Let's talk about brains

Discussion in 'Intelligence & Machines' started by hlreed, Jan 7, 2003.

Thread Status:
Not open for further replies.
  1. hlreed Registered Senior Member

    Messages:
    245
    I know this is intellegence and machines. Machines are ok.
    Intellegence is a bastard noun derived from the adjective intelligent. It is abstract on two levels. However, it should be obvious that what produces intellegent behavior derives from brains, a good noun we can discuss. Also we can make it into a mathematical object. Also, machine is promoted to animachine.

    Observe some brains in animals. Note that they come in all sizes and that they are always between the animal's sensors and motors. The data that flows from sensors to brain to motors is a data stream. (n[j-1] -> n[j] -> n[j+1] forever)

    Immediately we can propose a reason for brains, or a job if you like. The job of a brain is to mediate between sensors and motors.

    Inside the brain there are neurons. A neuron has more inputs than outputs (one), so it is a function. Let A be the axon, (output) and Sn be the synapses, (n is a number from 0).
    Let A = N(S1,S2...Sn) where N is some neuron function.

    Sensors are attached to the body of the animachine. They read some aspect of nature and convert it to a data stream. Let E be the environment or nature. Let O be an output. Then a sensor is
    O = S(E) ; where S is some conversion.

    Motors are also attached to the body. Motors read data streams and convert them to actions. E = M(I) where M is some conversion and I is an input.

    Now we can see that brain must be a list of sensors, motors and neuron functions.

    We can start at the least brain, since brains are from 0 to n, and build neuron functions.

    Is this not more fun than flapping about intelligence?

    Harold
     
  2. Google AdSense Guest Advertisement



    to hide all adverts.
  3. kmguru Staff Member

    Messages:
    11,757
    Hardware designer having fun? By all means start building the brain....

    One itsy bitsy problem...what is the algorithm that a neuron uses to do its work? If we can lick that, we are almost there....
     
  4. Google AdSense Guest Advertisement



    to hide all adverts.
  5. hlreed Registered Senior Member

    Messages:
    245
    Let O be output and Ik be inputs for k = 1 to n

    O = N(I1,I2....Ik)

    Find the smallest neuron.
    O = N(I) ; now functional possibilities are few

    Since data streams are the medium we have a data flow.

    O = d(I) ; is derivative of I (dx/dt = d(I) since dt = 1)
    O = i(I) ; is integral of I
    O = n(I) ; is negative of I
    O = Not(I) ; is arithmetic not of I (if I = 0 then O = last I else 0)
    And so on.

    If O = N(I1, I2) ; then the functions can be completed. (n input trees can be made with 2 input trees)

    O = I1 + I2 ; addition
    O = I1 - I2 ; subtraction
    O = I1 o I2 ; arithmetic OR. (O = max(I1,I2)
    O = I1 a I2 ; arithmetic AND. (O = min(I1,I2)

    These all make n input HalTrees and can match any neuron's inputs.

    We can also make a new kind of logic functions on data streams.

    O = I1 z I2 ; O = I2 if I1 = 0, else O = 0
    {z, g, l } = zero, greater than 0, less than 0

    and so on. I have about 28 node functions and many agents and structures.

    All this makes an algebra that makes brain lists. Every function here is a computer, so every function is a hardware node.

    Yes, we can build brains. Perhaps even better brains than the wet sloppy mess we have.

    Harold
     
  6. Google AdSense Guest Advertisement



    to hide all adverts.
  7. kmguru Staff Member

    Messages:
    11,757
    Ahh! a binary brain...I happen to have one under my desk....
     
  8. hlreed Registered Senior Member

    Messages:
    245
    Binary will not cut it. Binary can only do Boolean logic, no matter what you throw at it. You must get at least to integer before you can capture direction, for example.

    Direction = Left - Right ; gives direction
    Doing = Do - Don't ; gives instruction
    Isit = this is that ; detects equality

    Wasit = this was that ; detects and remembers equality

    Noun2 = adjective + Noun1
    Noun3 = property + Noun2
    Noun4 = attribute + Noun3

    All this maps to action, detection, language and we have not got to HalTrees or neurons yet.

    That thing under your desk has no brain. Its discription in this language is: Decode = Fetch ; where Fetch works on a linear register which is a single digit. (This is the basis. Decode also has commands that change the state of Fetch, but that does not add brain.)

    Thanks for the comments. We are getting there.

    Harold
     
  9. kmguru Staff Member

    Messages:
    11,757
    Well....I used one of these desk top devices to control an outdoor heater just as a human would do it -

    For example, if it gets cold, open the valve slowly
    If it gets colder, open the valve quicker
    If it gets very cold, open the valve very quickly

    And the program worked perfectly. Saved the company $200K per year. The program also took human supervisory over rides and remembered the next time, it got into a similar situation. The program also had search parameters to find optimum best conditions from a set of value returns and took the best decision.
     
  10. hlreed Registered Senior Member

    Messages:
    245
    Sure you did. If then control has been used for many years. That can be binary.
    In the 1700's they invented the revolving drum control, which still exists as music boxes.

    What do you do if you have a million inputs to provide a single answer or number? Answer: you make a neuron. Have you seen any state machines or if then logic in nature? If so let me know.

    To make a neuron of O = I1 + I2 connect another node to I1 and yet another to I2. Keep doing that until you reach a million or whatever. Then you have a HalTree which you could call a PTree because it is plus. It will gather an aggregate of all the inputs.

    Trees combine to make trees. Neurons combine to make bigger neuronal structures. Fractal as can be, just like numbers.

    Bye.
    Harold
     
  11. spookz Banned Banned

    Messages:
    6,390
    will you give us members discounts?
     
  12. hlreed Registered Senior Member

    Messages:
    245
    Re discounts on brain kits for members. Sure I will think about it.

    You may have to come out of the closet to buy. That is, you will have to fill in a form on my web site.

    Harold
     
  13. kmguru Staff Member

    Messages:
    11,757
    Sounds great.

    So, when is it going to be ready?

    As long as the answer is other than 42...we will be happy....
     
  14. hlreed Registered Senior Member

    Messages:
    245
    It is aleady done. The hardware is a node and cable system. Cables carry the data streams and connect to nodes.

    In the equation Doing = Do - Don't ; there is no mention of the battery node that is connected, since it only provides power to all the nodes connected in that network.

    I have just gone to a 10 wire system. (Started with 14) The cables are flat ribbon cable with plugs. The nodes have headers.

    When you are building a brain, you start with a motor, connect a CNode and build the tree that controls the motor. It happens that motors are independent, even when they use some of the same inputs. So logically, you could have a team working on each motor and tree.

    Here is the cable:
    pin name
    1 VPP ;voltage.
    2 gnd ;ground
    3 D0 ;least significant bit
    4 D1
    5 D2
    6 D3
    7 D4
    8 D5
    9 D6
    10 D7

    In the header it is:
    1 2
    3 4
    5 6
    7 8
    9 10

    Nodes have one, two or three headers.
    For the analog part I use 4 wire telephone cables and connectors.
    That carries voltage, ground, signal and spare to the analog parts of sensors and motors. Most nodes are not much wider than the cable.

    With this system, your construction is 3 dimensional. You end up with a great mass of cables which looks like another great mass off cables.

    You should see how smooth 8 bit arithmetic motors are.

    Harold
     
  15. kmguru Staff Member

    Messages:
    11,757
    What does your A. brain do?
     
  16. hlreed Registered Senior Member

    Messages:
    245
    The code for Big Al, a human sized animachine has about about 280 nodes, so far. This is all first level control. It has ability to stand, sit, walk in direction. At present it can sit, but does not know where to sit. It knows how to walk, but not where to walk.

    Big wall here. Code can code anything. But you must have something to code. This is unknown territory, but at least we can experiment now. The problems to attack are sight sense to things and properties. Given things and properties I can make and compare concepts, which would then tell where to sit, etc.

    There are many ways to do vision. We need separate pixels, but have cameras.

    Everything in this system scales, so we can start at the bottom and replicate upward. For example, define an object as a self similar patch in the visual area. Then thing = pixell is picelr
    Now, we have to put that all together, all over the map into things here, things there, things front, etc.

    As you can see, lots of work to do, but now we have an experimental tool.

    Try it yourself. You can make code, then make spreadsheet of parts of the code, truth tables and all that. Binary truth tables are a gross and good way to see all the main actions of your code.

    Thanks for the feedback.

    Harold
     
  17. kmguru Staff Member

    Messages:
    11,757
    Do you have any pictures you can post?
     
  18. hlreed Registered Senior Member

    Messages:
    245
    On my web site there are pictures, booklets which I will send if you ask. Big Al is mostly code and is scattered all over. I don't have enough money to make the thing.

    http://www.halbrain.com
     
Thread Status:
Not open for further replies.

Share This Page