun to your head to participate in it.

> I did say that words that occur together will be brought closer together, there is nonadditional needed memory space for that.

Euhm, what? That must be the first time I heard someone use physical location of memory as a property

But serious, how will you do this? You can’t move pieces of data around on the disk to be closer to other pieces, but I assume you don’t actually mean this.

So, you must mean you’ll bring the words (or letters) virtually closer together. This means that a) you store the `location’ of each word as a property of that word, or b) you have an extra parameter that stores the distance between two words. In the first case you are right, moving items closer together does not increase memory use. However, if you move it closer to another word you also move it closer to other words and further away from some words, and this can’t be good.

In the second case (b), you WILL need extra storage for such a link between words. Such a link uses memory, and if we assume that English contains 200,000 words (not including misspelled words), you’ll have to store every link between each word. There are 200,000! / (2! * (200,000-2)!) = 2*10^10 possible combinations. Assuming that each link contains a pointer (32 bytes) to each word it links, and a double-value for storing the chance, this results in 1,6*10^14 bytes for all the links between all words (that’s 160 TB by the way). And if you are storing N-grams of letters this is even worse.

So, please tell me, how will you store the links between words/letters? If you move them `closer’ together, could you please explain in more (technical?) detail how? Thanks

Not quite, the memory component of the system is not a typical storage database. It won’t store all combinations of letters. It will only store 26 objects, one being each letter of the alphabet. The combinations are links that connect to other objects its related to. So a link can be generated for the object “w” that connects to the objects “o”, “r” and “d”.

Genesis, you might want to take a look at my neural network designer. It works exactly this way. My suggestion for you: better start testing some of your ideas in the real world so you can get a feel of how systems like this behave. Let me reviel a tip of the iceberg: less objects means more links, more objects means less links, both situations behave different, give different results and different timing.
I somehow picked up that you seem to think those links don’t have much of a cost to the system. Please test this, I think you might be surprised of the results.

Genesis - Oct 13, 2011:

Well my system isn’t intended to operate like that. It will accomplish this task the same way you do it. Recognize the legal words in that input and also recognize the gibberish. Seperate the intelligent words from the gibberish (How | esedf | about| ttyrl| now| frt) and tada.

same with “howyxareldyiu?”

Ah, but this is exactly what I meant with my example of “thinking” and svelte monarchs. (Seeing the word “thinking” could easily activate the concepts of “thin” and “king”...) My point is that there are subtleties to computer pattern matching vs. the wide variety of data we take into account to do this ourselves. You’d be surprised the crazy directions computers can go when they don’t have enough context or are allowed to direct themselves unguided. (Actually, I’m running into this a bit myself with recent forays into sentence construction recognition and generalization.)

Genesis - Oct 13, 2011:

C R Hunt - Oct 13, 2011:

How about all those relationships for the “concept” of “wo”?

So on and so forth.

I suppose you meant “who” what relationships are you talking about? I never outlined any. I did say that words that occur together will be brought closer together, there is nonadditional needed memory space for that.

My example was referencing your example of the word “word” and the learning method you described in your “who” example:

Genesis - Oct 13, 2011:

Later when an input is streamed to the memory, objects similar to it will activate. For example if I entered “who”. The object “w”, “h” and “o” will activate. each time those three objects activate together, their relational link gets stronger. After a while the objects will pass thresh-hold and a new concept will be born which is the word “who”.

You say that you intend for the program to recognize when letters are often found together, and from there reaching threshholds to group these letter combinations into words and so forth. Yet this is not simply a matter of “activating” letters. Otherwise “how” would be treated exactly like “who”. Spatial relationships are key (as you’ve mentioned). But this is not simply tracking neighboring letters. The letter “o” may be strongly connected to “w” (preceding) and “r” (following), but only when followed by “d” or “ry” or so on. Unless you store these relationships and their number of occurances, they can’t ever cross a threshhold, can they?

In the “wo” example I gave, I was thinking along the lines of building longer string combinations from the ground up. So ” word ” would strengthen ” w”, “wo”, “or”, “rd”, and “d “. Then once those became concepts, “wo” followed by “r” and preceded by ” ” could, and so on. This isn’t the only way to do this. One could also do as I described above and store all occurances and conditions (conditions meaning, what letters occur between two letters to make the connection strong/weak/etc.). But this would quickly grow into a mountain of data and doesn’t seem to be at all necessary for broad pattern recognition.

After all, one must balance the intended goal with the hardware available.

Concerning Andrew, I agree that his tone has been condescending and negative. But his mocking tends to come only in proportion to the level of absurdity of people’s claims. As others have commented as well, it is one thing to discuss ideas, and quite another to project unearned hubris coupled with clearly unrealistic goals. I think if you had struck a different tone—and conveyed a greater understanding of current NLP research—he would have as well. It would be a mistake not to acknowledge and consider some of his good advice:

Andrew Smith - Oct 13, 2011:

You should not confuse engineering with research. When you set out to build something that is well defined, then you can make a plan like that. But when you set out to discover something new, you have to first determine everything that is old. Once you know and understand best practice you will be in a position to improve on it, but until then you will just be blathering randomly. You have to understand the rules before you try to break them.

As far as your data storage claims go, I’m as lost as to what you mean as Mark and Andrew.

Genesis - Oct 13, 2011:

I think we hear all the time how “google” or other corporations has a bunch of computer scientists conjuring up new technologies and ideas in their secret classified lab.

But look again. That’s not the case. I think its pretty much firm that 80% of googles products does not belong to them. But were purchases thanks to a large sum of green. I used to respect google till I saw their list of no-novation. No genuine inspiration, no pure creation.

This is irrelevant. The percentage of work done behind the walls of Google’s offices or behind the walls of another company has nothing to do with my point. The point being: we are talking of large collaborative efforts of many very intelligent people focusing their efforts on specific, hairy problems. I believe this came up particularly in the context of object recognition from images/video. It is laughable to claim that acting alone, you will overcome the limitations their work has encountered, all while concurrently working on various other NLP challenges.

Mark tM - Oct 13, 2011:

> I did say that words that occur together will be brought closer together, there is nonadditional needed memory space for that.

Euhm, what? That must be the first time I heard someone use physical location of memory as a property But serious, how will you do this? You can’t move pieces of data around on the disk to be closer to other pieces, but I assume you don’t actually mean this.

lol, but I don’t think that is even possible since a memory address next to another could be in use by another program. You are right, its not what I meant.

Mark tM - Oct 13, 2011:

So, you must mean you’ll bring the words (or letters) virtually closer together. This means that a) you store the `location’ of each word as a property of that word, or b) you have an extra parameter that stores the distance between two words. In the first case you are right, moving items closer together does not increase memory use. However, if you move it closer to another word you also move it closer to other words and further away from some words, and this can’t be good.

Yes, in page 1. I pointed out that memory will eventually be stored in a 3d space.
if a word is moved and it has words remotely close to it, these words will be pulled along aswell. Think of it as the gravity of the initial word pulling whats close to it.

Mark tM - Oct 13, 2011:

In the second case (b), you WILL need extra storage for such a link between words. Such a link uses memory, and if we assume that English contains 200,000 words (not including misspelled words), you’ll have to store every link between each word. There are 200,000! / (2! * (200,000-2)!) = 2*10^10 possible combinations. Assuming that each link contains a pointer (32 bytes) to each word it links, and a double-value for storing the chance, this results in 1,6*10^14 bytes for all the links between all words (that’s 160 TB by the way). And if you are storing N-grams of letters this is even worse.

So, please tell me, how will you store the links between words/letters? If you move them `closer’ together, could you please explain in more (technical?) detail how? Thanks

This goes back to what makes an AI intelligent? Its been studied that knowing 2,000 english words would give you the ability to comprehend the millions of English texts.

So 23 GB of knowledge? Knowing every word in typical dictionary? say 200,000? I don’t think so.

An average high school graduates is said to only know about 10,000 words. Take a 5-6 year old, they know about 3,000 or so words. Or a toddler that knows about 200 and they are pretty intelligent ain’t?

Could a toddler do better than most chat-bots today? Absolutely!

Since intelligence is not knowledge. It doesn’t matter the amount of words an AI knows. It matters whether or not it can understand the concept of the word. (when an infant hears a word, it associate those words with objects and actions) “Grab that cup” is labeled to a sequence of grabbing cup with hand and griping with fingers.

You can store all the words in the English language and still lack intelligence. So no, I’m not storing 200,000 words, 100,000 for that matter. Nor do I need to store 10,000 either. 5,000? nope. 1,000? nope. A few hundred words that is intelligently mapped is enough.

An AI doesn’t have to have an English major to converse. Understanding is the key.

Links

Links do not last long. There’s a determined life-span for each link before its forgotten. This life span is determined by what kind test I’m running on the AI. It could be set to 60 seconds, 6 minutes or an hour. Each time that same link occurs, its life span is increased. When a link does not form into a concept before its life span runs out? Its forgotten.

What I mean by moving words closer together is that Concepts that have traits in common: For example, the concept of cup and water. Each time this concept activates together, they are brought closer. Another example is the concept of the sound of a car tire screeching to the concept of car in a particular situation. When you hear the sound (say behind you), you know that a car just went flying by.

Or when you hear it in general, you know someone just lost control.
Thats because the visual concept and the sound concept of that particular situation are so close, they have almost become one. So when one activate, the other does like wise.
So when we hear the screeching sound, we automatically anticipate to see the car that produced it.

What doesn’t happen:

Links do not last forever but for a short time.
All combination of a letter that makes a word is not stored. (that means only “book” is stored not “bkoo” or “koob”)

C R Hunt - Oct 13, 2011:

This is irrelevant. The percentage of work done behind the walls of Google’s offices or behind the walls of another company has nothing to do with my point. The point being: we are talking of large collaborative efforts of many very intelligent people focusing their efforts on specific, hairy problems. I believe this came up particularly in the context of object recognition from images/video. It is laughable to claim that acting alone, you will overcome the limitations their work has encountered, all while concurrently working on various other NLP challenges.

REDACTED COMMENT BELOW:
But, but, but- Google did not create Earth and the universe… God created it all, and HE did it all by himself!  /s

Genisus your problems seem to run much deeper than merely having no idea what you are talking about.

You really need to seek medical advice to rule out mental illness, or obtain appropriate treatment for it, before you start doing something far worse than making a comical nuisance of yourself online.

After pondering through the weekend about this and the design of the soon to be ‘ariadne’ site. I decided to merge phase two into phase one because if I continued in the path that I was heading towards, I would have to reprogram the entire pattern matcher function and probably most of the innate memory abilities to be able scale it up to phase two.

So instead of wasting time on the ledge, let me jump right in.

The issue now is, instead of text I’m now dealing with images. Which complexity is un-imaginable because a simple picture has multiple activities going on at the same time, yet you can only pay close attention (focus) to one of them at a time. Think about your ‘desktop’. Very much complex than a series of ‘text’. It’s very easy to find patterns and sequences of patterns in a text. Maybe not easy, but compared to a image, it is.

But its not just finding patterns. how do you store an image in such a way that you can find patterns in it and in such a way that you can reuse it intelligently? For example, a picture of your desktop.

These are the questions I’ve been trying to map out on paper. I have many ideas that I formulated over the weekend and will attempt to put them into code next weekend. Hopefully all goes well.

Well, I’ve given up trying to make you see the problems of your design, so all I can say is: go for it, keep us posted, and hopefully you can show us a working example soon

Mark ter Maat - Oct 17, 2011:

Well, I’ve given up trying to make you see the problems of your design, so all I can say is: go for it, keep us posted, and hopefully you can show us a working example soon

Well its normal to see a problem in something that you don’t possess full knowledge about. I haven’t even talked about 50% of the theory. I have only given the foundational basics. Anything I don’t explain further is because I don’t want to talk about it.

If I were to make my project open source then I will probably have to write out a 20 page PDF to even contain the full scope of the theory.

Anyway. Parts of the theory works (my test with texts). But even getting to the point of a working demonstration example for the public won’t be soon. (Not that I won’t be making progress till then, its that showcasing the progress in a youtube vid for example, would unveil whats actually happening behind the scenes. I can only describe the progress per-say” till the AI is to the point where I can port it to a real-world application)  One of the only ways I can probably demonstrate its full abilities is in the form of something like a chat-bot (or what I call a conversational talk engine).

But you have to understand that “talking” in itself is also a learned thing. We have chat-bots with built in feature to talk, we program into it how to talk, when to talk and what to say. But in real life, a baby has to learn “how to talk, when to talk and what to say” from its surrounding society.

But you don’t need 200,000 words plugged into its DB. You just need it to learn a few ground words.

I was checking out http://openmind.media.mit.edu/ the other day. It has over 1,000,000 statements and searching for “dogs” give you a good list of statements. But none of the words in those statements are grounded, therefore the whole million of them are quite useless. (unless they port to a system with grounded concepts of some of those words.)