His only criterion is whether the browser crashes or not. Somehow, it disturbs me more that IE doesn't crash; what precisely is the effect of the bad code then?
I suspect it detects the error and does what any web browser ought to do on detecting malformed markup: ignore it.
Yes. This is because it crashes IE. Whether this makes it a trojan or not is an interesting question, but the fact that it would was very obvious from the context...
Yep. Another example, that I've now seen several sites falling over with, is that MS's Javascript implementation will accept statements with no terminating ';' in many situations that no other implementation will.
Actually, MS have never used Visual Studio (whatever version) to compile any part of windows.
Visual Studio just isn't up to a project of that size.
Really? Many of the components that were included in the source code leak were of sizes similar to projects I've worked on with VS '97. Although, IIRC, there weren't any.dsp or.vcproj (or whatever) files in there, so maybe they don't use it anyway. Just saying, I don't think the size is the reason.
Not that I downloaded the source or anything. Wouldn't involve myself in such a blatant violation of MS's copyrights. Just word of mouth, kind of thing, you know?
Listen to me. I've written a (rudimentary) web browser. HTML is not executed or interpreted. It is parsed into data structures that are then processed to determine a screen area that is relevant for each one, and a description of their content is then drawn into that area.
Besides some vague mumblings about "potential" exploits, I didn't see any evidence in the article of deleterious effects caused by those browsers "crashing". And I didn't see any evidence that IE survived the same broken HTML without side effects. Anyhow, the evil you know about is better than the one you ignore. It's better to have the browser fail than to have it keep running with a possible exploit.
Listen -- if the browser behaves in a way that it isn't supposed to (i.e. crashing) when given invalid input, this is because it has done something wrong. By far the most likely symptom of doing something wrong in a program like this is crashing. Probably by either following a NULL pointer, a pointer to a data structure that has been freed, or attempting to free the same pointer twice. The latter two of these errors can almost always be exploited to execute malicious code. It is often difficult, but the merest possibility ought to concern you.
However, it is _highly_ unlikely that a program with a structure like the one I've described above (particularly one that is written in a compiled-to-native-code language without garbage collection, which I'm pretty sure IE is) could become subverted in some way by bad data, but continue running with the incorrect behaviour waiting to be run at some later point after the page in question is no longer being displayed. For this to happen, some data would have to be copied out of the internal representation of the HTML document (which will be deleted after the page is closed) and into IE's internal state. There is no reason for it to do this; in fact, there will be very little state held over between pages (history, user settings, and pointers to window related structures and resources like menus, toolbars, etc. probably account for almost everything), and anything copied from a document ought to be treated as 'suspicious' anyway.
The Mozilla team will either add stuff to quirks mode, or pass the site reference to evangelists.
Much more frequently the latter than the former. And guess what... it just _doesn't help_. I've tried to impress on several companies myself the reason why not working with Mozilla is a bad thing. Their response. "Sorry, we've optimized the site for Internet Explorer," or some shite like that. They just don't want to listen.
IE might not be fully standards-compliant, but it is a "modern" engine that supports DOM/CSS and probably has very little Mosaic code left in it.
If there is any, though, it's likely to be the lexical analyser, as that's the one area that hasn't really changed much at all with all the new features.
Interesting. I was under the impression that IBM initially wanted CP/M for the PC, but were put of dealing with Digital Research because they found the company difficult to deal with (there are various stories; one is that Kildall failed to turn up to a crucial meeting). If they were so fanatical about dealing with the "elite", surely they'd have gone to Gates first?
It seems like you're using the same definition of a non-deterministic system as I am, after all, because what I've been talking about all along is a system which depends on variables that are "below the threshhold of measurability", as you put it. Chaotic systems, where even a tiny, undetectable change in the inputs will change the output. There are no shortage of such systems available for input to a computer. Thermal noise generators are just one method used; these are believed to be completely unpredictable, if suitable biased.
Note that due to space-time granularity, there are no real world systems that have an infinite range of possible outcomes; only theoretical systems have this property. Of course, some real world systems approximate this.
But why is this even relevant? Why must an AI be non-deterministic?
True. Or, more relevantly to this discussion, MainSoft, who developed the Unix port of IE, and would almost certainly do the same for Office if MS ever chose to do it.
Again, you miss the point about what determines whether a system is deterministic or not. It has nothing to do with predictability.
Perhaps you missed the definition I quoted of 'deterministic' - 'a system whose time evolution can be predicted exactly.' It has everything to do with predictability; that's what the word _means_.
You didn't predict how it was going to happen the first time (or you would have written it differently). the program is deterministic, just not in the way you wanted it to be. Otherwise, you wold not be able to reproduce the bug.
No, however with the additional information I had from the bug report I was able to predict (with a fair degree of certainty) that taking those steps would result in the bug being exhibited, otherwise I wouldn't be using them to attempt to reproduce it.
However, I still don't understand what relevance determinism has to the matter of AI consciousness.
Yeah, sure, there's a finite number of possible states for each measurement. There are also only a finite number of possible outcomes of any action, whether it is deterministic or not.
The point is, though, that if I use a system which is closed and cannot be influenced in a predictable fashion by any external trigger, then for all practical intents and purposes that _is_ nondeterminstic. It is impossible (if the system has been designed correctly) to gain enough information about it and the world around it to predict specifically how it will react.
Do you believe that tossing a coin is deterministic? Rolling a die? In trying to predict the outcome of such situations, you will find it is impossible to make the measurements that you need to do it with enough accuracy. It is, literally, impossible to determine in advance what the outcome will be, hence non-deterministic.
This is the kind of system I am talking about.
As I tried to point out with the 4 coins example, not being able to predict the outcome has nothing to do with whether a system is deterministic or not.
Strange, I think it has everything to do with it:
deterministic
1. (probability) Describes a system whose time evolution can be predicted exactly.
(source: dictionary.com)
However, in your 4 coins example, there is somebody who knows the initial state and can therefore predict the outcome. In my example, there is nobody who knows the initial state, therefore the outcome cannot be predicted, therefore it is non-deterministic.
Yeah, I get that. It was more your idea of an experiment that tested whether an observer was required in the double slit experiment having been performed. I've never heard of it.
Am I the only one who thought that with 3 DVDs you could store most of the BBS systems and let readers find out what it all was for themselves?
And, of course, with 3 DVDs you could store well over 10 hours of video at a quality level that matches that used by most commercial DVD producers... so why not put this documentary on 2 discs?
MPEG2 is the format that DVB uses for transmitting its videos. I don't see the relevance of DivX here; this device is obviously intended for use as a digital video recorder and MPEG2 is the obvious format for it to use for this.
It is impossible to create a non-deterministic computer. Your example is off, in that you don't seem to understand what I mean by "deterministic". The same inputs will always result in the same outputs. In other words, if your sensor is provided the same input over and over again, and the system is in the same initial state, it will always generate the same output.
Yes, sure. But because it is impossible to arrange for the sensor in the system I describe to provide the same input over and over again, this is a purely hypothetical and rather irrelevant point. For all practical purposes, such an apparatus is non-deterministic.
Also, you still haven't defined why an AI would need to be non-deterministic to have consciousness. I am far from certain that humans behave non-deterministically. Chaotically, perhaps.
The two-slit experiments have been done in such a way as to test whether an observer is required.
Do you have a reference to that? I haven't seen any discussion of it before.
Here's an interesting theory that you might be interested in -- one of the methods for observation to take place without requiring consciousness. Ironically, it was proposed by the same Roger Penrose who believes the same thing you do, but for different reasons...
I see what you mean about the entropy of the data... in the end though, I think what's wrong here is that you're applying a subjective interpretation of entropy where an objective one is required. This page has an interesting discussion on what entropy is and isn't.
You use the word "simulate" in your first 2 points. Please remember that a simulation is NOT the real thing. You're falling into the same errors Kurzweil made.
In computational theory, if one model can simulate another they are said to be equivalent. I don't see any reason why a simulation of the process of intelligence should be any different from the real thing. If you can't tell the difference, what is the difference?
Any computer-based AI will never achieve consciousness, because all such AIs are still, by definition, deterministic systems - "I think because I'm programmed, at some level, to think."
It is more than possible to create non-deterministic computer systems. The process usually involves the use of a sensor that makes measurements so small that their outcome is impossible to predict in advance, and which are chaotic in nature so that it is impossible for them to be influenced consciously.
However, I see no evidence that consciousness requires non-determinism; can you supply me with any?
It's a major error when he goes on to use this mistaken premise to argue for nanotech being the route to improving brain performance.
All we're talking about is a minor error of scale. He's probably only out by a few orders of magnitude in terms of the brain's capabilities for learning. This doesn't change his fundamental arguments.
As for whether the consciousness inhabits the quantum world, consider that the existence of a conscious observer affects the outcome of experiments that depend on quantum effects (the "two-slit" experiments are the best starting point). Just as the existence of an observer affects, in some situations, whether the exact same state should be considered as an increase or decrease in entropy.
I believe that it's generally held these days that the presence of a conscience is not necessary to constitue an 'observation' under quantum physics. What precisely does constitute an observation is a matter that is, I believe, presently the focus of much debate, but there are many theories which do not require the presence of an intelligent observer.
I'm not quite sure I understand your point about entropy, though. I've never studied thermodynamics in great depth, and am unaware of the situation you mention.
And I have never in my life referred to anything by Penrose, so why bring it up?
I bring it up because AFAIK he was the first person to suggest that consciousness is a quantum-level effect, and has offered the only argument as to why this must be that I had previously heard (an argument which I dismiss as grounded in absurdity). I'll admit your reasoning is new to me, and while I'm not convinced by it, I'll grant that it makes more sense than Penrose's.
Slashdot Mirror
His only criterion is whether the browser crashes or not. Somehow, it disturbs me more that IE doesn't crash; what precisely is the effect of the bad code then?
I suspect it detects the error and does what any web browser ought to do on detecting malformed markup: ignore it.
Also fixed in Win2K + IE6 SP1, by the looks of things.
Yes. This is because it crashes IE. Whether this makes it a trojan or not is an interesting question, but the fact that it would was very obvious from the context...
MSIE was embracing and extending your new syntax.
Yep. Another example, that I've now seen several sites falling over with, is that MS's Javascript implementation will accept statements with no terminating ';' in many situations that no other implementation will.
Microsoft's is called JScript, not Javascript.
So that'll be why it calls it when I use a url of the form "javascript:[code]", or a tag like "<script language=javascript>" then?
MS's software all internally understands that this language is called Javascript. It's only MS marketing that decided to use a different name for it.
Actually, MS have never used Visual Studio (whatever version) to compile any part of windows.
.dsp or .vcproj (or whatever) files in there, so maybe they don't use it anyway. Just saying, I don't think the size is the reason.
Visual Studio just isn't up to a project of that size.
Really? Many of the components that were included in the source code leak were of sizes similar to projects I've worked on with VS '97. Although, IIRC, there weren't any
Not that I downloaded the source or anything. Wouldn't involve myself in such a blatant violation of MS's copyrights. Just word of mouth, kind of thing, you know?
Err, you may want to switch to extrans when posting HTML code. :)
Listen to me. I've written a (rudimentary) web browser. HTML is not executed or interpreted. It is parsed into data structures that are then processed to determine a screen area that is relevant for each one, and a description of their content is then drawn into that area.
Besides some vague mumblings about "potential" exploits, I didn't see any evidence in the article of deleterious effects caused by those browsers "crashing". And I didn't see any evidence that IE survived the same broken HTML without side effects. Anyhow, the evil you know about is better than the one you ignore. It's better to have the browser fail than to have it keep running with a possible exploit.
Listen -- if the browser behaves in a way that it isn't supposed to (i.e. crashing) when given invalid input, this is because it has done something wrong. By far the most likely symptom of doing something wrong in a program like this is crashing. Probably by either following a NULL pointer, a pointer to a data structure that has been freed, or attempting to free the same pointer twice. The latter two of these errors can almost always be exploited to execute malicious code. It is often difficult, but the merest possibility ought to concern you.
However, it is _highly_ unlikely that a program with a structure like the one I've described above (particularly one that is written in a compiled-to-native-code language without garbage collection, which I'm pretty sure IE is) could become subverted in some way by bad data, but continue running with the incorrect behaviour waiting to be run at some later point after the page in question is no longer being displayed. For this to happen, some data would have to be copied out of the internal representation of the HTML document (which will be deleted after the page is closed) and into IE's internal state. There is no reason for it to do this; in fact, there will be very little state held over between pages (history, user settings, and pointers to window related structures and resources like menus, toolbars, etc. probably account for almost everything), and anything copied from a document ought to be treated as 'suspicious' anyway.
Sorry, you've just got it plain wrong.
The Mozilla team will either add stuff to quirks mode, or pass the site reference to evangelists.
Much more frequently the latter than the former. And guess what... it just _doesn't help_. I've tried to impress on several companies myself the reason why not working with Mozilla is a bad thing. Their response. "Sorry, we've optimized the site for Internet Explorer," or some shite like that. They just don't want to listen.
IE might not be fully standards-compliant, but it is a "modern" engine that supports DOM/CSS and probably has very little Mosaic code left in it.
If there is any, though, it's likely to be the lexical analyser, as that's the one area that hasn't really changed much at all with all the new features.
Interesting. I was under the impression that IBM initially wanted CP/M for the PC, but were put of dealing with Digital Research because they found the company difficult to deal with (there are various stories; one is that Kildall failed to turn up to a crucial meeting). If they were so fanatical about dealing with the "elite", surely they'd have gone to Gates first?
It seems like you're using the same definition of a non-deterministic system as I am, after all, because what I've been talking about all along is a system which depends on variables that are "below the threshhold of measurability", as you put it. Chaotic systems, where even a tiny, undetectable change in the inputs will change the output. There are no shortage of such systems available for input to a computer. Thermal noise generators are just one method used; these are believed to be completely unpredictable, if suitable biased.
Note that due to space-time granularity, there are no real world systems that have an infinite range of possible outcomes; only theoretical systems have this property. Of course, some real world systems approximate this.
But why is this even relevant? Why must an AI be non-deterministic?
True. Or, more relevantly to this discussion, MainSoft, who developed the Unix port of IE, and would almost certainly do the same for Office if MS ever chose to do it.
Again, you miss the point about what determines whether a system is deterministic or not. It has nothing to do with predictability.
Perhaps you missed the definition I quoted of 'deterministic' - 'a system whose time evolution can be predicted exactly.' It has everything to do with predictability; that's what the word _means_.
You didn't predict how it was going to happen the first time (or you would have written it differently). the program is deterministic, just not in the way you wanted it to be. Otherwise, you wold not be able to reproduce the bug.
No, however with the additional information I had from the bug report I was able to predict (with a fair degree of certainty) that taking those steps would result in the bug being exhibited, otherwise I wouldn't be using them to attempt to reproduce it.
However, I still don't understand what relevance determinism has to the matter of AI consciousness.
Yeah, who'd have thought it -- the joke question got the most informative answer of them all, while the most serious one was just dismissed. :)
:)
(I'm reading the ACME paper now. Looks interesting.)
Jules, who writes his own editors too.
Yeah, sure, there's a finite number of possible states for each measurement. There are also only a finite number of possible outcomes of any action, whether it is deterministic or not.
The point is, though, that if I use a system which is closed and cannot be influenced in a predictable fashion by any external trigger, then for all practical intents and purposes that _is_ nondeterminstic. It is impossible (if the system has been designed correctly) to gain enough information about it and the world around it to predict specifically how it will react.
Do you believe that tossing a coin is deterministic? Rolling a die? In trying to predict the outcome of such situations, you will find it is impossible to make the measurements that you need to do it with enough accuracy. It is, literally, impossible to determine in advance what the outcome will be, hence non-deterministic.
This is the kind of system I am talking about.
As I tried to point out with the 4 coins example, not being able to predict the outcome has nothing to do with whether a system is deterministic or not.
Strange, I think it has everything to do with it:
deterministic
1. (probability) Describes a system whose time evolution can be predicted exactly.
(source: dictionary.com)
However, in your 4 coins example, there is somebody who knows the initial state and can therefore predict the outcome. In my example, there is nobody who knows the initial state, therefore the outcome cannot be predicted, therefore it is non-deterministic.
Yeah, I get that. It was more your idea of an experiment that tested whether an observer was required in the double slit experiment having been performed. I've never heard of it.
Am I the only one who thought that with 3 DVDs you could store most of the BBS systems and let readers find out what it all was for themselves?
And, of course, with 3 DVDs you could store well over 10 hours of video at a quality level that matches that used by most commercial DVD producers... so why not put this documentary on 2 discs?
Because USB is good enough for their application, and is compatible with a wider range of existing hardware, I suspect.
MPEG2 is the format that DVB uses for transmitting its videos. I don't see the relevance of DivX here; this device is obviously intended for use as a digital video recorder and MPEG2 is the obvious format for it to use for this.
Why obviously? Samba is _not_ the only implemention of SMB, a protocol for which the specs are publically available.
It is impossible to create a non-deterministic computer. Your example is off, in that you don't seem to understand what I mean by "deterministic". The same inputs will always result in the same outputs. In other words, if your sensor is provided the same input over and over again, and the system is in the same initial state, it will always generate the same output.
Yes, sure. But because it is impossible to arrange for the sensor in the system I describe to provide the same input over and over again, this is a purely hypothetical and rather irrelevant point. For all practical purposes, such an apparatus is non-deterministic.
Also, you still haven't defined why an AI would need to be non-deterministic to have consciousness. I am far from certain that humans behave non-deterministically. Chaotically, perhaps.
The two-slit experiments have been done in such a way as to test whether an observer is required.
Do you have a reference to that? I haven't seen any discussion of it before.
Here's an interesting theory that you might be interested in -- one of the methods for observation to take place without requiring consciousness. Ironically, it was proposed by the same Roger Penrose who believes the same thing you do, but for different reasons...
I see what you mean about the entropy of the data... in the end though, I think what's wrong here is that you're applying a subjective interpretation of entropy where an objective one is required. This page has an interesting discussion on what entropy is and isn't.
You use the word "simulate" in your first 2 points. Please remember that a simulation is NOT the real thing. You're falling into the same errors Kurzweil made.
In computational theory, if one model can simulate another they are said to be equivalent. I don't see any reason why a simulation of the process of intelligence should be any different from the real thing. If you can't tell the difference, what is the difference?
Any computer-based AI will never achieve consciousness, because all such AIs are still, by definition, deterministic systems - "I think because I'm programmed, at some level, to think."
It is more than possible to create non-deterministic computer systems. The process usually involves the use of a sensor that makes measurements so small that their outcome is impossible to predict in advance, and which are chaotic in nature so that it is impossible for them to be influenced consciously.
However, I see no evidence that consciousness requires non-determinism; can you supply me with any?
It's a major error when he goes on to use this mistaken premise to argue for nanotech being the route to improving brain performance.
All we're talking about is a minor error of scale. He's probably only out by a few orders of magnitude in terms of the brain's capabilities for learning. This doesn't change his fundamental arguments.
As for whether the consciousness inhabits the quantum world, consider that the existence of a conscious observer affects the outcome of experiments that depend on quantum effects (the "two-slit" experiments are the best starting point). Just as the existence of an observer affects, in some situations, whether the exact same state should be considered as an increase or decrease in entropy.
I believe that it's generally held these days that the presence of a conscience is not necessary to constitue an 'observation' under quantum physics. What precisely does constitute an observation is a matter that is, I believe, presently the focus of much debate, but there are many theories which do not require the presence of an intelligent observer.
I'm not quite sure I understand your point about entropy, though. I've never studied thermodynamics in great depth, and am unaware of the situation you mention.
And I have never in my life referred to anything by Penrose, so why bring it up?
I bring it up because AFAIK he was the first person to suggest that consciousness is a quantum-level effect, and has offered the only argument as to why this must be that I had previously heard (an argument which I dismiss as grounded in absurdity). I'll admit your reasoning is new to me, and while I'm not convinced by it, I'll grant that it makes more sense than Penrose's.