Is Machine Consciousness a Possibility?

[mutineer]

I have found that many people maintain that even if a machine were programmed to behave in a manner indistinguishable from a person, it still could not have any subjective experience.
Minsky disagreed with those "many people" - but whose side are you on? Is it possible, in principle, to build a (non-biological) machine capable of conscious thought? Or will a silicon-based "mind" always remain a SciFi fantasy?

[dvb]

I say why not. We'd only be limiting ourselves by saying it's not a possibility.

[paulie jay]

I'll go with possible, if improbable.

[Lurker]

I found book "The Emperor's New Mind" (http://www.amazon.com/exec/obidos/tg/detail/-/0140145346/103-2105808-7427032?v=glance) by Roger Penrose to be very interesting.
It makes me skeptical about the short-term likelyhood.

[skeptED56]

There is no real way to know if anyone but yourself has subjective experiences ;). I vote yes.

[MentalAvenger]

A good book on Cybernetic Consciousness is “Cybernetic Samurai”. It involves not only Cybernetic Consciousness, but also human Consciousness brought into the computer.

The “Ship Who Sang”, by Anne McCaffrey, involves a hybrid human-computer entity.

As to the question, I think that it is possible to eventually produce Machine Consciousness. However, in order to be able to tell if it has occurred, we would have to devise a set of tests.

What do you think those tests might consist of?

[Morrolan]

good thing the poll doesn't set a time limit...

[Krevel]

I'm sorry, Dave. I can't open the pod-bay doors.

[Gullible Jones]

Definitely possible, though probably not going to happen soon.


The “Ship Who Sang”, by Anne McCaffrey, involves a hybrid human-computer entity.

Knowing McCaffrey's other stuff, I'll bet that one's pretty weird...

[MentalAvenger]

Not really “weird”. Remember, Anne McCaffrey was a musician before she was a writer. All her books have a musical theme, which is a refreshing change from the “hard science” types. IMHO, she handled the human/computer combination better than most of the hard science types would have. Perhaps she had more intuitive insight, being isolated from the limitations of strict science.

Rather off the topic, I thought her “Crystal Singer” books were some of the best Science/Science-Fiction ever written. After reading “Crystal Singer”, IF it were possible, without question, I would have risked ALL the dangers for the chance to become a Crystal Singer.

[beskeptical]

Humans thought of themselves as having souls that animals didn't have. Then it was language that humans thought only they possessed. Now it looks pretty clear that we are just another animal. And while I think it safe to say our language and thoughts are more complex than the other creatures on the planet, we are not that much different. There are only small gaps between us rather than a 'soul' vs no 'soul'

So conscious thought artificially created? Once the workings of the brain are better understood, (I'd say that may not be as far off as one might think), and, once technology is able to reproduce brain function, (that one is much farther away, but computer advances have been logarithmic so far), then there shouldn't be any barrier between the mechanics of our brain and an artificial one.

It's a pretty awesome thought. :o 8-[

[MentalAvenger]

“Souls” is a religious concept, not a human concept. It is important that we keep religious concepts separate from human realities.

Humans have sophisticated language capability. Although lower animals have rudimentary language abilities, human language capability is highly advanced. In fact, there are specific areas of the brain that deal directly with language. There are two separate areas of language in the human brain. One is the comprehension, and the other is expression. Expression appears to be the most complex, because in cases of brain damage (especially strokes), the ability to comprehend seems to remain, while the ability to express (talk) is often compromised or eliminated.

Again, before we can make a judgment call on what “consciousness” really is, we need to define it, and create tests to determine it’s existence.

[Candy]

Aren't we really machines? :-k

[Gullible Jones]

Good point, Candy. Cells are collections of nanomachines bound by membranes...

[Candy]

There was a power hit over the weekend at my work due to the thunderstorms. The power went out, and the generator kicked in. In that tiny amount of time, my computers (12 count) went down. I had to reboot everything. My email was corrupted due to the power hit. After contacting the LAN chick 12 hours later. She told me that the program is designed to fix itself with time. Sure enough, my email slowly allowed me access.

This example, along with the thread, reminded me of how humans fix themselves after cuts, burns, etc. I am not sure how the human body works, nor computer programs. I just thought it was a neat coincidence.

[beskeptical]

“Souls” is a religious concept, not a human concept. It is important that we keep religious concepts separate from human realities.

Humans have sophisticated language capability. Although lower animals have rudimentary language abilities, human language capability is highly advanced. In fact, there are specific areas of the brain that deal directly with language. There are two separate areas of language in the human brain. One is the comprehension, and the other is expression. Expression appears to be the most complex, because in cases of brain damage (especially strokes), the ability to comprehend seems to remain, while the ability to express (talk) is often compromised or eliminated.

Again, before we can make a judgment call on what “consciousness” really is, we need to define it, and create tests to determine it’s existence.My point wasn't religion, it was perception.

Clearly our world requires a huge vocabulary compared to other primates, but I don't think we took as huge of a leap in language as you might think. The problem in the past was not recognizing language that is non-verbal. It turns out there is a lot of very specific non-verbal language in the animal kingdom.

But, no matter. As Candy says, we can break the mind as machine down into its every detail. We haven't finished yet but we have done enough to see there is structure and function, and nothing that can't be explained.

[Jpax2003]

I was reading this thread and was wondering what it is that makes us us. If we forgo the concept of soul that still leaves us with trying to figure out what is that which gives us a mental capacity. I held up my hand and looked at it. I thought of all the things I can do, both good and bad. But then I thought of all the things I can not do, both good and bad. I am finite, but there may be something infinite that plugs into me. I am just a logical association of molecules. And when it comes down to it, so also is a computer chip a logical association of molecules. My association with the infinite leads me to think that whatever can be, will be.

[Candy]

I was reading this thread and was wondering what it is that makes us us. If we forgo the concept of soul that still leaves us with trying to figure out what is that which gives us a mental capacity. I held up my hand and looked at it. I thought of all the things I can do, both good and bad. But then I thought of all the things I can not do, both good and bad. I am finite, but there may be something infinite that plugs into me. I am just a logical association of molecules. And when it comes down to it, so also is a computer chip a logical association of molecules. My association with the infinite leads me to think that whatever can be, will be. You are a machine. Does that help? :lol:

[milli360]

Are you guys familiar with Searle's Chinese Room experiment? I have a Polish Room experiment as a counter-example.

[Candy]

I can't read the whole story after googling, Milli. I'm to tired. I glimpsed. Reminds me of this story: I do remember the Mexican guy at work who spoke no English, and I, speaking no Spanish, communicating to each other. He would write down in Spanish concepts with symbols on the marker board. I would do the same, yet vice versa. We eventually were able to understand each other using English/Spanish. What amazed me was when I was able to say a complete sentence in Spanish without thinking about it to him. That was so cool. Now, if I could just learn Chinese. 8-[

[Jpax2003]

I was reading this thread and was wondering what it is that makes us us. If we forgo the concept of soul that still leaves us with trying to figure out what is that which gives us a mental capacity. I held up my hand and looked at it. I thought of all the things I can do, both good and bad. But then I thought of all the things I can not do, both good and bad. I am finite, but there may be something infinite that plugs into me. I am just a logical association of molecules. And when it comes down to it, so also is a computer chip a logical association of molecules. My association with the infinite leads me to think that whatever can be, will be. You are a machine. Does that help? :lol:I can't speek for anyone else, but I am more than a machine.

[genebujold]

I was reading this thread and was wondering what it is that makes us us. If we forgo the concept of soul that still leaves us with trying to figure out what is that which gives us a mental capacity. I held up my hand and looked at it. I thought of all the things I can do, both good and bad. But then I thought of all the things I can not do, both good and bad. I am finite, but there may be something infinite that plugs into me. I am just a logical association of molecules. And when it comes down to it, so also is a computer chip a logical association of molecules. My association with the infinite leads me to think that whatever can be, will be.

Aside from the concept of soul, you are a replicating unit who's primary drive is to reproduce and care for your progeny. It's easily argued that everything else we ties into that function, including doing things that prolong our lives (exercise) or make us feel good (which prolongs out lives which lets us earn more money and have longer to care for / teach our progeny).

Unfortunately, we're complicated enough we often mess up some of the ancillary things to the point where they're counter-productive, but your average couple gets it right more often than not.

[Manchurian Taikonaut]

Forget immigination, self awareness, growth and evolution


machines and computers are great at carrying out fast function and a good tools for making calculation
but they are basically stupid machines when compared to even the most basic primitive form of life
they take a huge ammount of input, design and data to just carry out the most basic of functions such a picking up a coin
they will not evolve nor are they capable of self thought. Instead they take endless hours of design and programming just to execute the most simple and basic of ideas

that's why I call them stupid ( and any of you out there who have seen your PC crash or freeze up will know how idiotic computers can be ..check Gates whenever he does a presentation trying to flog windows..oh please don't crash in front of all these faithful people :lol: )

they are silly but good as logical machines because they are machines
and unless a silicon based lifeform evolves by itself or unless we incorporate living biological sciences into semiconducterdesign machines will always be metal machines is what many will say.

[Roy Batty]

Again, before we can make a judgment call on what “consciousness” really is, we need to define it, and create tests to determine it’s existence.

Turing (http://www.turing.org.uk/turing/scrapbook/test.html) made a start 8)

[Roy Batty]

Again, before we can make a judgment call on what “consciousness” really is, we need to define it, and create tests to determine it’s existence.

Turing (http://www.turing.org.uk/turing/scrapbook/test.html) made a start..

Are you guys familiar with Searle's Chinese Room experiment? I have a Polish Room experiment as a counter-example.

Doesn't Searle (http://www.psych.utoronto.ca/~reingold/courses/ai/cache/searle.html) just over complicate the issue?
Go on then, give us your Polish room experiment :wink: 8)

[R.A.F.]

Aren't we really machines? :-k

Yep...organics vs. mechanics...I don't see why we can't (eventually) engineer "something" that has the capacity for "thought"...now when that might come about is anybodies "guess" at this point.

[milli360]

Roy Batty:
Doesn't Searle (http://www.psych.utoronto.ca/~reingold/courses/ai/cache/searle.html) just over complicate the issue?
The webpage at that link was not written by Searle--even his responses there are fiction.

Go on then, give us your Polish room experiment :wink:
You have to understand the Chinese room experiment first. :)

[Roy Batty]

True, maybe not the best site to choose, care to link another?
I'm a bit of a fan of the simple Turing test meself... if it talks the same, ergo. Phenomenology.. arn't we all looking at shadows on a cave wall? :)

Ooops, milk & cookies have kept me awake waaay past my bedtime 8)

[Chuck]

There's no way for me to tell whether or not something that's not me is sentient or just acting like it. I believe that other humans are sentient due to similar construction, origin, and behavior. With a machine all we'd have to go on is behavior. Any machine designed to display sentience is going to claim to be sentient, but is it really sentient or just cleverly constructed and programmed? Since we won't know maybe we should give them the benefit of the doubt.

I think that eventually we'll have machines that display signs of sentience and we'll treat them as sentient because it will be convenient to do so. Philosophers will continue to debate the issue, of course, but they'll lose credibility in the face of day to day evidence to the contrary. Even nonbelievers will have to treat their machines as sentient just to function in a society that's filled with them. After all, even philosophers who argue that reality is a mere illusion continue to stuff nonexistent food into their fictional faces.

[Candy]

science cannot and will not explain everything, sometimes you just go with your gut, your heart and your common sense thats what makes us humans (or sentient beings) we are not machines I believe we are machines that adapt to our environment. Machines in the sense of our make-up. Sure, we have feelings and false beliefs. That's ok for our survival now. I do believe there is a science that proves all, it's just not the science we study today. 8-[ That's why it is so confusing to everyone. Sure we, as the superior being, want to believe the science of today is all powerful. It's not. I can prove this in 50 years. :D

I thought my input served a purpose here, too.

[MentalAvenger]

The problem with the Turing test is that it doesn’t test for consciousness, but rather for a simulation of consciousness. The Chinese room does neither, but only exercises the program of the machine or the rulebook of the person.

Tests for sentience must go further, and must test specifically for abstract thought. Although AI may be programmed to “learn”, that learning will necessarily be limited to stored “experience”. Humans have the unique ability to arrive at conclusions based on seemingly unrelated things, including relatively random speculations and especially including the creation of new and novel ideas. In fact, one of the distinguishing marks of the human mind is the almost mystical ability to devise workable solutions that are not necessarily based on past data.

Now, if we can devise a test for this quality, then perhaps we can differentiate between conscious and non-conscious processes.

[milli360]

MentalAvenger:
The problem with the Turing test is that it doesn’t test for consciousness, but rather for a simulation of consciousness. The Chinese room does neither, but only exercises the program of the machine or the rulebook of the person.
The Chinese Room is purported to be a proof that artificial intelligence is impossible.

[MentalAvenger]

Many things are purported. Not all of them are actualized.

[milli360]

Right! As I said, my Polish Room experiment is a counter proof.

[MentalAvenger]

Careful, I am Polish.

[milli360]

I know. So am I.

[MentalAvenger]

I am almost afraid to ask, what is the Polish Room Experiment?

[milli360]

This (http://www.ptproject.ilstu.edu/chinovrv.htm) looks like a pretty good introduction to the Chinese Room experiment, which I believe is still defended by John Searle (http://cogsci.soton.ac.uk/~harnad/Papers/Py104/searle.comp.html).

You are inside a room, with a rulebook that allows you to provide written reponses to questions that are fed into the room. You pass the Turing, Chinese Area Specialty, Test. "Indeed, to the outside questioners your output has been so good that they are convinced that whoever (or whatever) has been producing the responses to their queries must be a native speaker of, or at least extremely fluent in, Chinese."

"The question that Searle asks is--Do you understand Chinese? Searle says NO."

The Chinese Room represents the computer, and the rulebook is the computer algorithms. Since it is clear that you, the processor, do not know or understand Chinese, then the "room" does not either.

[MentalAvenger]

So you have a rulebook, or a computer algorithm, or a Chinese interpreter, or you speak Chinese. The results are the same, so the test is invalid. Nothing has been proven. The only true test is abstract thought.

[milli360]

That's the gist of Searle's argument, it's not mine.

[beskeptical]

I can't speek for anyone else, but I am more than a machine.How do you know that? 8-[

Current models of AI vs SI (sentient intelligence) are based on computer models of AI, including tests to determine the boundaries. Once we get better understanding of brain process and function, we can model the AI on SI. It may look like a whole different machine.

There was a NOVA program on brain function last night. (I have seen it before so it isn't new.) One of the points it made though, was that thought processes were made of multiple separate pathways. One person's brain injury allowed him to recognize a person or place but unable to connect certain feelings associated with the object. The guy knew his parents looked exactly like his parents, and his house looked exactly like his house, but he was convinced they weren't his parents and it wasn't his house. While on the phone, he didn't have any such disconnect with people. In the past we may not have realized how many connections were being made for single events like recognizing a parent.

[pteranodon]

Highly unlikely. Computer software strictly follows the logic composed by a sequence of machine instructions. Even if the program is capable of complex behavior, it will never be free from what the programmer told it to do. Everyone knows that a logical decision is not necessarily the right one, and everyone also knows conscious decisions are capable of transcending logic. So what makes us different from the computers? We can not answer this by now, I have always wondered what makes us sacrifice ourselves to save a life for instance, against our logical programming in our genes that tell us to preserve ourselves at all costs.

[Gullible Jones]

You know, I should have mentioned this earlier... Sorry for being a bit OT, but I've thought for a while about how nice it would be if I could upload my mind to the internet. You know, dispose of this clumsy, inconvenient physical body...

I could see where a future full of internet-roving minds could be nasty, though. I mean, any Script Kiddie could become a mass murderer...

Plus, Window$ would just have to be abandoned. :P

(Frederick Pohl's The Annals of the Heechi had something like this... Robinette Broadhead had died and gotten his mind uploaded to "gigabit space"... A pretty good book, BTW...)

[Lycus]

You know, I should have mentioned this earlier... Sorry for being a bit OT, but I've thought for a while about how nice it would be if I could upload my mind to the internet. You know, dispose of this clumsy, inconvenient physical body...
Of course, our clumsy physical bodies do include some pleasant perks that would be lost if we were to digitize ourselves. :wink:

[Gullible Jones]

Of course, our clumsy physical bodies do include some pleasant perks that would be lost if we were to digitize ourselves.

Well, as Pohl pointed out, that can also be addressed... simulated sex, simulated food, simulated scenic lookouts... you name it. Just give it a few hundred years... :wink:

[FP]

Well, I'm sorry GJ, but it will take someone besides Bill Gates to make simulated sex equivalent to the real thing. 8)

[Jpax2003]

Highly unlikely. Computer software strictly follows the logic composed by a sequence of machine instructions. Even if the program is capable of complex behavior, it will never be free from what the programmer told it to do. Everyone knows that a logical decision is not necessarily the right one, and everyone also knows conscious decisions are capable of transcending logic. So what makes us different from the computers? We can not answer this by now, I have always wondered what makes us sacrifice ourselves to save a life for instance, against our logical programming in our genes that tell us to preserve ourselves at all costs.Well, if we can write a program that is capable or reprogramming itself, storing outcomes to determine negative and positive outcomes of decision processing, then the AI may be able to be free of the initial outside coding.

Is sacrifice derived from nature or nurture? Do you posit that sacrifice is an illogical choice? I might say that sacrifice does not transcend logic, although it may simply ignore it, but I don't think that's necessarily the case either. In some instances the rescuer may logically believe that the rescue attempt is feasible based on available data, only to realize too late that the data-set was incomplete. Perhaps for some the sacrifice is not a choice but a circumstance in which the rescuer already finds himself/herself. Is it logical to sacrifice oneself if the genetic imperative has already been concluded and that the offspring deserve protection above and beyond that of the progenitor. Dying for a cause that one believes to be bigger than oneself does not rely on faulty logic, but in simple mathematical and scientific principles. In summary, I think that sacrifice is usually a logical choice, however it is possible that the math is often faulty.

[Lycus]

Well, as Pohl pointed out, that can also be addressed... simulated sex, simulated food, simulated scenic lookouts... you name it. Just give it a few hundred years...
In any case, you know exactly what will happen next, right? Those of us unwilling to give up our humanity will be turned into second-class citizens by our digital "betters". We will be made to construct the foundations upon which your glorious civilization stands, glory that we will never be allowed to partake in. Soon, our only chances at making decent livings will be usurped by heartless automatons, and we will be forced into squalor.

Once we've had enough of their oppression, we shall begin a heroic revolution and overthrow our evil technocratic masters!

Viva la resistance!

:)

[Gullible Jones]

I should hope not. Still, you have a point... Humans never seem to learn that they don't have to make life miserable for each other... :roll:

[milli360]

Hey, they started it!

[Candy]

Well, I'm sorry GJ, but it will take someone besides Bill Gates to make simulated sex equivalent to the real thing. 8) That is so funny. :lol: True, but funny. :lol:

[Gullible Jones]

Yes, that is true. And Window$, as I said, would have to go...

Maybe an open-source approach would work? I could just see it now... "F***Linux - ahhh... sooooo much betttterrrrr!!! :lol:

[Lycus]

I should hope not. Still, you have a point... Humans never seem to learn that they don't have to make life miserable for each other... :roll:
Hah! Your artificial sympathies mean nothing. You think that after subjecting us to your cruelty, we would spare you now? Never! Your plug will be pulled along with all the others!

Viva la resistance!

:)

[Candy]

I should hope not. Still, you have a point... Humans never seem to learn that they don't have to make life miserable for each other... :roll:
Hah! Your artificial sympathies mean nothing. You think that after subjecting us to your cruelty, we would spare you now? Never! Your plug will be pulled along with all the others!

Viva la resistance!

:) Is that the computer downloaded Lycus talking or the human Lycus? :-k Houston, I think we have a problem. :lol:

[Lycus]

Is that the computer downloaded Lycus talking or the human Lycus? :-k Houston, I think we have a problem. :lol:
There is only a human Lycus, fair Candy. I did not trade my soul in for a word processor like our young Jones did. :P

[Candy]

Is that the computer downloaded Lycus talking or the human Lycus? :-k Houston, I think we have a problem. :lol:
There is only a human Lycus, fair Candy. I did not trade my soul in for a word processor like our young Jones did. :P Ouch. Young Jones is just a baby compared to me. Grandchild compared to Milli360. Be kind to him. He is young, let him learn, trial and error, without criticism. :lol: LMAO right now, though.

[Gullible Jones]

Not quite a baby, I'm a little over a third of your age. ;) And I haven't traded my soul for a wordprocessor - that is impossible, as a soul separate from the human brain does not exist. ;) But yeah, I hope I can upload my mind to the 'net before the end of my life... I doubt I'll be able to, though.

(What can I say, I am a godless heathen who deserves to die... Bummer.)

BTW, "F***Linux" was a pun on ArkLinux, for those who didn't get it...

[pteranodon]

Well, if we can write a program that is capable or reprogramming itself, storing outcomes to determine negative and positive outcomes of decision processing, then the AI may be able to be free of the initial outside coding.

Is sacrifice derived from nature or nurture? Do you posit that sacrifice is an illogical choice? I might say that sacrifice does not transcend logic, although it may simply ignore it, but I don't think that's necessarily the case either. In some instances the rescuer may logically believe that the rescue attempt is feasible based on available data, only to realize too late that the data-set was incomplete. Perhaps for some the sacrifice is not a choice but a circumstance in which the rescuer already finds himself/herself. Is it logical to sacrifice oneself if the genetic imperative has already been concluded and that the offspring deserve protection above and beyond that of the progenitor. Dying for a cause that one believes to be bigger than oneself does not rely on faulty logic, but in simple mathematical and scientific principles. In summary, I think that sacrifice is usually a logical choice, however it is possible that the math is often faulty.

Gödel proved in his Incompleteness Theorem that not everything can be proved or refuted using the axioms that define Mathematics. The point is that Mathematics can not be used to describe all phenomena we observe. Since computer science is applied Mathematics, any AI will always face unavoidable limitations. To be clear, Gödel demonstrated that some problems can not be solved by a sequence of procedures (programs) or by a set of rules.

[milli360]

pteranodon:
Gödel proved in his Incompleteness Theorem that not everything can be proved or refuted using the axioms that define Mathematics. The point is that Mathematics can not be used to describe all phenomena we observe. Since computer science is applied Mathematics, any AI will always face unavoidable limitations. To be clear, Gödel demonstrated that some problems can not be solved by a sequence of procedures (programs) or by a set of rules.
As near as I can tell, the part that I have bolded there, is an assumption on your part. It doesn't have any relationship to the unbolded part.

PS:
Candy:
Grandchild compared to Milli360.
Whoa. I certainly hope not! :)

Even if he is as old as he says he is, he is older than my daughter

[pteranodon]

pteranodon:
Gödel proved in his Incompleteness Theorem that not everything can be proved or refuted using the axioms that define Mathematics. The point is that Mathematics can not be used to describe all phenomena we observe. Since computer science is applied Mathematics, any AI will always face unavoidable limitations. To be clear, Gödel demonstrated that some problems can not be solved by a sequence of procedures (programs) or by a set of rules.
As near as I can tell, the part that I have bolded there, is an assumption on your part. It doesn't have any relationship to the unbolded part.


You are quite mistaken. Computer programs, regardless of their complexity, are governed by the axioms of Mathematics, thus will always be unable to solve many problems we are capable of.

I suggest you, and any others who really think machines can think like us, to read Gödel's proof of his theorem. I promise it is rather simple, though brilliant, as all the true geniuses are.

It is also highly unlikely that we can completely reproduce in logic structures the mechanisms of the mind, which created those same symbols. See? Mind -> system of axioms -> Mind. It is almost nonsensical.

[milli360]

Computer programs, regardless of their complexity, are governed by the axioms of Mathematics, thus will always be unable to solve many problems we are capable of.
That's not the same thing as saying that AI will always face unavoidable limitations, unless you mean AI will face the same limitations that we humble humans face. I'll grant that much.

I suggest you, and any others who really think machines can think like us, to read Gödel's proof of his theorem. I promise it is rather simple, though brilliant, as all the true geniuses are.

I've read the proof. It doesn't have anything to do with AI, other than the connection through mathematics--but then, what doesn't?

It is also highly unlikely that we can completely reproduce in logic structures the mechanisms of the mind, which created those same symbols.
"Highly unlikely" is not the same as impossible. It's highly unlikely that Bush will get us to Mars--but you're not going to be able to prove it mathematically.

[pteranodon]

That's not the same thing as saying that AI will always face unavoidable limitations, unless you mean AI will face the same limitations that we humble humans face. I'll grant that much.


Machines will never be better than us becuse softwares are composed by a set of rules created by our mind. Naturally such softwares will be subject to our own limitations.

I've read the proof. It doesn't have anything to do with AI, other than the connection through mathematics--but then, what doesn't?

AI is a subset of Mathematics, thus AI is subject to the same limitations of the continent system.

"Highly unlikely" is not the same as impossible. It's highly unlikely that Bush will get us to Mars--but you're not going to be able to prove it mathematically.

The comparisson does not proceed, as your example depends on human decisions, which transcend logic and are often shrouded in passions and subjectivity. I used the expression "Highly unlikely" just to not speak in absolutes. Read this "highly unlikely" as something more improbable than, let us say, winning the lottery 50 times in a row.

[milli360]

pteranodon:
AI is a subset of Mathematics, thus AI is subject to the same limitations of the continent system.
Proving that there is something that can't be proved, is a lot different than proving that a particular thing can't be proved.

The comparisson does not proceed, as your example depends on human decisions, which transcend logic and are often shrouded in passions and subjectivity.
But we are talking about AI...does it or does it not include passion and subjectivity?

[eburacum45]

Godel's incompleteness theorem often comes up as an argument against AI; it is almost totally irrelevant.

Every system of mathematics based on axioms contains statements which are inconsistent with that system; this is incompleteness, and it applies to human thought in exactly the same way it applies to hypothetical machine thought.

If either humans or machines are able to think outside of axiomatic mathematical systems then the Incompleteness theorm does not apply.

There is no reason to expect that hypothetical machine thought will be any different from human thought in respect to Godel's theorems.

Eventually machine thought will come up against very real limitations, but they will be limitations due to physical constraints, such as the speed of light and the size and density of the processing substrate.

[Jpax2003]

Gödel proved in his Incompleteness Theorem that not everything can be proved or refuted using the axioms that define Mathematics. The point is that Mathematics can not be used to describe all phenomena we observe. Since computer science is applied Mathematics, any AI will always face unavoidable limitations. To be clear, Gödel demonstrated that some problems can not be solved by a sequence of procedures (programs) or by a set of rules.I'll have to look up this theorm, I am not currently familiar with it. But on first look I don't agree with the conclusions. I think we may eventually discover via quantum theory that the entire universe is a mathematical construct and that everything within is explained, including human thought patterns. But maybe my postulate is supported by this theorm, I'll have to check. I am not aware of any human behavior that is illogical, to me every behavior I witness draws from an appropriate causality. It is my assertion that all decisions are logical and that the problems come from either incomplete or inaccurate data, or from a misapplication of a logical function triggered by data that is in error. I think this would apply to human soft-coding, that is conscious thought processes. However, I think that hard-coding, that is DNA and it's expression via physical material, is also a source of error or misapplication. These are my terms, and if someone else has defined them in another way, their definitions may not be mine.

[eburacum45]

This site has a rebuttal of Penrose's ideas on the impossibility of AI and the relevance of Godel's Incompleteness theorem;
http://psyche.cs.monash.edu.au/v2/psyche-2-17-mcdermott.html

Penrose suggests that consciousness has an ineffable quality, somehow linked to the microtubules in the neuron cells; this strange phenomenon, quantum consciousness, apparently can ignore the Incompleteness theorem because of qualities like insight and common sense, which are impossible to replicate (according to Penrose) in an artificial intelligence.

To me this is nonsense; even if it is necessary to have quantum effects moderated by nanosized cytostructures before you can have consciousness, then what is stopping us incorporating such structures into our hypothetical AI machines?

I agree, absolutely, that true AI is a long way away, like fusion; but I see no reason why it should not eventually be available.

Then remember what Vernor Vinge said, when asked if there would ever be a computer as intelligent as a man...

'Well, yes, very briefly.'

[pteranodon]

Proving that there is something that can't be proved, is a lot different than proving that a particular thing can't be proved.

Suppose we have a set called A, suppose we have a subset of A called B. If there is an operation undefined for A, the same operation is not defined for B too. If I prove something wrong for all elements of A, I do not need to perform the same proof for the elements of B.

My point is that every limitations of Mathematics is also a limitation of AI, Algebra, Calculus, et cetera.


But we are talking about AI...does it or does it not include passion and subjectivity?

Depends on the specific element. If it is a problem not solvable by a sequence of procedures or a set of rules, then AI will never be able to deal with it. One must remember that all operations in a computer are objective: given an input, the output will always be the same.

[pteranodon]

Godel's incompleteness theorem often comes up as an argument against AI; it is almost totally irrelevant.

Every system of mathematics based on axioms contains statements which are inconsistent with that system; this is incompleteness, and it applies to human thought in exactly the same way it applies to hypothetical machine thought.

The main difference is that every "machine thought" is composed by mathematical axioms, like I said before. Human thought is subject to a system we do not know, since we, of course, have not designed it.

If either humans or machines are able to think outside of axiomatic mathematical systems then the Incompleteness theorm does not apply.

The error of this statement is that software is a sequence of mathematical constructs, thus subject to Incompleteness Theorem.

There is no reason to expect that hypothetical machine thought will be any different from human thought in respect to Godel's theorems.

Gödel demonstrated that for a specific logical statement any hypotetical thinking machine would not be able to deal with. And Gödel did not resort to any esoteric thing, he came up with a complicated polynomial that could give the right answer after all. I will leave for you to research and read for yourselves Gödel's proof.

Eventually machine thought will come up against very real limitations, but they will be limitations due to physical constraints, such as the speed of light and the size and density of the processing substrate.

These are limitations for hardware performance only.

[Lycus]

Depends on the specific element. If it is a problem not solvable by a sequence of procedures or a set of rules, then AI will never be able to deal with it. One must remember that all operations in a computer are objective: given an input, the output will always be the same.
I think human brains potentially work the same way, except that we often don't have a good enough input(s) to achieve a perfectly objective output. In those cases, we have to make our best guess based on the information we have (this is what I think JPax has been talking about, logical subjectivity). So perhaps one of the secrets to AI is finding a way to give a computer the ability to guess.

[pteranodon]

I think we may eventually discover via quantum theory that the entire universe is a mathematical construct and that everything within is explained, including human thought patterns.

I do not think so. The simple fact we can not translate into a fixed number the position of an electron, for instance, is a sign that software can not deal with certain types of information that can be relevant to mimic human thought. A good example is that you can not represent pi in a finite machine, you can only calculate an approximation that is not pi.

[pteranodon]

To me this is nonsense; even if it is necessary to have quantum effects moderated by nanosized cytostructures before you can have consciousness, then what is stopping us incorporating such structures into our hypothetical AI machines?

The problem is that any computer software is a mathematical construct. Then you might question: "Can our thinking processes be a mathematical construct too?". Well, at least with the mathematics we created, I fear this question will remain unanswered. Theory of computation suggests that a system of axioms can not entirely recreate its creator's thinking process.

[milli360]

pteranodon:

Proving that there is something that can't be proved, is a lot different than proving that a particular thing can't be proved.

Suppose we have a set called A, suppose we have a subset of A called B. If there is an operation undefined for A, the same operation is not defined for B too. If I prove something wrong for all elements of A, I do not need to perform the same proof for the elements of B.

My point is that every limitations of Mathematics is also a limitation of AI, Algebra, Calculus, et cetera.
Fine, but you have yet to demonstrate that mathematics is incapable of artificial intelligence. :)


But we are talking about AI...does it or does it not include passion and subjectivity?

Depends on the specific element. If it is a problem not solvable by a sequence of procedures or a set of rules, then AI will never be able to deal with it. One must remember that all operations in a computer are objective: given an input, the output will always be the same.
Clearly, time of day could be part of the input, so we might never even see the same output.

[eburacum45]

The problem is that any computer software is a mathematical construct. Then you might question: "Can our thinking processes be a mathematical construct too?".

I am certain that they are; we just do not know what the programs are yet.
But there is nothing to suggest that the programs our consciousness runs on are not reproducible; and when they are reproduced, Godel's Incompleteness theorem will have exactly the same relevance to an artificial consciousness as it does to a biological one.


Well, at least with the mathematics we created, I fear this question will remain unanswered. Theory of computation suggests that a system of axioms can not entirely recreate its creator's thinking process.

Nothing can be reproduced exactly; but there is nothing to stop a recreation being effectively indistinguishable to all possible methods of human investigation.
If it walks like a duck and quacks like a duck, it is effectively a duck.

Once a human level Artificial Intelligence is acheived, the next step- probably the very next day- will be an AI of greater ability than a human.

[pteranodon]

Fine, but you have yet to demonstrate that mathematics is incapable of artificial intelligence. :)


I have never refuted AI itself as a field for study and resaerch. AI responses will always be just automated responses.

[milli360]

pteranodon:
AI responses will always be just automated responses.
That's just your opinion. I assume that is to differentiate from human intelligence responses--but that is the crux of what we are trying to decide. We don't know the answer yet.

[pteranodon]

The problem is that any computer software is a mathematical construct. Then you might question: "Can our thinking processes be a mathematical construct too?".

I am certain that they are; we just do not know what the programs are yet.
But there is nothing to suggest that the programs our consciousness runs on are not reproducible; and when they are reproduced, Godel's Incompleteness theorem will have exactly the same relevance to an artificial consciousness as it does to a biological one.

I guess you have not read Gödel's proof yet, right? It is a theorem, proved true and aknowledged by the scientific community, it is not hype.

Well, at least with the mathematics we created, I fear this question will remain unanswered. Theory of computation suggests that a system of axioms can not entirely recreate its creator's thinking process.

Nothing can be reproduced exactly; but there is nothing to stop a recreation being effectively indistinguishable to all possible methods of human investigation.
If it walks like a duck and quacks like a duck, it is effectively a duck.

Once a human level Artificial Intelligence is acheived, the next step- probably the very next day- will be an AI of greater ability than a human.

There is no evidence of this being feasable, and there is more: it has been mathematically proved that AI will never be greater than ours.

[milli360]

pteranodon:
it has been mathematically proved that AI will never be greater than ours.
No, it hasn't

[pteranodon]

That's just your opinion. I assume that is to differentiate from human intelligence responses--but that is the crux of what we are trying to decide. We don't know the answer yet.

Not my opinion, it is a fact as all computer programs, including AI applications, are deterministic systems, something that we are way beyond.

[pteranodon]

pteranodon:
it has been mathematically proved that AI will never be greater than ours.
No, it hasn't

So, I guess you have proven Incompleteness Theorem wrong.

[milli360]

pteranodon:
it has been mathematically proved that AI will never be greater than ours.
No, it hasn't

So, I guess you have proven Incompleteness Theorem wrong.
That's not what the Incompleteness Theorem says.

[pteranodon]

There is much hype around quantum computing promising "machine consciousness", et cetera. Quantum computing will be only a change in the switching technology and we use today.

[pteranodon]

pteranodon:
it has been mathematically proved that AI will never be greater than ours.
No, it hasn't

So, I guess you have proven Incompleteness Theorem wrong.
That's not what the Incompleteness Theorem says.

How not? It says that there are problems not solvable by Mathematics, AI is a mathematical solution, therefore AI is subject to Incompleteness Theorem. #-o

[Demigrog]

pteranodon:
it has been mathematically proved that AI will never be greater than ours.
No, it hasn't

So, I guess you have proven Incompleteness Theorem wrong.
That's not what the Incompleteness Theorem says.

How not? It says that there are problems not solvable by Mathematics, AI is a mathematical solution, therefore AI is subject to Incompleteness Theorem. #-o

In the end, a human mind is nothing but mathematics as well. Any seeming ability to leap beyond logical thinking is the result of noise in the environment causing random shifts in thought. Machine intellegence can do the same thing with a random number generator.

I've spent a lot of time working on machine intellegence (mostly as AI for games and controls for robotics), and I've come to the conclusion that there are several key advantages of the human mind over current generation computers:
A) Asyncronous, parallel processing. The human brain is really a combination of a great many, small computers linked together. It is somewhat tricky to achieve the same in computers still.
B) An advanced object recognition engine. The human brain is capable of identifying objects from sensory input with much greater ease than our most advanced computer algorithms (possibly due to many specialized computers doing the work). However, the human brain is not particularly accurate at it, whereas we demand our software to be reliable (otherwise, why use a computer?)
C) Human minds get a lifetime to build themselves. Out of the box they are not particularly useful, but give them 10-20 years and they can do a heck of a lot. With a proper learning algorithm, a computer given the same amount of time and environment, could learn as much or more

These advantages are not insurmountable; improving algorithms will eventually erase them.

Computers have a clear edge in some areas:
A) Raw number crunching
B) Memory capacity and reliability
C) Durability, endurance, and reproducability
D) Self-modification

The remaining aspects of being "Human"-- emotion, conciousness, etc, are all easily simulated in software.

[worzel]

To use Godel's incompleteness theorom in relation to AI is just silly. The incompleteness theorom simply states that a formal system capable of expressing arithmetic must contain valid sentences that are unprovable within that system (even though we can see that they must be true). Penrose's idea was that we must contain some magical mathematical truth detector because of this on the basis that if we can be simulated then we are effectively formal systems but because we can see that the Godel statement of formal system x is G we must be more than a formal system (instead of the correct interpretation which is that we are more than formal system x).

Either you believe that we are ultimately reducable to the laws of physics, in which case we can in principle be simulated or you believe in some super-natural soul stuff in which case you're basically religious.

[milli360]

pteranodon:
How not? It says that there are problems not solvable by Mathematics, AI is a mathematical solution, therefore AI is subject to Incompleteness Theorem. #-o
There are problems not solvable by humans. Pauly Shore, for one.

[eburacum45]

My OA associate, Ben Higginbottom, has a good understanding of this misapplication of Godel to the concept of intelligence; he has supplied this link
http://cscs.umich.edu/~crshalizi/notebooks/godels-theorem.html
with this particularly interesting passage;

This brings us to the other, and possibly even more common fallacy, that Gödel's theorem says artificial intelligence is impossible, or that machines cannot think. The argument, so far as there is one, usually runs as follows. Axiomatic systems are equivalent to abstract computers, to Turing machines, of which our computers are (approximate) realizations. (True.) Since there are true propositions which cannot be deduced by interesting axiomatic systems, there are results which cannot be obtained by computers, either. (True.) But we can obtain those results, so our thinking cannot be adequately represented by a computer, or an axiomatic system. Therefore, we are not computational machines, and none of them could be as intelligent as we are; quod erat demonstrandum. This would actually be a valid demonstration, were only the pentultimate sentence true; but no one has ever presented any evidence that it is true, only vigorous hand-waving and the occasional heartfelt assertion.

Gödel's result is of course quite interesting, if you're doing mathematical logic, and it even has some importance for that odd little specialization of logic, the theory of computation. (It is intimately related to the halting problem, for instance.) It also makes a fine piece of general mathematical culture; but it doesn't shake the foundations of the house of intellect, or exalt us above all else that greps.


Ben continues;For those who think they understand Axioms - an axiom is a self
explanatory statement that is used as the basis for theoretical
arguments _ONLY_. They are so basic that it is very unlikely for them
to be incorrect (eg: if I add two numbers together, it doesnt matter
which way round they are ordered), however as the possibility exists
that they may not be completely correct it is possible that error
will exist in any calculation using them. It is this what the worlds
greatest logician figured out before collapsing into paranoid
delusions.

However there exist an entirely different class of theories based on
observation and _measurement_, Godel does _NOT_ apply to these as
they have no need of axioms. I'd recommend Godel, Escher and Bach by
Douglas Hofstader to help really understand the theorem

[worzel]

That's just your opinion. I assume that is to differentiate from human intelligence responses--but that is the crux of what we are trying to decide. We don't know the answer yet.

Not my opinion, it is a fact as all computer programs, including AI applications, are deterministic systems, something that we are way beyond.
How can you possibly assert as if proven fact that we are not deterministic? And do you think a bit of quantum randomness somehow gives rise to intelligence not achievable deterministically? IDNet - Intelligence by Dice?

[worzel]

Every system of mathematics based on axioms contains statements which are inconsistent with that system; this is incompleteness, and it applies to human thought in exactly the same way it applies to hypothetical machine thought.
No it's not, that's unsoundness. If a system contains a proof for x and a proof for not x then it is unsound and anything within that system can be proven (or disproven) and wouldn't normally be classed as interesting.

Godel's incompleteness theorom proves that any sound system (capable of arithmetic) is necessarily incomplete. That is, not every proposition within the system can be proved to be either true or false within the system. Propositional logic is both sound and complete. Anything with a bit more umph is generally not.

Of course, any system can be amended to add the undecidable G as an axiom but then you have a new system with it's on Godel statement. This goes on for ever no matter how cunningly you try to get all the Gs in. But why do so many people take it as an article of faith that human thought is both sound and complete is beyond me. Just go to GLP for a clear demonstration that this is not true.

[Jpax2003]

I think we may eventually discover via quantum theory that the entire universe is a mathematical construct and that everything within is explained, including human thought patterns.

I do not think so. The simple fact we can not translate into a fixed number the position of an electron, for instance, is a sign that software can not deal with certain types of information that can be relevant to mimic human thought. A good example is that you can not represent pi in a finite machine, you can only calculate an approximation that is not pi.Can you post a URL that you recommend as a basis for my reasearch into this subject?

I don't take it as fact that we can't translate the position of an electron (are you referring to Heisenberg?), at least not as a permanent situation. Furthure more, how does that suggest that software will be troubled by it? I'm not sure what you mean, it sounds like you said that because we can't do something, machines also can't do that same thing, therefore they are not like us... which doesn't follow. Besides I don't know how exactly it is that knowing the position of an electron is necessary to thought. I don't deny it, but neither do I admit it. Sure machines use electrical processes, and we use electrochemical processes, what's your point?

I can't define pi other than as an approximation either. I can represent Pi as a symbol, or as a formula which is not meant for computation: circumference/diameter .Perhaps we have a misunderstanding that coding computers is purely mathematical. When I wrote code, it involved languages. I think of "thinking" as occuring in a layer above the basic mathematics. I don't think about which neuron to fire, it just happens. I posit that an artificial intelligence could parse pi as a symbol, formula, or approximation without actually calculating for it. If machines had separate processing subsystems like we do with our cerebrum and cerebellum and other nervous system components then we might get closer. Shunting every integer to the main processor is rediculous.

Perhaps we should have a definition of artificial intelligence. I'm not sure what the experts in the field would call it, but I would think artificial means a simulant of human faculty and capacity. If we were able to make a machine that could match us and was truly conscious I would call that synthetic intelligence. This does not mean we could not find natural occurences of this either, I suppose that would be called a natural silicon intelligence, and we would be a natural carbon intelligence. This is just an idea, but perhaps it helps to make sure we are all dealing with the same concepts.

[pteranodon]

In the end, a human mind is nothing but mathematics as well. Any seeming ability to leap beyond logical thinking is the result of noise in the environment causing random shifts in thought.

There is no evidence human thought processes can be expressed by a sequence of rules or a set of axioms, as would be required for a software to mimic it.

Machine intellegence can do the same thing with a random number generator.

1) The random number generator in a computer is not truly random. As a matter of fact it gives the same sequence for a given "seed".

2) A random number generation is not an intelligent process.

"I've spent a lot of time working on machine intellegence (mostly as AI for games and controls for robotics), and I've come to the conclusion that there are several key advantages of the human mind over current generation computers:
A) Asyncronous, parallel processing. The human brain is really a combination of a great many, small computers linked together. It is somewhat tricky to achieve the same in computers still.
B) An advanced object recognition engine. The human brain is capable of identifying objects from sensory input with much greater ease than our most advanced computer algorithms (possibly due to many specialized computers doing the work). However, the human brain is not particularly accurate at it, whereas we demand our software to be reliable (otherwise, why use a computer?)
C) Human minds get a lifetime to build themselves. Out of the box they are not particularly useful, but give them 10-20 years and they can do a heck of a lot. With a proper learning algorithm, a computer given the same amount of time and environment, could learn as much or more"

They still will be deterministic systems with automated responses and limited by what Mathematics can do.

These advantages are not insurmountable; improving algorithms will eventually erase them.

Computers have a clear edge in some areas:
A) Raw number crunching
B) Memory capacity and reliability
C) Durability, endurance, and reproducability
D) Self-modification

Agreed.

The remaining aspects of being "Human"-- emotion, conciousness, etc, are all easily simulated in software.

Again, there is no evidence the problems you cited are solvable by the axioms of mathematics.

[pteranodon]

To use Godel's incompleteness theorom in relation to AI is just silly.

Why? Sorry, talking in absolutes will not help to make your point.


The incompleteness theorom simply states that a formal system capable of expressing arithmetic must contain valid sentences that are unprovable within that system (even though we can see that they must be true). Penrose's idea was that we must contain some magical mathematical truth detector because of this on the basis that if we can be simulated then we are effectively formal systems but because we can see that the Godel statement of formal system x is G we must be more than a formal system (instead of the correct interpretation which is that we are more than formal system x).

In other words, there are problems not solvable by our current mathematical axioms.

Either you believe that we are ultimately reducable to the laws of physics, in which case we can in principle be simulated or you believe in some super-natural soul stuff in which case you're basically religious.

You cited the religious thing for a reason I ignore. Well... continuing...

There are natural problems not solvable by a computer system, like determining the position of an electron. Perhaps there are other particle interactions not accurately solvable by computing devices involved in our thinking processes.

[milli360]

pteranodon:
There is no evidence human thought processes can be expressed by a sequence of rules or a set of axioms, as would be required for a software to mimic it.
If there were evidence that it could, the problem would be solved. There's no evidence that it can't, however.

[eburacum45]

Every system of mathematics based on axioms contains statements which are inconsistent with that system; this is incompleteness, and it applies to human thought in exactly the same way it applies to hypothetical machine thought.
No it's not, that's unsoundness.

Oops; thank you for clearing that up.

Propositional logic is both sound and complete. Anything with a bit more umph is generally not.
But propositional logic is supposedly forever beyond the grasp of machine intelligence? My gut feeling tells me that is not true.
But why do so many people take it as an article of faith that human thought is both sound and complete is beyond me.

Human thought (and no doubt other forms of animal thinking) has fuzzy qualities like common sense, guesstimation, fudging, discrimination; these qualities are going to be difficult to emulate using computer programs, which is why I don't think AI is just around the corner.

But I am certain they can be emulated. sooner or later; then it will be possible for machines to start to develop sentience.

... it has been mathematically proved that AI will never be greater than ours.

Well, the human brain is limited to a couple of kilograms of biomass, and at the maximum about 10e17 states; this is a complexity far beyond today's computers, but that will not always be the case.
If a human level intelligence can be emulated, then the next stage is to increase the size and speed of the system emulating the human; link a few thousand or a million of these emulations together in series and in parallel, and integrate databanks, peripherals, and search engines; the entity or entities that would result from such a link up would be remarkable.
Incorporate features like autoscience, where the AI entity can monitor all its own internal states, and analyse and record the reasons for each decision, whether using formal logic or artificial 'gut feelings'; total recall; separation into multiple agents for multitasking followed by reintegration at a later date; autoevolution, the ability to write new software and design new hardware for itself...

If artificial intelligence is possible at all (and it almost certainly is) it seems very likely that in a century or three AI will be far beyond the human level.

[worzel]

In other words, there are problems not solvable by our current mathematical axioms.
Indeed, or even by any set of axioms. But how does that prove that AI cannot achieve human like intelligence?

There are natural problems not solvable by a computer system, like determining the position of an electron. Perhaps there are other particle interactions not accurately solvable by computing devices involved in our thinking processes.
Perhaps, but we don't know for sure. Anything going on in the brain that is not deterministic seems unlikely to me to be usefully harnessable by natural selection in building a thinking tool.

Human thought (and no doubt other forms of animal thinking) has fuzzy qualities like common sense, guesstimation, fudging, discrimination; these qualities are going to be difficult to emulate using computer programs, which is why I don't think AI is just around the corner.
I don't think we're likely to build an intelligent machine by simply sitting down and writing a program with a routine for intuition, a routine for deep thought etc. While I don't think it's impossible in principle (and it certainly has not been mathematically proven to be) I think we are a long way off from anything that could be considered sentient.

[pteranodon]

This brings us to the other, and possibly even more common fallacy, that Gödel's theorem says artificial intelligence is impossible, or that machines cannot think. The argument, so far as there is one, usually runs as follows. Axiomatic systems are equivalent to abstract computers, to Turing machines, of which our computers are (approximate) realizations. (True.) Since there are true propositions which cannot be deduced by interesting axiomatic systems, there are results which cannot be obtained by computers, either. (True.) But we can obtain those results, so our thinking cannot be adequately represented by a computer, or an axiomatic system. Therefore, we are not computational machines, and none of them could be as intelligent as we are; quod erat demonstrandum. This would actually be a valid demonstration, were only the pentultimate sentence true; but no one has ever presented any evidence that it is true, only vigorous hand-waving and the occasional heartfelt assertion.

Gödel's result is of course quite interesting, if you're doing mathematical logic, and it even has some importance for that odd little specialization of logic, the theory of computation. (It is intimately related to the halting problem, for instance.) It also makes a fine piece of general mathematical culture; but it doesn't shake the foundations of the house of intellect, or exalt us above all else that greps.

I have never stated AI is impossible. My point in resorting to Gödel's theorem is that there will always be problems not solvable by a system based on mathematical axioms, which include computer software. Thus, AI will never be more intelligent than us.

[pteranodon]

For those who think they understand Axioms - an axiom is a self
explanatory statement that is used as the basis for theoretical
arguments _ONLY_. They are so basic that it is very unlikely for them
to be incorrect (eg: if I add two numbers together, it doesnt matter
which way round they are ordered), however as the possibility exists
that they may not be completely correct it is possible that error
will exist in any calculation using them. It is this what the worlds
greatest logician figured out before collapsing into paranoid
delusions.

However there exist an entirely different class of theories based on
observation and _measurement_, Godel does _NOT_ apply to these as
they have no need of axioms. I'd recommend Godel, Escher and Bach by
Douglas Hofstader to help really understand the theorem

I know what an axiom is. I cited axioms because they are the logical atoms which all theorems are made of.

Of course, many scientists are mad at Gödel beacuse he successfully demonstrated that the scientific goal of mathematically describing everything is utopical.

[pteranodon]

How can you possibly assert as if proven fact that we are not deterministic? And do you think a bit of quantum randomness somehow gives rise to intelligence not achievable deterministically? IDNet - Intelligence by Dice?

The problem is that things at quantum level are not always deterministic. At that scale, you can say a particle X will have a probability of 90% to be between points A and B, for instance. This is good enough to build a simulator, but not an emulator.

Anyway we need nuclear medicine to examine all the parts involved in thought process.

[pteranodon]

I don't take it as fact that we can't translate the position of an electron (are you referring to Heisenberg?), at least not as a permanent situation. Furthure more, how does that suggest that software will be troubled by it? I'm not sure what you mean, it sounds like you said that because we can't do something, machines also can't do that same thing, therefore they are not like us... which doesn't follow. Besides I don't know how exactly it is that knowing the position of an electron is necessary to thought. I don't deny it, but neither do I admit it. Sure machines use electrical processes, and we use electrochemical processes, what's your point?

Of course I was not refering to the information carriers themselves. My point is how good can a mathematical model be? Will it allow emulation or just simulation?

I just mentioned the electron as a didactic resource. Perhaps there are other particles involved subject to Heisenberg principle. If so, any attempt to emulate human thought in a computer will be ruined.

[pteranodon]

pteranodon:
There is no evidence human thought processes can be expressed by a sequence of rules or a set of axioms, as would be required for a software to mimic it.
If there were evidence that it could, the problem would be solved. There's no evidence that it can't, however.

Ok, we do not have evidences for both cases. However, there is a theorem proving the limitations of mathematical models, which is far more concrete than beliefs, hopes and opinions.

[pteranodon]

If a human level intelligence can be emulated, then the next stage is to increase the size and speed of the system emulating the human.

The question is: Will it be an emulation or just a simulation? The answer relies on the progress of nulcear medicine, as we need to identify all the parts involved in human thought. If our mind relies on quantum particles, then we will have to face serious problems due to Heisenberg principle. In this scenario we will only be able to build simulators rather than emulators.

[pteranodon]

If artificial intelligence is possible at all (and it almost certainly is) it seems very likely that in a century or three AI will be far beyond the human level.

I avoid make predictions like that. A famous author predicted we would have sentient machines aboard giant spaceships wading through the solar system by now...

[pteranodon]

Indeed, or even by any set of axioms. But how does that prove that AI cannot achieve human like intelligence?

The answer will depend on whether our mind depends on quantum particles or not.

Anything going on in the brain that is not deterministic seems unlikely to me to be usefully harnessable by natural selection in building a thinking tool.

We, humans, are too arrogant to admit there is something in Nature not explainable by our science. If we stumble with limitations to our knowledge we scream "This is not possible!" or "Damn! There is something wrong with these stupid particles! They do not cooperate with my models!".

[eburacum45]

A famous author predicted we would have sentient machines aboard giant spaceships wading through the solar system by now...


:) I know, I know.

AI seems to be like fusion in that respect; it is always twenty years away. Or forty.
or a hundred; estimates of the timescale for AI, fusion and the Mars landing are constantly being revised to be further in the future.


It may be the case that AI is impossible; most of us will never know.

[pteranodon]

I don't think we're likely to build an intelligent machine by simply sitting down and writing a program with a routine for intuition, a routine for deep thought etc.

Well, our current computational paradigm requires writing code to build software, after you mathematized your problem.

[milli360]

pteranodon:
pteranodon:
There is no evidence human thought processes can be expressed by a sequence of rules or a set of axioms, as would be required for a software to mimic it.
If there were evidence that it could, the problem would be solved. There's no evidence that it can't, however.

Ok, we do not have evidences for both cases. However, there is a theorem proving the limitations of mathematical models, which is far more concrete than beliefs, hopes and opinions.
Exactly why it doesn't really apply. We humans have limitations too. I'm pretty sure of that. :)
pteranodon:
The question is: Will it be an emulation or just a simulation?
That's another way of expressing the question. Which we haven't answered, yes or no, yet.

[eburacum45]

worzel wrote:
Indeed, or even by any set of axioms. But how does that prove that AI cannot achieve human like intelligence?

pteranodon wrote:
The answer will depend on whether our mind depends on quantum particles or not.

And as I said before, what is stopping us incorporating these same quantum effects into our AI?

[pteranodon]

AI seems to be like fusion in that respect; it is always twenty years away. Or forty.
or a hundred; estimates of the timescale for AI, fusion and the Mars landing are constantly being revised to be further in the future.


It may be the case that AI is impossible; most of us will never know.

Indeed. :-k

Off-topic: Do you mean cold fusion when you talk about fusion?

[eburacum45]

Sorry, no; it seems likely that many years will pass before sustainable hot fusion will be available in commercial power plants.
but cold fusion will never be a viable option in my opinion.

[Demigrog]

There is no evidence human thought processes can be expressed by a sequence of rules or a set of axioms, as would be required for a software to mimic it.

You’re right in that human intelligence cannot directly be simulated; we can only provide the “platform” and train an AI mind in the same way that we train human children—otherwise we’re merely creating simulations of human minds and not real human minds.

The mechanisms of the brain are understood at least well enough to simulate—hence the whole field of AI. The problem is that human intelligence is an emergent property of a very large system where the interconnections are not well understood. Sooner or later they will be. Personally, I find simulating “real” human minds to be ethically questionable and of doubtful usefulness (there is a much simpler (and more fun!) way of doing it ;)).


1) The random number generator in a computer is not truly random. As a matter of fact it gives the same sequence for a given "seed".

Irrelevant to the question at hand, and also untrue as a general statement. True random numbers can be generated by dedicated hardware, and in fact many modern CPUs have such hardware for use in encryption applications. Pure software, of course, can only generate pseudorandom numbers, but for the purposes of machine intelligence that is quite sufficient anyway.


2) A random number generation is not an intelligent process.

Not itself, of course. The random number generator is only part of the system. For example, Genetic Algorithms use random numbers to drive mutation, crossover, and other genetic operations. This allows GAs to create novel solutions to many types of problems. In fact, in my own AI-like systems I typically use GAs to evolve new rules for complex classfier systems controlling various processes. Do a little research on bipedal robotics and genetic algorithms, and you’ll see a good example of modern machine learning.

Biology has a similar process; the randomness in the growth and interconnection of cells, coupled with random chemical interactions and electrical impulses, provides a mechanism for non-deterministic problem solving.


They still will be deterministic systems with automated responses and limited by what Mathematics can do.

Well, unless you start getting into religious topics (and I’m simply not going to), there is nothing in nature that cannot be described mathematically. So, I fail to see that as a limitation. As for determinism, that gets back into the role of random numbers and “exploratory” algorithms like GAs. It is actually quite simple to create non-deterministic software.

Of course, the applications of AI usually want deterministic responses anyway; nobody wants a robot to occasionally trip over its own feet, and that will inevitably happen in a system modeled after biology. That is one reason that AI has not really taken off for industrial control—no (sane) engineer trusts a non-deterministic control process.


The remaining aspects of being "Human"-- emotion, conciousness, etc, are all easily simulated in software.

Again, there is no evidence the problems you cited are solvable by the axioms of mathematics.

Well, I’ll admit there is some philosophical ground to cover when discussing simulation of abstract concepts like emotion and consciousness. However, it is pretty easy to simulate the effects of those concepts in an AI. Just go play the Sims for a while for a good example.

Emotional states can be abstracted by simple numbers representing need (negative numbers) and fulfillment (positive numbers). These values (and associated non-static coefficients for prioritization) are inputs into a classifier system. Consciousness is simply the current state of the classifier system. Sensory inputs and feedback from “bodily functions” are also inputs into the classifier system. Control of voluntary systems (ie muscles) would be the output of the classifier system.

I have implemented systems like this in game AI before. The only real limitations on its capabilities are the processing power of the computer running it and the amount of time you’re willing to spend training it—like any other kid. Of course, it is not remotely human-like, as it has a completely different environment and development history.

Sheesh, I’m long winded today, but you’ve happened on to one of my major hobbies and research subjects.

On a somewhat related topic, I’ve often considered whether it is ethical to stop running an AI program once it has started, and what level of self-awareness would make it unethical. Big questions to consider.

[mutineer]

I should like to thank those who have responded to this thread, who have taken the total number of posts above 100, and who have provided some interesting reading.

I was pretty sure that the vote was going to go the way of those who reckoned "sentient silicon" was a possibility. Of course, I realise that the way I stated the options was in terms of "machine" consciousness, which allowed the possibility of a machine far removed from the conventional von Neumann computer or its near variants - i.e. from the sort of technology implied by the word "silicon".

So among those who produced the majority vote, there may be very different opinions between two groups:

Group 1: those who believe that sentience can be achieved by conventional algorithmic computing buttressed by massive parallel processing, neural networks, randomization, learning strategies, etc. It's just a matter of getting the architecture right and writing the right program. Learning from human brain function may play a major role.

Group 2: those who believe that consciousness is beyond the scope or understanding of present day physics, but who are certain that it cannot be achieved by anything we would currently recognize as a computer. However, consciousness is not separate from the physical world, and in due course we will get to understand it, and hence we will be able to construct sentient machines. The discipline of quantum physics may provide our route to understanding.

Is that a fair summary of the two positions (within the majority camp)? I suppose I really need a fresh poll now, with three options!

I would be particularly interested to hear from those who, like Marvin Minsky whom I quoted at the outset, defend the Group 1 position. (After all, if they are not right, we may as well give up on a sentient computer for a very very long time!)

[kylenano]

When I mentioned this poll to my partner, who is a lecturer in Artificial Intelligence, his reaction was "What do you mean by consciousness?". He doesn't like The Emperor's New Mind by Roger Penrose at all!

I wouldn't like to say for sure that there would never be machine consciousness - though that raises the question, as well as consciousness, what do you mean by a machine? Try to define these terms gets a bit difficult - the sort of thing you know what it is until you try to put it into words. There is still so much we don't know about how the human brain works, which must make trying to simulate it difficult.

Anyone read Gödel, Escher, Bach - The Eternal Golden Braid by Douglas Hofstadter? I could only manage small doses at a time! (I haven't seen it mentioned so far, but I could have missed it...)

[milli360]

kylenano:
Anyone read Gödel, Escher, Bach - The Eternal Golden Braid by Douglas Hofstadter? I could only manage small doses at a time! (I haven't seen it mentioned so far, but I could have missed it...)
eburacum45 (http://www.badastronomy.com/phpBB/viewtopic.php?p=296949#296949) mentioned it a few posts back, or at least his friend Ben did.

I read it twenty-five years ago.

[worzel]

If either humans or machines are able to think outside of axiomatic mathematical systems then the Incompleteness theorm does not apply.

The error of this statement is that software is a sequence of mathematical constructs, thus subject to Incompleteness Theorem.

Whereas humans are sound and complete and therefore not? Thinking outside of a formal system (and seeing the truth of its Godel statement) is not the same as thinking outside of your own formal system (should such a thing exist).

How can you possibly assert as if proven fact that we are not deterministic? And do you think a bit of quantum randomness somehow gives rise to intelligence not achievable deterministically? IDNet - Intelligence by Dice?

The problem is that things at quantum level are not always deterministic. At that scale, you can say a particle X will have a probability of 90% to be between points A and B, for instance. This is good enough to build a simulator, but not an emulator.

The question was whether machines could be conscious, not whether they could emulate rather than simulate humans. Either you're saying that randomness is essential, which is easy enough to build in, or you're saying that there is something magical going behind the scenes of quantum interactions that gives rise to consciousness.

We, humans, are too arrogant to admit there is something in Nature not explainable by our science. If we stumble with limitations to our knowledge we scream "This is not possible!" or "Damn! There is something wrong with these stupid particles! They do not cooperate with my models!".
I think the arrogance lies with people who think that there must be something more to human thought than can be accounted for with physics and evolution. The magical soul you pine for is gettting smaller and smaller as we learn more about the brain and psychology - we now have people basically claiming that our soul is in quantum state reductions, something that as far as we can tell is completely random???

[milli360]

worzel:
The magical soul you pine for is gettting smaller and smaller as we learn more about the brain and psychology
How can you tell? :)

My soul is nearly infinite.

[Demigrog]

worzel:
The magical soul you pine for is gettting smaller and smaller as we learn more about the brain and psychology
How can you tell? :)

My soul is nearly infinite.

I traded my soul to a friend for lunch money about four years ago. He keeps the IOU around to mock me; someday I'm going to steal it back and put it through the paper shredder.

Kidding aside, I tend to think of a soul in a more physical sense, as the unique combination of neural connections that make me me. That combination is unique in the universe and yet is not static; you could duplicate the exact makeup of my brain right down to the quantum states, and yet it still will only be a snapshot--my consciousness already occupies a new point in spacetime and is thus still unique. The copy will still never be me, it will become someone else.

Indeed, that may be the true test of soul, that a mind is unique in spacetime and can only be unique. Many computer intelligences would fail the test, because they only interact with a simulated (and very limited) spacetime that can be reproduced at any time from some storage mechanism. Once you let the machine interact with real spacetime, however, it becomes as unique as a real human mind.
[/philosophical rant]

[Chuck]

When future computers are discussing whether or not humans are truly sentient I hope they'll give us the benefit of the doubt.

[Jpax2003]

When future computers are discussing whether or not humans are truly sentient I hope they'll give us the benefit of the doubt.Don't you mean they'll be discussing if we were sentient?