Robots are going to take over

[billvon]

I agree that the current economic crisis was largely caused by greedy humans; but I said that it would not have been likely to occur IF connection machines, well trained on prior loan applications and the corresponding repayment records, had made the "lend" or "not lend" decisions.

I disagree. Any machine is only as good as its inputs. If its inputs are mortgage default details during good economic times, the machine's outputs will only be valid during good economic times.

The idea that machines can more accurately predict things like market trends, foreclosure rates, citrus crops etc than people is attractive, but history has demonstrated that that is not so.

 

[fedr808]

I disagree. Any machine is only as good as its inputs. If its inputs are mortgage default details during good economic times, the machine's outputs will only be valid during good economic times.

The idea that machines can more accurately predict things like market trends, foreclosure rates, citrus crops etc than people is attractive, but history has demonstrated that that is not so.

good point.

 

[Billy T]

Well by your definition of "learning" any system that can alter itself in such a way to increase the efficiency/success of said system.

Close to what I would define learning as.

Lets emphasize the words "limited fields". So what if it can do a dozen jobs many times better than I can. I can potentially (and I stress that word) perform thousands, even millions of different tasks much better than it can, including constructing a better version of said ANN.
The fact is that the amount of tasks an individual person can potentially perform relatively well versus how many an individual ANN can potentially perform is many orders of magnitude greater.

Yes you can set up this straw horse and knock it down easily; but I never made any such claims. In fact many times I stated the humans were more versatile, etc.

Tell me, are ANN's self aware?

No, but again I never said they were.

If you delude yourself into thinking that this ability can really be considered artificial intelligence let alone evidence of significant learning then you need to learn what an ANN is.

Again no I don't think or suggest connection machines are, or ever will be able to, achieve what can be considered AI.

... If every node on an ANN is comparable to a real neuron then tell me, what exactly is the size of an ANN containing 50-100 billion nodes and 1000 trillion connections (which at this point may not even be possible to create that many connnections, but I don't know that off the top of my head).

Your "if" condition is false, but if there were a one to one correspondence then yes a connection machine would be at least as big as a small house.

It is precisely to try to avoid one reaching the conclusion that connection machines are artificial neural networks, ANNs, that I refuse (perhaps the only human still so stubborn) to call them neural networks. They don't resemble in any meaning full way the neural network of even a bat's brain. Which BTW can do some processing of acoustical information in real time that man’s best computers, filling a typical two person office room would need at least an hour to achieve. (Or may be too complex for man to ever write code for a digital machine that can do what the bat does in real time? For example a bat can not only tell which of dozens of bugs in a cluster of flying bugs is the tasty one from 25 feet away, but optimize his complexly changing sequence of chirps for tracking that one bug! Note he needs to make three Doppler frequency adjustments: one as the chirp source (the bat) is moving, one for the bug's motions, and one for his later motion from a different location as the echo returns! That much man can understand and do but that is the easy part of the bat's processing.)

Rather than directly reply to rest of your post I remind you that my ONLY reason for making any reply was to correct your errors, intended or not, as I am the self appointed "Sheriff of Nonsense" at Sciforums and also that I have never claimed connection machines can do any of the straw horses you have destroyed.

 

[fedr808]

Close to what I would define learning as.
Yes you can set up this straw horse and knock it down easily; but I never made any such claims. In fact many times I stated the humans were more versatile, etc.
No, but again I never said they were.
Again no I don't think or suggest connection machines are, or ever will be able to, achieve what can be considered AI.
Your "if" condition is false, but if there were a one to one correspondence then yes a connection machine would be at least as big as a small house.

It is precisely to try to avoid one reaching the conclusion that connection machines are artificial neural networks, ANNs, that I refuse (perhaps the only human still so stubborn) to call them neural networks. They don't resemble in any meaning full way the neural network of even a bat's brain. Which BTW can do some processing of acoustical information in real time that man’s best computers, filling a typical two person office room would need at least an hour to achieve. (Or may be too complex for man to ever write code for a digital machine that can do what the bat does in real time? For example a bat can not only tell which of dozens of bugs in a cluster of flying bugs is the tasty one from 25 feet away, but optimize his complexly changing sequence of chirps for tracking that one bug! Note he needs to make three Doppler frequency adjustments: one as the chirp source (the bat) is moving, one for the bug's motions, and one for his later motion from a different location as the echo returns! That much man can understand and do but that is the easy part of the bat's processing.)

Rather than directly reply to rest of your post I remind you that my ONLY reason for making any reply was to correct your errors, intended or not, as I am the self appointed "Sheriff of Nonsense" at Sciforums and also that I have never claimed connection machines can do any of the straw horses you have destroyed.

Okay, if you never made a claim that ANN's have comparable learning capabilities. Then why say this:

THE MACHINE LEARNS, how to decide

Why do you say that, when in post 84 there are several examples of them learning to solve problems and control complex processes even better than humans can?

yet you later quoted me

{post 85, in full}: The fundamental problem isn't that computers CANNOT learn, its that they cannot do it a fraction as well as we can.

To which you replied:

My "real point" as you say, in several posts, was just to show your post 85 claim is false or at least overstated. In many limited fields they learn to solve problems better than humans do,even some problem humans can not learn how to solve well enough to write an algorithm for conventional digitial computers.

You now understand that*, so you have been moving the goal posts - Now demanding flexibility, heuristic ability, creativity, etc.

. Were talking about artificial intelligence which is a hypothetical prerequisite to a robotic "takeover" as per the OP. If AI is a prerequisite then it requires the heuristic abilities, some creativity, and significant flexibility. This isn't about me, its about the artificial intelligence and the construction of such a thing.

Look, when I made my whole argument about size and efficiency of a large scale ANN you snapped back with the completely irrelevent argument that a chip can be made much smaller then the human brain. A completely irrelevent red herring, if you want to cite that as an argument in this discussion then I will cite the dust mite who has a much, much, much smaller brain/nervous system.

Rather than directly reply to rest of your post I remind you that my ONLY reason for making any reply was to correct your errors

Really? Because thus far you have conceded to my points and corrected YOURSELF far more often then you have corrected me. Actually, you never even corrected ANY error I made in the original statement that triggered this impotent retaliation of yours.

You have all but explicitly conceded my original conclusion, now, are you done wasting my time?

 

[Billy T]

Okay, if you never made a claim that ANN's have comparable learning capabilities. Then why say this: {the machine learns to decide}...

you are putting words in my mouth I never claimed that connection machines have "comparable learning capabilities" That partial quote from post 95 is clearly speaking of only the task the machine was trained on. Here is the paragraph you quoted the end of:

"... {connection machines} don’t suffer from greed for loan placement fee or the smile of a pretty girl, etc. They just use loan repayment histories to decide. Note no human programed the machine to make these decisions - It learned how to decide. The very same machine (with different inputs and historic data) could learn to give advice to person 1 about marrying person 2. etc. THE MACHINE LEARNS, how to decide. ..."

Obviously I am telling that the machine's ability is NOT comparable to a humans as it can ONLY decide on the class of problems it was trained on.

To twist the few final summary words in to a claim of being comparable learning capabilities is a gross distortion of what I said.

You also claim you have corrected my posts, but unlike me when making the claim that you made several errors, you do not even list one. I listed several of your errors and explained why what you stated was wrong. In one sentence you made three separate false claims. After they were refuted, I said in jest: "Three strikes and you are OUT"

Please give at least one false statement I made. I don't think there is even one, but I can repeat a list of at least five you made.

 

[spidergoat]

Those little RC helicopters are doing amazing things. There is a video on newscientist.com that shows three of them working to assemble a structure. They sound just like large bees.

 

[fedr808]

you are putting words in my mouth I never claimed that connection machines have "comparable learning capabilities" That partial quote from post 95 is clearly speaking of only the task the machine was trained on. Here is the paragraph you quoted the end of:

"... {connection machines} don’t suffer from greed for loan placement fee or the smile of a pretty girl, etc. They just use loan repayment histories to decide. Note no human programed the machine to make these decisions - It learned how to decide. The very same machine (with different inputs and historic data) could learn to give advice to person 1 about marrying person 2. etc. THE MACHINE LEARNS, how to decide. ..."

Obviously I am telling that the machine's ability is NOT comparable to a humans as it can ONLY decide on the class of problems it was trained on.

To twist the few final summary words in to a claim of being comparable learning capabilities is a gross distortion of what I said.

You also claim you have corrected my posts, but unlike me when making the claim that you made several errors, you do not even list one. I listed several of your errors and explained why what you stated was wrong. In one sentence you made three separate false claims. After they were refuted, I said in jest: "Three strikes and you are OUT"

Please give at least one false statement I made. I don't think there is even one, but I can repeat a list of at least five you made.

Wow, you just don't give up do you?

Billy T are you really trying to pick a losing fight?

By the way, your example about marriage, it would never work, ANN's tend to stop working well when you have too many inputs, and the amount of inputs, variables, etc... for a marriage are massive.

Obviously I am telling that the machine's ability is NOT comparable to a humans as it can ONLY decide on the class of problems it was trained on.

And that is why as I said in my original point, which you called an error:

The fundamental problem isn't that computers CANNOT learn, its that they cannot do it a fraction as well as we can.

Billy T, you literally are saying the exact same thing I said from the beginning.

You accepted my conclusion and henceforth, I win. Now, are you done wasting my time? Please stop trying to twist words to fit your argument, when it gets to the point where you are twisting your own words around rather then mine it becomes pathetic.

Every time you tried to correct an error your explanation was either erroneous, off topic, or bad science.

 

[Billy T]

... By the way, your example about marriage, it would never work, ANN's tend to stop working well when you have too many inputs, and the amount of inputs, variables, etc... for a marriage are massive. ...

True you could probably collect a couple of hundred facts that might be important to predicting the success or failure of a planned marriage. (That would be more information than a marriage councilor usually has available, but is conceivable.) But 200 inputs would still less than the data collected for the connection machine that was used to improve the consistence of the paper produced by a paper factory.

There is very little cost (less than a dollar, at most) to adding another input node to a connection machine, much less than the cost of the transducers that convert physical facts into electrical voltages. I already gave many of the facts that were transduced to voltages for the paper plant's production. (Things like pH, temperatures, flow rates, etc. at many dozens of points along the process, including the final drying rollers speeds etc.) I forget how many input nodes were used, but certainly more than 200.

Your statement that "ANNs tend to stop working well when you have too many inputs, …" is more unsupported false nonsense.

Remember that once the connection machine has been trained, the connection weights between all pairs of nodes are known. Thus, the connection machine can be replaced by a resistor divider network with each resistor's value inversely related to the known connection strength between the corresponding pairs of nodes.

Thus you are stating that a resistor divider network ceases to properly divide the current flow if there are too many resistors - Clearly more nonsense!

-------------------

I was never in a contest with you but if you want to claim that “you won”, that is fine with me. From my every first post until this last one I have ONLY been correcting your errors (as Sciforum’s “Sheriff of Nonsense”). You have kept me quite busy as in each of your new posts, like the one I am now correcting, you tend to make at least another nonsense statement.

You say you have corrected my posts more than I have corrected yours, but you continue to ignore my request for you to list even one false statement I have made – that is understandable as there are none. I.e. that claim of yours is also false.

Wow, you just don't give up do you? ...

Normally I continue to correct nonsense I notice but on a few occasions as Sheriff of Nonsense, I have given up after many tries to get the poster to understand his errors. You are diving me close to that point now.

 

[fedr808]

True you could probably collect a couple of hundred facts that might be important to predicting the success or failure of a planned marriage. (That would be more information than a marriage councilor usually has available, but is conceivable.) But 200 inputs would still less than the data collected for the connection machine that was used to improve the consistence of the paper produced by a paper factory.

There is very little cost (less than a dollar, at most) to adding another input node to a connection machine, much less than the cost of the transducers that convert physical facts into electrical voltages. I already gave many of the facts that were transduced to voltages for the paper plant's production. (Things like pH, temperatures, flow rates, etc. at many dozens of points along the process, including the final drying rollers speeds etc.) I forget how many input nodes were used, but certainly more than 200.

Your statement that "ANNs tend to stop working well when you have too many inputs, …" is more unsupported false nonsense.

Remember that once the connection machine has been trained, the connection weights between all pairs of nodes are known. Thus, the connection machine can be replaced by a resistor divider network with each resistor's value inversely related to the known connection strength between the corresponding pairs of nodes.

Thus you are stating that a resistor divider network ceases to properly divide the current flow if there are too many resistors - Clearly more nonsense!

-------------------

I was never in a contest with you but if you want to claim that “you won”, that is fine with me. From my every first post until this last one I have ONLY been correcting your errors (as Sciforum’s “Sheriff of Nonsense”). You have kept me quite busy as in each of your new posts, like the one I am now correcting, you tend to make at least another nonsense statement.

You say you have corrected my posts more than I have corrected yours, but you continue to ignore my request for you to list even one false statement I have made – that is understandable as there are none. I.e. that claim of yours is also false.Normally I continue to correct nonsense I notice but on a few occasions as Sheriff of Nonsense, I have given up after many tries to get the poster to understand his errors. You are diving me close to that point now.

Choice of model: This will depend on the data representation and the application. Overly complex models tend to lead to problems with learning.

I said "working well" for a reason mr. Billy T "nonsense". That is all I said and all I meant.

Yah, right, sheriff of nonsense, like that BS you believe about Russian jets. Class A bull.

You lost billy T, you literally admitted my original point over again. Instead of wasting my time mr. Sheriff, just give up, its getting to be pathetic.

You are not the sheriff of nonsense, please. Just because you pinned a dinky plastic gold star onto your chest means nothing. Your know it all attitude plus your overinflated ego leads to your erroneous "facts". Every single error you corrected your correction was either irrelevent, idiotic, or wrong, and for many of them you later CONCEDED that I was right in the first place and that you were merely wasting time.

Pathetic.

 

[Lady Historica]

AHHH THe Machines!!! Press the delete sequence button!!!

$$ \frac{4\pi}{\frac{3\pi}{\frac{2\pi}{\pi}}} = \frac{sqrt{4}}{\frac{\sqrt{3}}{\frac{\sqrt{2}}{2}$$

 

[fedr808]

haha. good point. Just get them to solve for pi and it should go into an infinite loop filling up all DRAM space and soon taking up hdd space eventually wearing down and breaking the hdd.

 

[smashingrobots]

Nobody knows the future of robots, what you can do is to make predictions

 

[fedr808]

ummm... okay?

Kind of obvious. If we can't predict the weather 12 hours ahead with any degree of accuracy how do you expect us to predict something else generations into the future?

 

[Pete]

haha. good point. Just get them to solve for pi and it should go into an infinite loop filling up all DRAM space and soon taking up hdd space eventually wearing down and breaking the hdd.

Something that did that would be a computer, not an AI. An AI might get bored after a while and dump the task, or tell you after a while that the task is beyond it, or even just tell you to go f&@# yourself!

Kind of obvious. If we can't predict the weather 12 hours ahead with any degree of accuracy

But we can!

... how do you expect us to predict something else generations into the future?

Predictions can always be made... How good they are is the tricky thing.

Generally speaking, any good prediction will have a particular probability distribution attached (but popular media doesn't like that - which is why weather forecasts get a bad rap), and the further ahead the prediction the broader the distribution.

But for technosociocultural forecasts like the theme of this thread... I suspect you're pretty much right. It seems difficult (impossible?) to separate the good predictions (if there are any) from the wild-ass guesses.

 

[fedr808]

Something that did that would be a computer, not an AI. An AI might get bored after a while and dump the task, or tell you after a while that the task is beyond it, or even just tell you to go f&@# yourself!

But we can!

Predictions can always be made... How good they are is the tricky thing.

Generally speaking, any good prediction will have a particular probability distribution attached (but popular media doesn't like that - which is why weather forecasts get a bad rap), and the further ahead the prediction the broader the distribution.

But for technosociocultural forecasts like the theme of this thread... I suspect you're pretty much right. It seems difficult (impossible?) to separate the good predictions (if there are any) from the wild-ass guesses.

Really? We can predict the weather to the third or even fifth significant figure (after decimel point). Bull. Look, where I live it can snow in february and hail in the end of april.

We cannot predict the weather with a significant degree of reliability, what percentage of weather predictions are erroneous?


But the fact is were humans, we look to the future, we base our future off what we can do in the present from what we learned in the past.

Human beings live in three different times, if we weren't to try and predict the future then we would not be who we are.

 

[Pete]

Really? We can predict the weather to the third or even fifth significant figure (after decimel point).

No, but that's not what you said. You said it can't be predicted "with any degree of accuracy"... which is Bull.

Look, where I live it can snow in february and hail in the end of april.

And that can be predicted at some probability level 12 hours in advance.

We cannot predict the weather with a significant degree of reliability, what percentage of weather predictions are erroneous?

What, the actual predictions with error bars, or the media sanitized versions?
Either way, you'll find a statistically significant correlation between forecasts and measurements.

 

[fedr808]

No, but that's not what you said. You said it can't be predicted "with any degree of accuracy"... which is Bull.

And that can be predicted at some probability level 12 hours in advance.


What, the actual predictions with error bars, or the media sanitized versions?
Either way, you'll find a statistically significant correlation between forecasts and measurements.

Im saying relative to many of the problems in science. No electrical engineer working on a 28 nm chip would accept two significant digits let alone the accuracy of weather prediction.

Im not saying its not accurate, what I am saying is that it is not nearly as accurate other forms of science would demand.

 

[Pete]

Perhaps you're confusing accuracy with precision. How precisely can an engineer predict the ultimate tensile strength of a particular beam?

 

[fedr808]

you know what pete, I'm done with debating accuracy and precision and what not.

Its all contextual and relative.

 

[birch]

i was just watching a movie called "time of eve" and it's about a society that has robots mingling with people. they function a lot like maids.

when robots are made to look and function like people, there could be some psychological issues with living people in how they view them and how they treat them. even if they are not actual feeling beings, the problem with treating them like objects with no respect can turn into a mindframe of abuse of robots. in some way, they can be the discriminated. the irony is even though it has not real effect on the robot, it can have an effect on people's thinking process and perhaps not always for the positive.

the more sophisticated the robot and look to function like living beings, they may need some rights or laws to protect them or at least treat them with some level of dignity.