Some people seem to have some strange ideas about what is being done here. The reality is far less interesting:
Say you want to know what sort of image would result in “Banana.” Start with an image full of random noise, then gradually tweak the image towards what the neural net considers a banana [...] By itself, that doesn’t work very well, but it does if we impose a prior constraint that the image should have similar statistics to natural images, such as neighboring pixels needing to be correlated.
Even less exciting, naturally, are the swirly ones and the ones with the mixed images. (I presume the anthropomorphism in the quote is to tickle the Kurzweil nuts and "science" reporters.) Very few difference, but it makes prettier pictures -- start with a photo instead of noise, tweak it differently.
In this case we simply feed the network an arbitrary image or photo and let the network analyze the picture. We then pick a layer and ask the network to enhance whatever it detected. Each layer of the network deals with features at a different level of abstraction, so the complexity of features we generate depends on which layer we choose to enhance. For example, lower layers tend to produce strokes or simple ornament-like patterns, because those layers are sensitive to basic features such as edges and their orientations.
Don't read the comments after reading the article. You'll lose all faith in humanity.
Next up: The amazing Markov chain AI that composes incredible music!
in what way would that computer not understand blocks and milling?
The question is how could it? All the computer gets is symbols -- the best it can do is relate those symbols to other symbols. The meaning isn't inherent or carried along with them, after all. You may want to look in to the symbol grounding problem. That should help clear things up a bit for you.
I do have to explain this to you. You'll find that the majority of those disagree with the CRA, completely agree with the essential part. Harnad, for example, would be included among those. David Chalmers also opposes the CRA (see his laughable paper on subsymbolic computation) while completely accepting the argument. (If this seems unclear, consider that everyone who denies the CRA on some variation of the robot reply necessarily accepts the argument.)
It's a strange thing indeed. It's very rare to find someone who thinks the CRA is wrong who doesn't also accept the argument!
It's probably why you'll find so few condemnations that actually address the crux of the argument (syntax is insufficent for semantics). Most CRA opponents accept that premise without question. Chalmers and the Churchlands are notable here in that they actually attempt to address it. Chalmers thinks it can be avoided, and the Churchlands (through literal hand-waving) simply deny it.
On the sidelines, where you'll find magical thinkers offering variations of the systems reply, there's a strange sort of denial. To accept the systems reply is also to conceded the argument entirely as it necessarily introduces non-computational aspects. For the systems reply folks, computation is necessary but not sufficient. (To remind you: All I've claimed is here is that computation alone is insufficient. )
Hopefully, that should make it clear to you that while many oppose the CRA very few actually disagree with the argument.
Warning: if you bring up Searle and the Chinese Room, I'm going to tear it to pieces
I seriously doubt that. From what I can tell from your other posts, you don't even have a basic understanding the argument. Though if you really don't want to talk about that, there are other approaches we could take. Computationalism is dead for more than one reason, after all.
I'm suggesting that computation alone will likely be sufficient.
It's a fine belief. Kurzweil has made a good living selling that belief to his followers. The problem, of course, is justifying it. If I pray to Ray hard enough, will I become a true believer as well?
As I'm not interested in weird sci-fi religions, I'd expect you to be able to show computation to be sufficient. Now, I have very strong reasons to believe that it is impossible. I also know that if you could accomplish such a feat, then fame and fortune await you. Needless to say, I have no expectations.
Searle is completely discounting the idea of an emergent phenomenon.
On the contrary, that is precisely what Searle proposes.
Searle doesn't even try escaping into any sort of mysticism.
Why would he? It doesn't seem necessary to his argument.
He claims that we know that intentionality is biological, which is not actually something I've found anywhere else
It's true that that conclusion based on a set of metaphysical assumptions; though they're also the same assumptions physicists necessarily operate under. You can disagree with those, which is fine, but it doesn't alter Searle's argument at all. You can accept it or reject it, however, in either case, you still need more than mere computation.
Searle does get into the semantics vs. syntax distinction in one of these digressions, but doesn't really develop it.
? That's the bulk of the argument. The CRA is designed to illustrate the fact that syntax, by itself, is insufficient for semantics. I think you have things reversed.
Which is, of course, silly nonsense, as I explained earlier. You introduce cameras, for example, but cameras don't magically give a computer sight. All the input from a camera consists of mere symbols -- indistinguishable from any other input. In the context of the Chinese Room, rather than the proposed video screen, the data flowing in from the camera would simply be more symbols. Symbols, obviously, no more infused with semantic content than any other set of symbols.
It's a rhetorical trick, though not a very good one as it's trivially easy to expose. It's why no one takes the robot reply seriously. It concedes the argument, and hopes no one will notice.
AGI does not imply consciousness, intentionality, qualia, or any of the other difficult issues in philosophy of mind - it's only about the ability to preform tasks.
Sense when? "AGI" has been synonymous with "strong AI" since the term's inception.
you seem to think that it was some kind of knock-down, indisputable argument.
Well, so far, no one has been able to offer a satisfactory reply. As an argument against computationalism, it's quite convincing. There's a reason it's still seriously discussed 35 years on. Had that pillar been knocked down, we'd all know the standard reply -- and the world would be a much more interesting place! The implications such a revelation would have for linguistics alone are staggering.
Computionalism is long dead. Searle, in no small part, is responsible for that.
and nowhere (there or in the literature) is Searle's conclusion treated as definitive.
It's not my fault that you're completely unfamiliar with the literature. I mention Stevan Harnad in another post, who's work I'll offer as just one of many counterexamples. (I mentioned him specifically as it's likely other Slashdotters are at least passingly familiar with him.)
If you don't, you'll end up hurt - badly
I had no idea that roving gangs of failing undergraduate philosophy majors were so dangerous!
And boolean logic is strictly insufficient for arithmetic
That's simply not true.
but it cannot connect those words to any kind of real-world phenomenal experience
You're trying to talk about the symbol grounding problem (see Harnad for a discussion about how this relates to Searle's Chinese Room).
But give the man in the room a bunch of photos and sound recordings and scratch-and-sniff panels and so on
This was addressed by Searle in the original 1980 paper. I recommend you read it. Put simply, it's a weak attempt to smuggle semantics in through the back door. Replace the photos, sounds, and smells with some equivalent in the form of symbols and you'll find that nothing changes.
As we have ever-improving image-recognition algorithms for computers. We have computers that can observe empirical phenomena and connect them to verbal symbols.
You're just still connecting symbols to other symbols. Consider this for a moment: I give you a unilingual dictionary and an exhaustive corpus of texts written in the same language. You have all you need to identify relationships between the symbols. The most you could hope to produce would be a grammar. You'd never be able to determine the meaning of any of the words.
There's a reason the argument has stood-up to 35 years worth of very harsh scrutiny. I recommend you read more about it. It will be painful for you, as it's difficult to see much-cherished beliefs crushed by something so simple. Just be glad it wasn't your career, as it was for so many others.
You're a bit off there. The crux of Searle's argument is that syntax, by itself, is insufficient for semantics. A point that is both obvious and seemingly irrefutable. If you believe that you can show otherwise, fame and fortune await.
You can never state that something is not sufficient just because no one has figured it out yet.
That it's logically impossible is reason enough! Consider for a moment a simple example: I claim that it is impossible to clear 5 lines simultaneously in a game of Tetris. Would you say that claim is nonsense and it's only a matter of time before someone figures it out? Of course not. You can clearly demonstrate that it is an impossibility. The same is true for computational approaches to AGI -- they are logically impossible.
Why do you believe in such silly nonsense when it's clear that those beliefs are pure fantasy?
If you'll take a moment to read the comment to which you replied, you'll find that I've said exactly that -- and that that point is completely irrelevant to the fact that computation is insufficient.
I'm sorry that reality does not conform to your fantasy. Please, at least do a little bit of reading. I don't even know how to begin to address your confusion.
I have no idea. Neither does anyone else. That doesn't change the fact that computation alone has been show to be insufficient. That's pretty well established.
No, I'm suggesting, on a rather well-established basis, that computation alone is insufficient. This is all assuming that the mind is a product of the brain. Whatever the brain does to cause consciousness, it can not be by mere computation alone.
It has nothing to do with atheism. You don't need to posit a god. Hell, you don't even need to posit a soul or other supernatural concept. Computation alone is insufficient. To claim otherwise necessitates that you be able to demonstrate that syntactic content is sufficient for semantic content. So simple and obvious is this point, that it can lead you to only one conclusion.
On this particular issue, they're both worthless. AGI by purely computational means simply isn't possible. We've known this for decades. Only lunatics like Kurzweil and the under-informed believe otherwise; a belief has no basis in reality.
You might as well discuss the sociological impact of a zombie apocalypse. It's just as meaningful.
I haven't eaten trans fat in decades and I haven't even been trying. What has transfat in it that you want?
Try finding fortune cookies without... Looked everywhere and gave up. Selfishly I find myself cheering for the FDA ban because I want fortune cookies without trans fat.
You claim to have been able to avoid trans fat for decades without even trying. He wants to buy fortune cookies without trans fat. He claims that they don't exist. The ban means he will be able to buy fortune cookies without trans fat. Something he can not do currently.
That fortune cookies can be made without trans fat is completely irrelevant.
Slashdot Mirror
That's ... not even close to what they did.
Some people seem to have some strange ideas about what is being done here. The reality is far less interesting:
Say you want to know what sort of image would result in “Banana.” Start with an image full of random noise, then gradually tweak the image towards what the neural net considers a banana [...] By itself, that doesn’t work very well, but it does if we impose a prior constraint that the image should have similar statistics to natural images, such as neighboring pixels needing to be correlated.
Even less exciting, naturally, are the swirly ones and the ones with the mixed images. (I presume the anthropomorphism in the quote is to tickle the Kurzweil nuts and "science" reporters.) Very few difference, but it makes prettier pictures -- start with a photo instead of noise, tweak it differently.
In this case we simply feed the network an arbitrary image or photo and let the network analyze the picture. We then pick a layer and ask the network to enhance whatever it detected. Each layer of the network deals with features at a different level of abstraction, so the complexity of features we generate depends on which layer we choose to enhance. For example, lower layers tend to produce strokes or simple ornament-like patterns, because those layers are sensitive to basic features such as edges and their orientations.
Don't read the comments after reading the article. You'll lose all faith in humanity.
Next up: The amazing Markov chain AI that composes incredible music!
in what way would that computer not understand blocks and milling?
The question is how could it? All the computer gets is symbols -- the best it can do is relate those symbols to other symbols. The meaning isn't inherent or carried along with them, after all. You may want to look in to the symbol grounding problem. That should help clear things up a bit for you.
I do have to explain this to you. You'll find that the majority of those disagree with the CRA, completely agree with the essential part. Harnad, for example, would be included among those. David Chalmers also opposes the CRA (see his laughable paper on subsymbolic computation) while completely accepting the argument. (If this seems unclear, consider that everyone who denies the CRA on some variation of the robot reply necessarily accepts the argument.)
It's a strange thing indeed. It's very rare to find someone who thinks the CRA is wrong who doesn't also accept the argument!
It's probably why you'll find so few condemnations that actually address the crux of the argument (syntax is insufficent for semantics). Most CRA opponents accept that premise without question. Chalmers and the Churchlands are notable here in that they actually attempt to address it. Chalmers thinks it can be avoided, and the Churchlands (through literal hand-waving) simply deny it.
On the sidelines, where you'll find magical thinkers offering variations of the systems reply, there's a strange sort of denial. To accept the systems reply is also to conceded the argument entirely as it necessarily introduces non-computational aspects. For the systems reply folks, computation is necessary but not sufficient. (To remind you: All I've claimed is here is that computation alone is insufficient. )
Hopefully, that should make it clear to you that while many oppose the CRA very few actually disagree with the argument.
Warning: if you bring up Searle and the Chinese Room, I'm going to tear it to pieces
I seriously doubt that. From what I can tell from your other posts, you don't even have a basic understanding the argument. Though if you really don't want to talk about that, there are other approaches we could take. Computationalism is dead for more than one reason, after all.
I'm suggesting that computation alone will likely be sufficient.
It's a fine belief. Kurzweil has made a good living selling that belief to his followers. The problem, of course, is justifying it. If I pray to Ray hard enough, will I become a true believer as well?
As I'm not interested in weird sci-fi religions, I'd expect you to be able to show computation to be sufficient. Now, I have very strong reasons to believe that it is impossible. I also know that if you could accomplish such a feat, then fame and fortune await you. Needless to say, I have no expectations.
Searle is completely discounting the idea of an emergent phenomenon.
On the contrary, that is precisely what Searle proposes.
Searle doesn't even try escaping into any sort of mysticism.
Why would he? It doesn't seem necessary to his argument.
He claims that we know that intentionality is biological, which is not actually something I've found anywhere else
It's true that that conclusion based on a set of metaphysical assumptions; though they're also the same assumptions physicists necessarily operate under. You can disagree with those, which is fine, but it doesn't alter Searle's argument at all. You can accept it or reject it, however, in either case, you still need more than mere computation.
Searle does get into the semantics vs. syntax distinction in one of these digressions, but doesn't really develop it.
? That's the bulk of the argument. The CRA is designed to illustrate the fact that syntax, by itself, is insufficient for semantics. I think you have things reversed.
Really?
Yes, really. Even your beloved wikipedia agrees with me on that point. At this point, you're just in denial.
Even Stevan Harnad, who thinks the CRA is just obviously correct, knows that he's in the minority.
Citation needed. Or are you claiming to be a psychic that can read his mind? You sure as hell haven't read any of his work.
A computer can have more than syntax.
Which is, of course, silly nonsense, as I explained earlier. You introduce cameras, for example, but cameras don't magically give a computer sight. All the input from a camera consists of mere symbols -- indistinguishable from any other input. In the context of the Chinese Room, rather than the proposed video screen, the data flowing in from the camera would simply be more symbols. Symbols, obviously, no more infused with semantic content than any other set of symbols.
It's a rhetorical trick, though not a very good one as it's trivially easy to expose. It's why no one takes the robot reply seriously. It concedes the argument, and hopes no one will notice.
AGI does not imply consciousness, intentionality, qualia, or any of the other difficult issues in philosophy of mind - it's only about the ability to preform tasks.
Sense when? "AGI" has been synonymous with "strong AI" since the term's inception.
you seem to think that it was some kind of knock-down, indisputable argument.
Well, so far, no one has been able to offer a satisfactory reply. As an argument against computationalism, it's quite convincing. There's a reason it's still seriously discussed 35 years on. Had that pillar been knocked down, we'd all know the standard reply -- and the world would be a much more interesting place! The implications such a revelation would have for linguistics alone are staggering.
Computionalism is long dead. Searle, in no small part, is responsible for that.
and nowhere (there or in the literature) is Searle's conclusion treated as definitive.
It's not my fault that you're completely unfamiliar with the literature. I mention Stevan Harnad in another post, who's work I'll offer as just one of many counterexamples. (I mentioned him specifically as it's likely other Slashdotters are at least passingly familiar with him.)
If you don't, you'll end up hurt - badly
I had no idea that roving gangs of failing undergraduate philosophy majors were so dangerous!
The consciousness that is created by these computations is what provides the semantic meaning.
If you want to make magical claims, that's fine, though it's pointless to discuss the issue further.
It successfully proves that syntax is not all it takes to have consciousness, sure, you need more than syntax
That was the entire point. We're talking about computationalism, after all.
Give him that, and he will actually understand Chinese.
You need more than texts. Humans need more than texts. Machines need more than texts.
Obviously. Syntactic content alone is insufficient.
You, apparently, agree with me completely.
If you don't have specific reasons why it is impossible
I've made that clear already. Did you miss it?
And boolean logic is strictly insufficient for arithmetic
That's simply not true.
but it cannot connect those words to any kind of real-world phenomenal experience
You're trying to talk about the symbol grounding problem (see Harnad for a discussion about how this relates to Searle's Chinese Room).
But give the man in the room a bunch of photos and sound recordings and scratch-and-sniff panels and so on
This was addressed by Searle in the original 1980 paper. I recommend you read it. Put simply, it's a weak attempt to smuggle semantics in through the back door. Replace the photos, sounds, and smells with some equivalent in the form of symbols and you'll find that nothing changes.
As we have ever-improving image-recognition algorithms for computers. We have computers that can observe empirical phenomena and connect them to verbal symbols.
You're just still connecting symbols to other symbols. Consider this for a moment: I give you a unilingual dictionary and an exhaustive corpus of texts written in the same language. You have all you need to identify relationships between the symbols. The most you could hope to produce would be a grammar. You'd never be able to determine the meaning of any of the words.
There's a reason the argument has stood-up to 35 years worth of very harsh scrutiny. I recommend you read more about it. It will be painful for you, as it's difficult to see much-cherished beliefs crushed by something so simple. Just be glad it wasn't your career, as it was for so many others.
You're a bit off there. The crux of Searle's argument is that syntax, by itself, is insufficient for semantics. A point that is both obvious and seemingly irrefutable. If you believe that you can show otherwise, fame and fortune await.
Empty hand-waving, naturally, won't convince anyone.
There's no reason that [...]
There is a reason! You've simply chosen to ignore it: Syntax, by itself, is insufficient for semantics. Wishful thinking won't change that.
You can never state that something is not sufficient just because no one has figured it out yet.
That it's logically impossible is reason enough! Consider for a moment a simple example: I claim that it is impossible to clear 5 lines simultaneously in a game of Tetris. Would you say that claim is nonsense and it's only a matter of time before someone figures it out? Of course not. You can clearly demonstrate that it is an impossibility. The same is true for computational approaches to AGI -- they are logically impossible.
Why do you believe in such silly nonsense when it's clear that those beliefs are pure fantasy?
You are just a machine, get over it
If you'll take a moment to read the comment to which you replied, you'll find that I've said exactly that -- and that that point is completely irrelevant to the fact that computation is insufficient.
I'm sorry that reality does not conform to your fantasy. Please, at least do a little bit of reading. I don't even know how to begin to address your confusion.
What else other than computation is required?
I have no idea. Neither does anyone else. That doesn't change the fact that computation alone has been show to be insufficient. That's pretty well established.
No, I'm suggesting, on a rather well-established basis, that computation alone is insufficient. This is all assuming that the mind is a product of the brain. Whatever the brain does to cause consciousness, it can not be by mere computation alone.
I don't know why you find this so troubling.
It has nothing to do with atheism. You don't need to posit a god. Hell, you don't even need to posit a soul or other supernatural concept. Computation alone is insufficient. To claim otherwise necessitates that you be able to demonstrate that syntactic content is sufficient for semantic content. So simple and obvious is this point, that it can lead you to only one conclusion.
I don't worry about AGI for the same reason I don't worry about vampires.
Hubris? Sure. I'm of the opinion that when science conflicts with religion, that it should be religion, not science, that should adapt.
You're free to believe any nonsense you like. Just understand that it has no rational basis.
On this particular issue, they're both worthless. AGI by purely computational means simply isn't possible. We've known this for decades. Only lunatics like Kurzweil and the under-informed believe otherwise; a belief has no basis in reality.
You might as well discuss the sociological impact of a zombie apocalypse. It's just as meaningful.
It's like you weren't even paying attention!
I haven't eaten trans fat in decades and I haven't even been trying. What has transfat in it that you want?
Try finding fortune cookies without ... Looked everywhere and gave up. Selfishly I find myself cheering for the FDA ban because I want fortune cookies without trans fat.
You claim to have been able to avoid trans fat for decades without even trying. He wants to buy fortune cookies without trans fat. He claims that they don't exist. The ban means he will be able to buy fortune cookies without trans fat. Something he can not do currently.
That fortune cookies can be made without trans fat is completely irrelevant.
The smart ones only know java
Years ago, Slashdot used to disparage the kiddies who used Java.
Just assume that it's not worth your time. It's much easier that way.