[spydaz]

Hiya ,
I am spydaz , currently designing a AI brain ...
Mainly focused on understanding what the user is taking about .
Have made loads of AI program's yet they all seem to come to that age old problem ... Doe the computer know what your actually saying or is it just some snazzy predictive program ( tricking you into believing that it knows ) when actually it's just detecting some odd words and some querys ...

After loads of study ....

Loads of different program's such as Hal .. Verbot etc ... And Jarvis (lol)
I have finally figured out how to achieve this goal ..

Grammatically and algorythmicaly ...

Age old question "what is a cat?"
Simple answer "it's an animal" / "somebody's pet" / "somebody's else's dinner"

Thanks for reading this ... Hopefully I will get some insight on this forum ...

Ps: I also have a minor website http://www.spydazweb.co.uk
And some mini tutorials on how to begin building an AI in vb... Slowly I will do the whole series's on YouTube ... But it's very time consuming so patience and I will complete them ....

Interesting video : is the one I've done on plugins and updates !

Enjoy !

I hope this satisfys curiosity's !!

[ranch vermin]

Hey Spydaz,   I check this forum every now and again myself.

The crowd here are fairly experienced, (The athena post is a good talk about logic) Myself... ive been doing these 'video nets' of mine for about 2 years now, since June 2013, after watching Jeff Hawkins talk about his Hierarchical temporal memory system.  It started off from there, but its evolved into its own thing now.

Im in a theory rutt at the moment->
When it comes to making a system logical,  im finding using some data chain predictor, or using the logic paradigm (like NLP),  your still stuck in a recording only situation.  (your ai is more recording than learning)  it needs to be dictated everything exactly,  and it has no creativity. 
I swear forming a logical tree is a bit better tho,  but im yet to work it out how exactly...   Im actually busy on a more low level symbol vision topic.  Just tracking moving surfaces for now, and ive got my hands full.  (Further symbols would be extracted from there, to go into some logical database knowledge collection.)

[spydaz]

Recording information ... Learning knowledge through textual information (speech) .... The response is not too important at this moment except for answering questions .... Determining subject / predicate / object is a single technique effective

 "A cat has four legs" , "cats are animals " , "some people keep cats as pets", "some cats are lions and some cats are tigers"
Then ask what is a cat ....
It's inferred that some cats are pets all cats are animals , all cats have four legs ....

Learning about a cat and inference ... This form of knowledge gathering is learning also .......
In large texts lots of information can be learned ...

Before you can get to here ...

Understanding grammar and predicate logic and inference is key ...
Plus being able to program it....

Edit:
I also think that the creative response would also be along the lines of quoting some combination of subject predicate related information to the current topic or even mention some inference that also fits the current topic ...

People also generally randomly think or are triggered to think many different things in a conversation but only say small portion of what they are thinking ... Unless a direct question is put to them .

I often think what would I say to that input ....
As when we look at statements we often create some random statement like "the sun is hot" but to that what would be the expected reply... "Ok" probably .... So sometimes only acknowledgement is required ...

Visually :- recording an incident ... The breaking the scene down ... Probably starts with detection ... But also the techniques used to create a movie scene "set" lighting, actors, props, actions , dialogues , locations, background and foreground, time line & storyboard ... These would be how I would record a scene for diagnosis and information gain ... Store the information then interrogate and compare ...
But at this time vision is too complex for me ....
Interesting stuff (opencv)

[ranch vermin]

The whole thing has to be extracted from the sensor,  automatically.
Its pretty hard to make one of these things.

say you had a video of a guy beating another man to death.   theres many things to draw from this,  not just "men with angry faces beat up other men in the park"   theres also, picking up combat tips, why possibly he did it, how to get away from this guy if it happened to you, lots of things, tactics being drawn  just by seeing one situation.

Im not saying I know that a symbolic extraction system couldnt do it,  but I cant nail on the exact process,  it seems difficult.

*EDIT*   dont let me get you down if you want to do this,  maybe it is simple im just confusing myself.

[ivan.moony]

hi spydaz

I'm curious, how do you learn new knowledge in your AI? I assume that there should be two main components: 1. recognizing input patterns by already known facts and 2. learning new patterns.

I'm particularly interested in learning new patterns, as that precedes recognizing process.

[ranch vermin]

Ivan.       My system (I can only speak for myself, in this wild west ai stuff.)   actually solves 1 and 2 together.   

It will derive all its binary symbols from the sensor, using an extraction program.   Then it stores it as an assert,   If its already seen the data, its already in it, maybe itll +1 commonness, but learning new patterns is just what its constantly doing... makes no difference if its seen it before or not, its just consuming all of it.

I guess your coming from, a spacially id'ing point of view, where your machine cant recognize things at first,   where im currently at it, theres no id'ing process. (unless you needed a little help, conglomerating things together)  It only knows exactly what you programmed it TO DETECT.  You have to say 'look for 2 pixels one brighter than the other'...  you tell it  'measure how far they are apart'  if it was a chat program you detect if its a question or a story or a statement of knowledge, and then its a matter of continual assertion of these detections,  you detect behaviours,  you give it its entire varience of DIFFERENCE with an average jo program, making it completely impossible it could be thinking.   and then you record in what differences depend on what differences statistically.

Its a long story to get someone to understand my point of view,   but im over recognition, ive seen it mastered already.   Im more interested in a simulation of the sensor in animation.

[8pla.net]

Welcome spydaz,

I am hosting an Artificial Neural Network, on my website. 
Machine learning like this follows the same principles as a brain in biology.

[Art]

Welcome aboard Spydaz (Leroy). It's been quite some time since I've seen your name in the forums. Hope you've been doing well. We've a great group of like-minded (for the most part) friends here so join in when the mood hits.

 

[spydaz]

Ivan hi,
I built a program which extracted patterns from sentences ie ;

(Noun) , (predicate), (noun)
A cat ) , is an , animal
A cat), is a , animal
A car, can be used as a, vehicle.

Extract just between two nouns

Then I decided that I'm still not understanding the relationships so ,

I created category's animal , person, place
Then put the relevant items in the category's and did it again .

Category , is born in a , category
This gave me more diverse information , but the retrieval of the information is still hard to extract ....
Both of these principles had actually been done before as.(Hearst) algorythm and the snowball algorythm ...
But still it's not enough as the information needs to be rated as true or false and false information can be learned . But the gathering of the patterns learned as the full sentence is also saved as a patern
A #noun #predecate #noun while jumping
A cat went onto the mat while jumping

I thought the best way maybe to study the user request much deeper and build decision trees based on the question words who, what why, when ... This helps to decide which predicate is being requested .....

What is noun (is a , defined as)
Noun then fill the missing argument ..... But still hashing that bit out ....
But the lists are useful as a basis of testing and collecting some data as one can tailor the list to capture only specific predicate patterns then tests can be produced if it exists already .....

The sentence logic is quite complexed but start with a basic grammar ie just noun , verb, adjective, adverb then build your sub grammar rules as you can detect them or need to detect them when something does-not fit your grammar ...
It's a learning process ...

[spydaz]

Thanks art , been doing a degree in machine learning/ business intelligence .... Now want to concentrate on linguistics and logic and inference logic ...
Using .net now upgraded the programming styles .... Zabaware gave that initial push ..... Now I don't need to use it as a plugin

So this time I'm building a smart and intelligent learning AI ....
Real AI

I will probably put some of my old bits on my website (vb6)
Thanks for the welcome

[DemonRaven]

Have you checked out the personality forge? It has some of the things you are talking about.  I find articles about how the brain works interesting as well as it gives us a idea on how to program a AI.