Singularity is gona get U all Humans !

What do U think about Singularity ?

  • Nothing

    Votes: 8 28.6%
  • Its a real danger to human specie

    Votes: 4 14.3%
  • I dont care , that will never happen

    Votes: 12 42.9%
  • I am sure God will help us and we will fight it off

    Votes: 4 14.3%

  • Total voters
Billiard balls, too?

Certainly - in the sense that one can be made to affect the positioning of another or multiple others in chain-fashion. Action/reaction with the initial action supplied by an external force.

But not in the sense that they can learn and affect their own positions on order to achieve whatever. :D
When is it programmed? Each node is programmed before being connected to the other nodes in the neural net....
More proof of your ignorance! - You must know nothing about neural networks. Normally, each node in layer 1 (the input layer) is connected to all nodes in the second layer by a resister. Likewise all the outputs from the second layer nodes are connected to all the input nodes of the third or output layer (assuming a three layer net) by a resister. How do you program a resister? LOL

Typically, all resister have slightly different randomly chosen values initially.

One of the first methods of learning was proposed by Hebb - If the response on the first trial in the training set is a correct response (50/50 chance if only yes/no type of out put - two out put nodes) then more active paths (smaller interconnect resister) are strengthened. If the first response was wrong, then the more active paths are weakened. Then another member of the trianing set is presented to the input and if wrong same as before, if right same as before untill all membes of the training set have been presented. --- Repeated this thru all members of the set many times, repeating trials on each mamber, and eventually on the training set the performance is good, if not perfect. (More details I will not go into about how to have a good chance that the net will be able to solve the type of problem you are interested in.)

Neural nets are very useful precisely in complex problems with many variables - So complex that no human knows how to solve and certainly no human can program a solution to a problem so complex that he does not understand how to get any solution to the problem.

Sometimes, if the problem is not too complex, it is possible to examine the final values of the resisters in the fully trained net and then humans can understand how the network solved the problem - perhaps then it is even possible to program a von Neumann machine to solve the problem.

Programming a solution, when you do not know how to achieve a solution is clearly impossible - another "proof that" you are wrong - neural networks are not programmed because (1) usually humans do not know how to solve the problem and (2) you can not program the inter connection resistors!
Last edited by a moderator:
.... A neural net is nothing more than a lattice of machines connected together, each containing the software (programming) enabling them to communicate with each other, make decisions based on existing rules (programming) and create their own new rules (programming)....
Amazing ignorance!!!!!!!!!!!!

Nodes in the net are interconnected only by resisters whose values can be increased or reduced by the net itself, depending on the results of each trial during the training period. There are no programmed computers.

Hebb's, and as far as I know (25+ years ago anyway), all learning is very "behavioristic" I.e. punish* the active parts of circuit when the net gets wrong answer and strengthen them when the net get it right, trial by trial on the training set until either (1) the performance is acceptable (to the human wanting a solution) OR (2) performance on the training set is no longer improving as training continues.

There are no programmed computers and NO PROGRAMMING. - Just an electrical circuit wired up. - Is a radio receiver programmed too? -LOL

PS to help you get at least a slight idea about what you so ignorantly have been posting: A "node" in the intemediate and output layers is typically the summation junction or input of a high gain op-amp. All the prior layer outputs, via variable resistors, come to the summation junction of each intermediate and and output node. There usually are threshold circuits also. It is not a sub-computer in a network of computers as you are obviously assuming in your ignorance as to what neural networks are.
*By the net itself, depending on the results of each trial during he training period. All the human does is tell net if answer was correct or wrong (and start the next trail)
Last edited by a moderator:
That is NOT what Singularity is! I myself am a Singularitarian (EG, try to help the advance of singularity) so let me explain:

Singularity is the theory and practice of the exponential growth of science. The more you have, the faster it can advance due to better methods of research and development. The point of Singularity is where Biological and Mechanical mesh (which is already happening- prosthetics anyone?)

There will be no "super robots" taking over the world- instead it will be cybernetic implants helping to increase our brains NATURAL abilities. After all, the average dipshit uses what... 10% of their brain at best? Most "geniuses" only use around 13%... c'mon! Singularity is when we learn to "unlock" this latent potential. It IS the next step in human evolution because it is the only practical step left.

Singularity does NOT grant eternal life (that would be foolish)- instead, it grants you life at 125% efficiency UNTIL THE DAY YOU DIE. That's it... you decide "Oh, well, my affairs are in order and I've done all I wanted to do, time to pass on" And you are allowed to die with DIGNITY and HONOR! Not this slowly melting away in a bed hooked up to IV's and machines bullshit. You die with your boots on man!

Singularity is when NOBODY lies because people can TELL when you are lying! Think about that for a second- politicians and big business would HAVE to start getting their shit together or else people will tear em a new one! Crime will decrease drastically because criminals would have no way to hide. Privacy would be 100% and yet 0% at the same time!

Nobody would suffer the ravages of cancer, aids, and other such atrocities. Disease would NOT be 100% removed, as that defeats the purpose of living- you can NOT have good without it's equivalent bad! Instead, the recovery time is cut in half or more and methods are in place to enable you to carry on even while sick should you choose to do so.

Singularity is about embracing religion- a true religion. EVERY religion is the same damn thing with different interpretations. You cannot deny that christianity, catholocism, buddhism, jewdism, and paganism (the five main religions of the world) have some seriously striking similarities. Most differences are superficial - angels, lesser gods, demons, etc can all be the same thing described by different people. God, Buddha, Allah, etc are all the same "being", just as different people have seen him/her/it (burning bush anyone?) Religion is a set of guidelines set to enable humanity to survive... a set of guidelines that so many people are twisting to their own stubborn ends.

Humanity will be lucky to survive to singularity because we are so hellbent on destroying ourselves...
U all humans based on primitive instincts, (programmed Neural Nets) U are doing it again (the functioning) ie.

I am better than u, i know more than u, my knowledge is far superior than yours.

Inshort u r not evolving because u all are struck in your frame of evolution.

Now think what happens when the generation starts to evolve while thinking. No need to wait for the offsprings.

"Singularitys Gona Get U all" :itold:
You sir are an insult to Ray Kurzweil and singularitarians everywhere...
Thanks for that Billy, I'd never actually looked into the details of how a neural net is wired up, I just knew the basic idea. I knew that there were ways to handle the interconnections through hardware, but also assumed that the nodes could be any number of different mechanisms. Some of which could utilize hardware such as your example. Some of which might incorporate something along the lines of a microprocessor to route traffic.

Interesting read actually getting into the nuts and bolts of it though.
A neural net should be just that- neural. Biological components (like neurons) have a slower rate of transfer than most current methods, but they don't suffer from having to go only one way, breaks in the circuit can be self repairing, and all you need is some glucose and it's happily working :)
Have you heard of HTMs? Similar to neural networks but modelled on the structure of the human neocortex.
I bet most neural nets for the last ten years or more have actually been programmed simulations of neural nets on von Neumann machines. Processors have just become so cheap, fast and powerful that other techniques are nowhere near as economical.

I bet most neural nets for the last ten years or more have actually been programmed simulations of neural nets on von Neumann machines. Processors have just become so cheap, fast and powerful that other techniques are nowhere near as economical. psik
I have been out of the field for 25 years, but suspect your may be correct; however, if there are many separate factors to input there may be some speed advantages to a physical neural network.

For example, if you want to control a paper mill, and measure the Ph, the temperature, the viscosity, moisture content, and a dozen other observable at each of 25 points along the process stream converting pulp to paper and finally ending up with the tension in the rollers taking up the paper coming out, say 1000 inputs in total, then (I think) that much input may be a problem for a von Neumann machine, but just a large number of voltage connection points for the 1000 input nodes of the parallel inputs to layer one of the physical neural network.

I guess the approach one would use in a simulated neural network is to rapidly sample each of the 1000 pieces of concurrent input information and store them all in memory and then process that one "time-slice of data."

I really know nothing about how a lot of simultaneous data is processed by a von Neumann machine. Please someone who does, tell me if my guess is approximately correct, or if not, how are a 1000 different concurrent inputs handled. How would a von Neumann machine, even using a simulated neural network (because no one understands the relative importance or inter actions of all those parameters to program the control - the simulated neural network must learn how to do it well by getting feedback such as "paper too thin", "too yellow" etc. and learn how to adjust the connection strengths between the internal nodes levels to get feed back "Paper is good quality") Once it can accurately predict set of inputs that do produce the "good quality paper" result, then it can be empired to control the heating along the flow stream, the bleach added etc. - I.e. when it is well trained*, it takes over the control of the plant.
*No human will ever be able to understand as well how to adjust the plant parameters (temperatures, flow rate, bleach, etc.) to compensate for unexpected variables, the neural network does "understand" how to control various plant factors to keep the inputs to the well trained neural network such that it is making a "good quality paper" prediction almost all the time.

Many complex chemical processes have feed materials that can vary. - For example, one lot of hops or yeast may differ from another when making beer. When humans, even those with years of experience control the production process, the best results are due in large part to luck.
U seem to have a vague idea of neural networks may be 25 years old.

Whenever u give examples do u think in each example u dont have to program the flow of inputs and the results ?
And do u think that the machines understand the task without any programming ?

Whats interesting is that u dont care about Genetic Algorithms at all.
... Whenever u give examples do u think in each example u dont have to program the flow of inputs and the results?
To respond with the same pulp to paper plant example of neural network, NN,:

Yes, it is necessary to give the NN inputs (information). For example, one of the stations along the flow of pulp will have a transducer that converts the local temperature into a voltage and that voltage must be connected by a wire to ANY one of the input resistors of the first layer node inputs of the NN.

I hope you understand that the NN is not a digital machine, it works on analogue votages* - another reason why your concept of it needing to be programmed is so silly, ill informed.

Are you now calling that connecting of the wire "programming" ?
If yes, then as it is getting dark, I just "programmed" a light bulb!
LOL ....How silly can you get !!!!!!!!!

It is true that the wire from that particular temperature transducer must remain connected to that same input resister throught the entire training period.
*As I have previously stated, sometimes threshold circuits are used, mainly to eliminate noise, and if the output is binary (Yes or No, etc.) then a "winner take all" mutual competive feed back is often employed. - Sort of a competive bistable "flip flop" like circuit.

I.e. You want to avoid outputs like: "Yes voltage 7.6V" and "No voltage 5.6V." Instead you want: "Yes voltage 10V" and "No voltage 0V."
Last edited by a moderator: