CYC or COG?

Discussion in 'Intelligence & Machines' started by Rick, Nov 11, 2001.

Thread Status:
Not open for further replies.
  1. Rick Valued Senior Member

    Messages:
    3,336
    there are presently two different approaches for AI projects:
    CYC&COG,illustrate these two beautifully;
    CYC
    ==============================================
    In AUSTIN,TX,computer scientist Douglas b.LENAT is programming a computer called CYC(spoken as psych) with the basic facts about the worldeveryone knows,such as falling is dangerous etc,there goal is to create a 12-year old guy.the problem is 12-yr olds know a lot!!!!Lenats collegues have spent almost a decade feeding instructions into CYC,the computer now stores over one million rules!!!!!however he depends upon his programmer for feeding instructions,and can't learn on his own.
    COG
    ==============================================
    A different approach is taken by MIT prof.Rodney.A.brooks,who believes that Lenat is tackling the problem wrong way.in natural world intelligence arose as competition for survival increased.
    the path to true AI as brooks says is minimal pre-programmed knowledge,what they put is massive sensory input(sight hearing and touch)coupled with artificially programmed desires,for brooks intelligence isnt reasoning,but a set of behaviours as organisms interact with the enviornment.to make this point they have constructed a series of insect-like robots that have learned to crawl accross the fields strewn with boulders,they even steal soda from students desks!!!!the latest project COG(designed the same way.
    ****************************************************
    Now,the point is that we are still trapped...if go the CYC way,the robot would be to lousy and will be easy to cheat and could't respond easily....
    if we go the COG way,off course very advantageous(coupled with time consuming)but they would certainly after some time outlaw us(WITHOUT Asimov's laws off course!!!!).really a fix!imagine COG telling me what to wear on a date

    Please Register or Log in to view the hidden image!

    Please Register or Log in to view the hidden image!

     
    Last edited: Nov 12, 2001
  2. Google AdSense Guest Advertisement



    to hide all adverts.
  3. kmguru Staff Member

    Messages:
    11,757
    The COG is the way to go. Time consuming is not a problem. Just speed up the learning process. However my feeling is that creating a single AI will be very time consuming and may not be fruitful at all through COG system.

    So, CYC or COG : Neither will be successful.

    Try a different approach.
     
  4. Google AdSense Guest Advertisement



    to hide all adverts.
  5. Rick Valued Senior Member

    Messages:
    3,336
    Hi,
    yet another approach towards achieving AI is creating a computers that mimic the structure of human brain,Called the NEURAL NETWORKS (or simply neural nets),these computers are composed of hundres or thousands of tiny processors that are interconnected in multiple ways,just like neurons in human brain,(although i believe Stryder knows much more than i do about them

    Please Register or Log in to view the hidden image!

    )
    these neurons in human brain are connected by current conducting nerve fibres called gangia.often neurons has many ganglia attatched to it.scientists believe that when people learn things some connections are reinforced and grow stronger.

    in neural nets,thousands of computer processing units are connected in multiple ways.these NEURAL NETS arent programmed;they are trained.the net learns by trial and error(similiar to idea of COG),just as humans do.an incorrect guess weakens a particular pattern of connection,a correct way reinforces a pattern.after training is finished neural net knows how to do a particular task.in fact neural nets exhibit electromagnetic waves that are surprisingly similiar to human brain,but none of these nets has ever approached the complexity of even an animal brain.
    APPLICATIONS:
    Right now neural nets are comparing signatures made in banks with stored signatues,monitering state of aircraft engines,predicting stock market trends...but gradually they"ll find even wider uses.
     
  6. Google AdSense Guest Advertisement



    to hide all adverts.
  7. kmguru Staff Member

    Messages:
    11,757
    There is a minor problem with neural net. Human brain has multi layered processing system where the operating system and the automation system is built-in. Otherwise a baby will be dead before the neural network learns to control the glucose level and other thousands of sub systems in the body. Similarily, without a good OS, a crude neural net will do only one specific task for which it is designed (like the signature verification).

    But, it is a start. A lot more work needs to be done before we will know if we are heading in the right direction.

    BTW, one does not need thousands of hardware processors to create a neural net. You can do that in software using a single hardware processor. Future systems will use a three-way processor (for redundancy and fault-tolerance) with nerons simulated using software. That way, the AI can add more neurons as needed - just by thinking err...processing.
     
  8. Chagur .Seeker. Registered Senior Member

    Messages:
    2,235
    Confused ...

    Are neural nets being discussed or expert systems?

    There is a marked difference.
     
  9. Rick Valued Senior Member

    Messages:
    3,336
    Good point.

    hi KM,
    i agree that there has to be some pre-programmed knowledge for neual net systems.

    hi chagur,

    an expert system as you mentioned is pretty different from neural nets,as they rely on knowledge base, a dbase of represented knowledge,it solely relies on it,but neural nets trains itself which is pretty different.but i dont get the reason for your question about expert sytem,why did you mention it?please do answer to clarify more.
     
  10. Rick Valued Senior Member

    Messages:
    3,336
    Genetic algorithms:another way.

    Hi,

    another approach as i understand towards real AI could be genetic algorithms.

    GENETIC ALGORITHMS
    ==============================================
    intelligence is a by-product of evolution,isnt it?through the mechanism of natural selection,so why not try and create AI by creating lab conditions in which most intelligent programs survive.that is the object of gene algorithms.

    according to biological science,organism try and survive.occasional errors in the genetic code induces mutations,leading to changes,sometimes advantageous,sometimes not.organism with advantageous code dominate because they have opprtunities to reproduce.

    gene algorithms mimics nature in following fashion: a no. of algorithms are placed inside a computer system enviorment are given potential to mutate in random ways.all algorithms compete to try and solve the problem,over time an algrithms emerges as best and thus survives,but still there's no AI connection here as AI requires algorithms that mimic the behaviour of intelligent beings,but such algorithms are difficult to discover,still gene algorithms provide us with a new concept of AI.
     
  11. Chagur .Seeker. Registered Senior Member

    Messages:
    2,235
    Thanks, zion ...

    Prior to my replying, zion, you posted the following:
    Which as far as I am concerned, answered my question ... Neither CYC or COG are AI but rather sophisticated Expert Systems
     
  12. Stryder Keeper of "good" ideas. Valued Senior Member

    Messages:
    13,101
    From what I gather from a friend of mine participating within a course, there are two types of study, one of which Chagur has mentioned but I'm not sure to the context as it conflicts with the statement my friend made.

    Knowledge Based Systems:
    These are the artificial intelligence systems that are fed a broad band of information for the system to syphon relevant information.

    Expert Systems:
    Are systems fed information on a relevant subject and are used in such activities as diagnosing engines, or perhaps medical ailments. (Basically understanding something like an expert would)

    This kind of nibbles at Chagur's definition of an Expert system, by his usage.

    (of course feel free to search
    http://www.botspot.com for relavent AI information.)

    I personally like the understanding of using Neural Networks.
    Many for the Many paths and parallel processing effect, but because they can be very effective if a mixture of a fractal algorythm and processor pairing is devised.

    I've mentioned manytimes of a Hyperthetical device that would transcend time, well this is pretty much possible using a neural network, but only in very small incriments and a neural network using a parallel processed matrix of systems could organise all these small "Quantum leaps" into one single bound.

    (I mentioned processor pairing, as you could run the main body on something that only processes at 500Mhz, and then loop information through processors that run at 1Ghz or more. Of course the true power of a neural network allows a usage to go far beyond these speeds)

    I would write more information on this subject, but at the moment I'm undergoing negotiations for the development of a decent paper on the topic.
     
  13. kmguru Staff Member

    Messages:
    11,757
    I have already done expert systems with a neural net sub-system for industrial uses. We are proposing the same system for a DoD project. But I am still tinkering with AI. That is why I said the same path will not help to develop AI. It has to be a radical design.

    If any of you have any thoughts, ideas, let me know.
     
  14. Rick Valued Senior Member

    Messages:
    3,336
    Hi :conergence may be solution to true AI.

    i think path to truest AI could be convergence of what we"ve already discuused here,including what logic i"ll describe,just now(it actually mimics theory of probability in various respects).
    ==============================================
    FUZZY LOGIC
    ==============================================
    The term FUZZY LOGIC suggests inaccurate logic,but reference may sound more accurate than odd yes and no type.with fuzzy logic its possible to express a proposition with varying degrees of confidence about truthfulness.you hear it all the times during that CNN wether crap.Fuzzy logic helps make the expert systems and other AI-based system softwares work better by automating forms of reasoning that people use.an application of Fuzzy logic is the circuitory that enables a handheld video camera to adjust jigling picture.the circuit figures out what probably should and should not move and adjust the image accordingly.
    ==============================================
    BYE!.
     
  15. Merlijn curious cat Registered Senior Member

    Messages:
    1,014
    I hope I am not too late with this

    Okay, I have been away for a while. I could not reply, and now I don't know where to start.

    Zion, have you ever heard of Dynamic System Theory (DST)?
    DST uses a set of interacting (hence System) differential equations (hence Dynamic) to describe the state of the intelligence. There are, as in NN, no descrete representations of knowledge used. It looks like a promising field in both cognitve psychology, as well as in AI. Some (e.g. Tim van Gelder -and I agree) argue that NN's are a subdomain of DST. (btw I strongly believe COG uses DST).

    If you like to know more, ask.

    Merlijn
     
  16. Rick Valued Senior Member

    Messages:
    3,336
    Re: I hope I am not too late with this

    could you ellaborate more please.?
    how do you describe intelligence by the means of Differential equations??.

    Please Register or Log in to view the hidden image!

    Please Register or Log in to view the hidden image!

    Please Register or Log in to view the hidden image!

    Please Register or Log in to view the hidden image!

    Please Register or Log in to view the hidden image!

     
    Last edited: Nov 30, 2001
  17. Merlijn curious cat Registered Senior Member

    Messages:
    1,014

    Please Register or Log in to view the hidden image!

    Erm... DEs?
    I am not sure I know what you mean.
    I will write a reply when I figure that one out, or you've replied.

    cheers
    Merlijn
     
  18. Rick Valued Senior Member

    Messages:
    3,336
    Merlijn,

    you forgot about the differential equations that you were going to mention????i am still waiting...

    bye!
     
  19. kmguru Staff Member

    Messages:
    11,757
    Unless there is a different definition, DST as the name implies is nothing more than Nonlinear System Dynamics. So people have been playing in this field for a loooong time including fuzzy logic.

    We are not there yet...but moving at a snails pace for sure...
     
  20. Rick Valued Senior Member

    Messages:
    3,336
    Okay so i got this idea while writing a post to Kms thread "LIFE INSIDE A COMPUTER"...

    this may become a real path to AI.all we lack today is a software that emulates Human intelligence.so all we got to do is create them.

    but creating such a software is a daunting task.once we understand the uploading concept of human mind,we can do loads of things .like we download a persons brain softwares(with some changes off course) to a Robot a mechanical device etc..

    what do you think?

    bye!
     
  21. kmguru Staff Member

    Messages:
    11,757
    I have a feeling (can not prove it yet) that we have two brains (not the two hemis) - one is the consious mind that has emotions and personality and self aware, the other one is pure logic processor. That is why, when someone lies, you can pick up the truth from brain wave apttern or changes in thermal imaging on the face or GSR.

    So, the first step is to isolate the subconscious and transfer that. Then provide an emotion layer. If you transfer the whole thing, say your brain, you in the computer may get so mad...
     
  22. Rick Valued Senior Member

    Messages:
    3,336
    Km,confused...

    Hi,

    i thought sub-consious mind is capable of storing virtually everything,so it may consist of a permanent memory accesible to cortex for learning and making futuristic decisions.the other consious mind being somewhat just a processor of temporary as well as permanent memory...

    but i think the only need for sub-concious mind is some pre-requisite knowledge for actions performed <i>naturally</i> and only concious mind will provide true survival aspect to the Virtual Being...

    just speculating,i"d like to be clarified...

    bye!
     
  23. kmguru Staff Member

    Messages:
    11,757
    My opinion is that SCM is the processor devoid of emotions like Startrek Data. The CM is like his emotion chip. When we throw tantums etc, it is governed by CM. By keeping CM in synch with SCM, you get better results. Yogis, Buddhist etc spend lots of hours controlling the CM and harmonizing with SCM. SCM can not lie provided the information is correct. It has direct access to the databanks that is filled from the five senses.
     
Thread Status:
Not open for further replies.

Share This Page