OK, Billy, I've tried to be cordial throughout this whole discussion but I suppose it's time to just cut to the chase. I ask you one again - provide solid evidence of your claim of non-programming. Not just "your word" but solid, dependable references. The thing is, I'm MUCH more interested in facts than I am in debating.
Singularity is gona get U all Humans !
- Thread starter Singularity
- Start date
OK, Billy, I've tried to be cordial throughout this whole discussion but I suppose it's time to just cut to the chase. I ask you one again - provide solid evidence of your claim of non-programming. Not just "your word" but solid, dependable references. The thing is, I'm MUCH more interested in facts than I am in debating.
I have given at least five examples of Connection Machine applications following only exposure to training sets, no programming. - You and others know it is impossible to prove the non-existance of programming or non-existance of unicorns or non- existance of anything!
Please instead you simply tell when and how a neural network computer is programmed as you claim it is. - That is an easy request, not an impossible one, as you ask of me. I admit I can not prove the non-exisitance of unicorns or the non-existance of programming of connection machines.
Just tell me when and how it is programmed, as you say it is.
You're totally avoiding my request, sir. Even though there are many here - including yourself - that seem to enjoy endless debates, I do not.
And I've never once asked you to prove a negative - how dare you?????
My request is quite simple, if you actually know what you're talking about you should be able to provide some links that give solid, documented, somewhat detailed information on the things you are claiming to have existed long ago. I'm NOT asking for references to things that do not exist! Sheesh!
Now - can you do it or not? If you can I'd really like to see it.
I searched “neural networks" at:
Amazon.com: 48 books about them.
Ebay: 140 items for sale. Some are machines and many are used BOOKS about them.
In the few minutes I searched elsewhere I did not get any explicit statement that they are not programmed. I did find some text from Barons (at answer.com) which comes very close:
“…The system gets positive or negative response to output from the operator and stores that data so that it will make a better decision the next time. While still in its infancy, this technology shows promise for use in accounting, fraud detection, economic forecasting, and risk appraisals. The idea behind this software is to convert the order-taking computer into a "thinking" problem solver. …”
Most computers are von Neumann machines that “take orders” or “follow instructions” or are “programmed to do their job;” but Neural Networks are not. - They only are told (on each trial of the “training set” examples), if they guessed correctly or not. If “not,” then they modify themselves to guess better next time. Eventually, after many attempts on the training set, they do very well (~100% correct on the training set examples) and often better than 90% correct on similar trials with input cases never seen before.
From Wikipedia (http://en.wikipedia.org/wiki/Neural_networks):
Real life applications
“The tasks to which artificial neural networks are applied tend to fall within the following broad categories:
Function approximation, or regression analysis, including time series prediction and modelling.
Classification, including pattern and sequence recognition, novelty detection and sequential decision making.
Data processing, including filtering, clustering, blind signal separation and compression.
Application areas include system identification and control (vehicle control, process control), game-playing and decision making (backgammon, chess, racing), pattern recognition (radar systems, face identification, object recognition and more), sequence recognition (gesture, speech, handwritten text recognition), medical diagnosis, financial applications, data mining (or knowledge discovery in databases, "KDD"), visualisation and e-mail spam filtering. …”
AND:
“…neural networks have been applied successfully to speech recognition, image analysis and adaptive control, in order to construct software agents (in computer and video games) or autonomous robots. Most of the currently employed artificial neural networks for artificial intelligence are based on statistical estimation, optimization and control theory….”
I knew they were never programmed, but as have not been active in the field for more than 25years, I did not know that now they even construct programs for video game machine and robots. - Despite what Wiki states, I bet they are really just used there as fast parallel machines of amazing power, which no one understands (how they do their task) and no one has programmed. !!!
I know that to be the case of the neural networks that contol the paper quality at pulp to paper mills - no human can understand all the variables through out the plant's process in real time like a neural network can. It is more of an art than a science and Neural Networks do it better than even the old timer with 30 years of experience can. -I gave this example many posts ago.
Neural Networks machines may be only “software” as they are often simulated in a von Neumann machine, before physical construction. Then of course they require software, use programming, etc., but it is all for the von Neumann machine. I.e. even then, there is no software or program executed by the simulated neural network.*
OK now I have given you a lot from the literature supporting my point (Neural networks are not programmed yet very useful computers.) Can you give even one example of a neural network program, or even tell when it is done?
Answer: No you can not. If you do reply, it will be again to duck the issue, and demand something more from me, etc.!
------------------------------------------------
* I have some old neural network simulation I wrote approximately 30 years ago, but they are useless now as on the old 5.25 inch floppy disks. - They never were worth cost of converting and it would be nearly impossible to do so now. also they are in language Fortan or PL1 - I forget which.
I note in passing that there is a lot to be said for books. I can still get information from some written 300 years ago, but this 30 year old information can no longer be used.
Last edited by a moderator:
Thanks Hypewader for reference link explicitly stating:
"They are non-programmed adaptive information processing systems that can autonomously develop operational capailities in response to an information environment..."
I only searched briefly - I knew I was correct. If he wants to duck issue and refuses to learn, or even respond to my simple request for even one example of a program used by neural network - it is not worth more of my time to try to help him.
In earlier post I told of my hearing Terry Sejnowski, I think in about 1980, give talk at JHU/APL on his "NETtalk" neural network - that started several of us at JHU/APL seriously thinking and working on connection machines. (It was immediately obvious to us that connection machines probably could manage the defense of a navy ship against many incoming missiles better than any human could, so we were not "boondogeling" on the navy's money.) That Terry joined JHU in 1982 is for sure see:
http://en.wikipedia.org/wiki/Terry_http://en.wikipedia.org/wiki/Terry_Sejnowski
This proves my other claim that connection machines have been used for about 30 years. Terry was not the first by at least a decade, but he did make an important advance in them.
"They are non-programmed adaptive information processing systems that can autonomously develop operational capailities in response to an information environment..."
I only searched briefly - I knew I was correct. If he wants to duck issue and refuses to learn, or even respond to my simple request for even one example of a program used by neural network - it is not worth more of my time to try to help him.
In earlier post I told of my hearing Terry Sejnowski, I think in about 1980, give talk at JHU/APL on his "NETtalk" neural network - that started several of us at JHU/APL seriously thinking and working on connection machines. (It was immediately obvious to us that connection machines probably could manage the defense of a navy ship against many incoming missiles better than any human could, so we were not "boondogeling" on the navy's money.) That Terry joined JHU in 1982 is for sure see:
http://en.wikipedia.org/wiki/Terry_http://en.wikipedia.org/wiki/Terry_Sejnowski
This proves my other claim that connection machines have been used for about 30 years. Terry was not the first by at least a decade, but he did make an important advance in them.
Last edited by a moderator:
No, Billy, I've never been one to "duck" anything! And I've got you pegged to be that way, also.
And I'm glad to see you finally decided to get with the program (if you'll pardon the little pun).
That was some good information that both you and Hyperwaders cam eup with - well done!
I would still like to point out two things - first a quote from the book Hyper linked to:
Taken from: Section 3.2 Expert Systems
"There are three main components of an expert system: the rulebase, the factbase and the inference engine. ... The inference engine is a formal implementation of one or more reasoning mechanisims."
Care to guess what those "reasoning mechanisims" are?
Second: Guess what's on those floppies of yours? It sure isn't old copies of Life magazine.
My point is this - and you may think I'm splitting hairs here, and if so, so be it. Even though such systems DO learn and DO make correlations and DO improve on past mistakes, they will not - make that NEVER - get past the bootstrap point without some underlying software (program) to drive the process!! I've never once said or implied that they wee given everything in the form of data and programming. Certainly, such systems can learn from their mistakes once they are given a set of rules (instructions/program) to compare their results with previous data AND a target outcome.
Without those basic functions - which are in program form - they are nothing but metal and silicon. And that's all I've been saying all along. Not that they were constantly being reprogrammed or anything like that. And in the end, it certainly does not constitute Intelligence. It's just a machine with a feedback loop that's doing precisely what it was originally programmed to do.
And I'm glad to see you finally decided to get with the program (if you'll pardon the little pun).
That was some good information that both you and Hyperwaders cam eup with - well done!
I would still like to point out two things - first a quote from the book Hyper linked to:
Taken from: Section 3.2 Expert Systems
"There are three main components of an expert system: the rulebase, the factbase and the inference engine. ... The inference engine is a formal implementation of one or more reasoning mechanisims."
Care to guess what those "reasoning mechanisims" are?
Second: Guess what's on those floppies of yours? It sure isn't old copies of Life magazine.
My point is this - and you may think I'm splitting hairs here, and if so, so be it. Even though such systems DO learn and DO make correlations and DO improve on past mistakes, they will not - make that NEVER - get past the bootstrap point without some underlying software (program) to drive the process!! I've never once said or implied that they wee given everything in the form of data and programming. Certainly, such systems can learn from their mistakes once they are given a set of rules (instructions/program) to compare their results with previous data AND a target outcome.
Without those basic functions - which are in program form - they are nothing but metal and silicon. And that's all I've been saying all along. Not that they were constantly being reprogrammed or anything like that. And in the end, it certainly does not constitute Intelligence. It's just a machine with a feedback loop that's doing precisely what it was originally programmed to do.
Then, for the fifth time, I ask when is the neural network programmed?
Is it during its construction (the wiring?) If that is your answer, how does that different from the production of a baby with original "wiring" which it also modifies as it learns? Are you then calling the wiring of all computers “programming” - I.e. in a despirate effort to not be exposed as knowing nothing about neural networks, you are even killing the very useful distinction between hardware and software! - I.e. claiming all wiring is “programming” - even my flashlight’s wiring!!!! How silly can you get?
Why not just admit you were wrong? Are you stretching the concept of "programming" to such an extent that Babies are "programmed" by their initial brain wiring? - That is not what most mean by “programming.”
With the help of Hypewaders, I have given you exactly what you asked me to. - A clear statement by expert that neural networks are NOT programmed.
Can give even one example of the programming (you claim it is done). I can not prove that expert Hypewader found is correct. I can not prove neural network programming does not exist. I can not prove unicorns do not exist either.
One can never prove a negative. So long as you (now for the sixth time) refuse to either give an example of neural network programming (even one line of code for it is OK) and refuse to tell when it is done (in the life cycle of a machine) THEN YOU ARE DUCKING the issue and question.
Chill down Bill T, everyones not a AI expert.
Instead tell him that if what he says is programming then humans are programs too.
BTW, what about your ducking of my question ?
Instead tell him that if what he says is programming then humans are programs too.
BTW, what about your ducking of my question ?
Wow, comon singularity, you sure bring up some enlightened points, but frankly we are pretty effing far off from this kind of event. No Nanotubes is not going to magically make this possible.
If you just think about the amount of programming necessary for "instinct" - YIKES.
Frankly i'm not going to worry until I encounter A.I in a video game about 100X better that "Gal Civ II:dread lords".
If you just think about the amount of programming necessary for "instinct" - YIKES.
Frankly i'm not going to worry until I encounter A.I in a video game about 100X better that "Gal Civ II:dread lords".
Depends on what machine your playing that game on, how about IBM Deep Blue ?
Then, for the fifth time, I ask when is the neural network programmed?
Is it during its construction (the wiring?) If that is your answer, how does that different from the production of a baby with original "wiring" which it also modifies as it learns? Are you then calling the wiring of all computers “programming” - I.e. in a despirate effort to not be exposed as knowing nothing about neural networks, you are even killing the very useful distinction between hardware and software! - I.e. claiming all wiring is “programming” - even my flashlight’s wiring!!!! How silly can you get?
Why not just admit you were wrong? Are you stretching the concept of "programming" to such an extent that Babies are "programmed" by their initial brain wiring? - That is not what most mean by “programming.”
With the help of Hypewaders, I have given you exactly what you asked me to. - A clear statement by expert that neural networks are NOT programmed.
Can give even one example of the programming (you claim it is done). I can not prove that expert Hypewader found is correct. I can not prove neural network programming does not exist. I can not prove unicorns do not exist either.
So long as you (now for the sixth time) refuse to either give an example of neural network programming (even one line of code for it is OK) and refuse to tell when it is done (in the life cycle of a machine) THEN YOU ARE DUCKING the issue and question.
Wow! Chill, Bill, before you have a heart attack!
When is it programmed? Each node is programmed before being connected to the other nodes in the neural net.
But if you are so bull-headed as to not recognize the fact that software IS programming, then there's nothing I can do to help you.
Want to see a reference to software/programming in a neural net, eh? Then just read this: http://en.wikipedia.org/wiki/Neural_network_software
As I said before, I never duck an issue.
Ha ha!
I wonder if you can understand why you posting this is so funny?
You're talking out your ass, yes?
Yes. You are.
Heh.
The question, I suppose, is did you actually read the link you posted and grossly misunderstood it to the point where you think it actually backs up your point? Or did you just post a link blindly because it has certain keywords in the title?
Either is not really flattering for you.
Last edited:
Sorry, nexus, nothing funny about it. 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).
They are NOT - as Billy (and you, apparently) seems to visualize them - little nodes of intelligence that somehow "spring to life" without having been given some original directions (programming). Connect all the nodes you like - even thousands or millions of them - tell them nothing (no original programming) and they will just sit there like dumb rocks. I cannot fathom why that is so difficult to understand.
Then you don't understand that your link is for software that simulates a neural network?
The point Billy is making is perfectly valid. The 'programming' of a neural network exists in the hardwiring between the various nodes.
Each individual node might have some sort of programming to determine what it will do when a signal crosses its threshold (although this too could easily be dealt with through hardware, not software), but even so then you'd be talking about node programming and not programming of the network.
The neural network is an emergent process. It emerges from the nodes.
So, yes, it 'springs to life'.
As do you.
The point Billy is making is perfectly valid. The 'programming' of a neural network exists in the hardwiring between the various nodes.
Each individual node might have some sort of programming to determine what it will do when a signal crosses its threshold (although this too could easily be dealt with through hardware, not software), but even so then you'd be talking about node programming and not programming of the network.
The neural network is an emergent process. It emerges from the nodes.
So, yes, it 'springs to life'.
As do you.
Oh.
And since I'm here. Might as well chime in on the topic.
The Singularity (tm) is sensationalistic garbage.
And since I'm here. Might as well chime in on the topic.
The Singularity (tm) is sensationalistic garbage.
Yes, I did realize it was about simulating a neural net, but it's still valid in the point that EACH node must still be told (programmed) initally in order for it to do anything.
And I never once said or even implied that the network itself needed programming - I'm talking about the individual nodes themselves. They absolutely cannot "spring to life" on their own.
Then you're not talking about neural net programming, are you?
By the way, would you consider a neural net based on hydraulic valves and some sort of liquid to be 'programmed'? Just curious about your semantics.
(The valves, that is, not the network at large. We've already established that neural networks aren't programmed.)
By the way, would you consider a neural net based on hydraulic valves and some sort of liquid to be 'programmed'? Just curious about your semantics.
(The valves, that is, not the network at large. We've already established that neural networks aren't programmed.)
Correct, not the net as a unit.
And yes, one consisting of hydraulic valves. servos, pumps, cylinders, etc. would most certainly be programmed - fully analogous with hard-wired programming and the ability to be changed.