[danlovy]

Noam Chomsky Has Weighed In On A.I. Where Do You stand?
There is a debate about the path to cognition. Will it be a statistical big data approach or a more biologically directed strategy?

What does this group think?


http://danlovy.com/ai-essays/noam-chomsky-weighed-stand/

[keghn]

 I say yes. A cognitive machine can be made. I have a complete model for a cognitive artificial intelligence.

[danlovy]

Can you post a link?

[keghn]

 Here is a is a good fraction of it:
https://groups.google.com/forum/#!topic/artificial-general-intelligence/f5yCbo3XALE

https://groups.google.com/forum/#!forum/artificial-general-intelligence

[Korrelan]

Scalable Bio- based intelligence first... then just let it loose on the big data.

They are both actually Bio-based approaches.  Even the convolution/ pooling networks 'big data' uses are based/ inspired by biological systems. 

In my opinion they won't achieve human level condition from current 'big data' schemas.  They are missing something fundamental to cognition, and the only way to obtain/ achieve it is full Bio-based engineering/ modelling.  They are currently taking just one small approximate function CNN of the brain and trying to recreate the whole. 

If we reproduce enough tyres and improve them enough... perhaps one day we will get a car... err nope.

Hoping that the output from a system will one day hopefully generate/ improve the algorithms that drive it is nonsensical to me.  Intelligence is not embedded in the data, the data is a product of intelligence... you can't work backwards.

In a true AGI system data improves intelligence, a CNN is just a filter, an algorithm to sort and find correlations and the data it receives does not improve its performance.

You can not separate intelligence and data, for an AGI to be intelligent is requires data during its development, but it’s not the human formatted data that ‘big data’ is using. 

That data has been altered/ filtered by human written programs to make it easy to understand and digest.  Humans have generic methods we apply to information; we have to because our written language is so limited. 

Although the words you are reading now make sense to you, they are many levels of abstraction away from how your cognitive system understands them; you can’t work back from these words to figure out how to build the system that generated them.  Ok you can figure out some of the rules the system uses to generate the text… but then you hit the wall, the problem of irreducibility.  This is just one of the hurdles/ reasons why I don’t think ‘big data’ schemas will ever achieve human level intelligence/ cognition.

‘Big data’ is just a phase, a catchword to fish/ temp funding by the money making/ financial wing of the corporations who are really, secretly waiting/ hoping that the next bit of insight gleaned by the Bio driven neuroscientists will further increase their fortunes.

 

[ivan.moony]

Neural networks are a kind of assembler for programming mind processes. I'm working on a replacement for neural networks that would have some interesting properties regarding to finding logic proofs. It is about bidirectional walking through a knowledge tree. Moving downwards would be finding solutions to problems, while moving upwards would be seeking what problem setups result with a specific solution. Today's artificial NN recognition process would be equivalent to moving upwards through a knowledge tree.

[keghn]

 @danlovy really like to hear you  feed back on my work, or any one else?. But O.K. If you not interested in my work.
 Gong to write a big paper. Need get all the confusion out of my model.
 A CNN is a supervised detector. That is for a down and dirty trick ok. I have a unsupervised detectors.
 If use a Tensorflow CNN in a AGI system it need
to be paired with temporal data base when it activated. And a Generative NN to recreate rung the bell, or to say what
activated the CNN. 
 Down and dirty would record a snap shot and cut out the object that activated the detector NN. With a image editor.

[danlovy]

As with any human endeavor, truth usual resides somewhere in the middle.

My theory -
https://www.linkedin.com/pulse/industrial-strength-evolution-genetics-ai-dan-lovy/

Video version(s)
Episode 1 (27 min)


Episode 2( 10 min)


Episode 3 (3 min)


follow here:
http://danlovy.com/critter/

[danlovy]

@danlovy really like to hear you  feed back on my work, or any one else?. But O.K. If you not interested in my work.
 Gong to write a big paper. Need get all the confusion out of my model.
 A CNN is a supervised detector. That is for a down and dirty trick ok. I have a unsupervised detectors.
 If use a Tensorflow CNN in a AGI system it need
to be paired with temporal data base when it activated. And a Generative NN to recreate rung the bell, or to say what
activated the CNN. 
 Down and dirty would record a snap shot and cut out the object that activated the detector NN. With a image editor.

Keghn -
I looked at the video and some of your links.  I'm confused and am having trouble understanding your approach.  Is there a summary link?

[keghn]

 Thank you @Danlovy. I guess it is back to the drawing board?

[danlovy]

Thank you @Danlovy. I guess it is back to the drawing board?

I watched the horse video,  I'm not sure what I'm seeing.

[ranch vermin]

For GA's, I dont think you have to worry about chromosomes, and anything natural,  just reinforcement learning should work.

The big hurdle is after youve mastered the basic physics of things,   you have to get things worthwhile for the robot to do,  because otherwise hes stuck in a limbo of not understanding the "game" we are in.

[keghn]

 In this cognitive machine there is a system for consciousness. Simply a a pointer that indexes through video. And then a
pointer that index through a single image frame.
 All video is then rebuilt up form smaller pieces so that objects can be identified. So that object can be tracked form one frame
to another with index pointer. The focus. A consciousness pointer.   

 The consciousness index focus pointer as a x and y, and also a z move through time, or through video, or new built video.   
 The pointer can select a atomic piece form a image like a cat or cup and use x, y, and z, and track the flight and movement from
one frame to another or use different x, y, z, and more dimensions, to morph form on object to another. Or, third to use as
a atomic piece selector and build new video in empty frames. Like in a way image editor works. 
 At the same time it make it's own internal language that it will learn to match to external spoken language.

 To find atomic pieces of thing in video and images i well use some thing like a detector NN.

YoloV2, Yolo 9000, SSD Mobilenet, Faster RCNN NasNet comparison: 


Segmentation with PSPNet-50 ADE20K - 4K dashcam #2:

[ranch vermin]

Sounds like a magic machine,  where does the cocacola flavour go in its database?

[keghn]

 Data base is built up in a unsupervised fashion. Three methods can be used.
 First). Is  to look for two or more matching patterns at different parts of the raw data with Kolmogorov Complexity algorithms:
https://en.wikipedia.org/wiki/Kolmogorov_complexity

 And together with:

 Frieze group  pattern:
https://en.wikipedia.org/wiki/Frieze_group
http://www.scientificcomputing.com/article/2015/09/patterns-are-math-we-love-look 

 The second). Is to have a internal doodle board and draw something on it and then go look for it in the data.
 A genetic algorithms evolve doodle form simple to complex.

 The third). Is  to random set weight in very simple neural network. And then go see if it will detect something in the data.
 To see what it detects, a doodle board generates image in evolving way until detector NN activates.
 The detector NN is evolved from simple to complex. This type is a Unsupervised Generative adversarial Neural Network 

 Once a pattern is detected a conscious focus can track it, manipulate it, and compare it with others. The detected is converted into internal
symbol and stored in memory to build up a language of synesthesia symbols and also used programming code instruction.

 The AGi  will be running code form the very fist time, initial starter code.  But with no detected symbols it is running a search
program to look for them. For the AGI to run write it need to use code along with symbols. New internal symbol will be used
as new instruction in its own ever growing code. So there will be a working memory that program will built and then jump into them like
function calls. But there will be a back step system if the code crashes. 
 This code will be it soul to be up loaded, at some time in the future.