So Transmeta are finally going to be ready to *say* something. The funny thing is that their patents are quite revealing about what they're up to - a speeded up version of the self-modifying FPGA technology that has occasionally spawned 'new era' claims. I'm not saying that their chip is just an FPGA, but that the effect is meant to be much the same: a metamicrocode that can be optimised in near-real time by a JIT-like (or is dynamic compiler a better term than JIT?) compiler and scheduler.
Please though, don't beleive all the speed hype. Remember, it was a year ago or so when 1GHz sounded astonishing, but now it's almost boring for those chiller guys. The thing is going to be *flexible* not *necessarily* fast.
Curiosity killed the cat, but who ever saw a cat reading a patent application?
Just look at the site - the whole thing is an obvious troll. I can see that much even having been up since 07:00 GMT Wednesday (now 09:10 GMT Thursday). Rob must really be pushing the edge these days - he doesn't usually fall for this kind of thing.
Sorry to seem harsh, but the first level filter seems to have failed bigtime.
Hi, I would have emailed this but you don't give an address, so I hope you're the kind of person who looks for replies to their comments;-)
Good question, why recommend a book on three such aparently unrelated topics? Well, Kurt Godel is mainly famous for his "incompleteness theorem", in which he showed c. 1931 that there will always be valid statements of a formal system that cannot be proved wither true or false within that system. For some time "Fermats Last Theorem" was thought to be of this type. The relevance of this is that it can be shown to be equivalet to the "Halting Problem" for universal turing machines.
A Turing Machine is a very generally specified computer with infinite memory and infinite time to carry out its calculations. a Universal Turing Machine is a TM that emulates all other TMs (where a TM is the machine + a program + data). So the haltng problem is the question, "is there a TM that, when fed the specs of another TM, says whether the TM it has been given will produce an answer?" To which Turing showed that the answer is "No".
We can see that the question about TMs (the halting problem) is equivalent to the question about mathematical statements (the "decidability question") that Godel posed and solved.
Escher comes into this by virtue of the fact that his art poses deep questions about the relationships between different levels of understanding. What exactly is the foreground and what is the background of a picture like "Swans"?
Bach comes into it because the fundamentally mthematical nature of much of his music is fascinating for the author, and provides the source for some very complex and original analogies in the text - try the chapter "Crab Canon" in which the structure of a Bach Canon is translated into a dialogue (with one interjection).
Now really you should have put the disclaimer as the title. If you're religious then the original question is meaningless, and if you're not, then the phrase 'turing machine' would be a lot more helpful than "lossless accurate finite state machine/automaton".
Can a turing machine emulate brain states? Most everyone in the field thinks so except Roger Penrose ("the emperor's new mind"), who has a bee in his bonnet about wanting to recover free will by invoking quantum gravity (LOL IMO, but then writing off the first guy to tile the plane without repetitions is perhaps getting above station).
So next time you start a post with "I dont know dick but...", how about stopping just there?
Oh yeah, I had a point: You make many comments stating the noncomputability of brain functions ("you can't represent X as bytes"), and then go and concur with the idea that the question has an answer.
suggested reading: Godel, escher, Bach, an Eternal Golden Braid, by Douglas Hofstadter
First off, although I don't do this stuff day to day anymore, I did do my degree in cognitive science, so this is the kind of stuff we were expected to think about.
The question as asked is ambiguous, since a 'hard disk' is an object whose contents have no semantics, just syntax, that is they have no meaning except when interpreted by another entity (you, me, sendmail, etc). OTOH the information that is stored in the brain is a mixture of semantics and syntax that we do not yet understand.
To clarify this I would like to give a very/. analogy. Consider a minimal working Linux installation (basic hardware, kernel, shell). Let us say that the system has a 10GB hard disk. It is obvious that most of the HD is almost completely devoid of information, since it has no context, but is just blank. However the portion of the HD that contains the base software is very information rich, but only in the conext of the surrounding hardware (bios etc), and perhaps more importantly, in the context of the wider environment of/.ers and others who know what Linux is and what it can do, and can interact with it.
Now here we have drawn three levels fairly clearly - the HD, the bios and other hard/firmware, and the rest of the world. But in the case of a human brain it is not at all clear where (or if) one can draw these distinctions, so only a complete description can suffice (i.e. we are not able to summarise the state by means of external references)
In this event we can rephrase the question as follows:
Given the required processing capacity, what amount of storage would be necessary to provide the same information processing capacity as a human brain?
Now here we hit astronomical numbers. The question is equivalent (check Turing, Church, etc) to asking how many bits it would require to store a complete description of a human brain at a given instant. This is certainly a smaller number than a precise description of the state of every subatomic particle in a brain (i.e. less than the memory required for a Star Trek transporter), but is still pretty big.
Back of an envelope? A very conservative envelope? If a brain state could be described by the states of each neuron and each connection in the brain, and each of those took 16 bits (which is almost certainly a gross underestimate), and there are ~10^12 neurons and ~10^3 connections each on average (Churchland and Sejnovki, 1992), then that is 16*10^15.
Slashdot Mirror
So Transmeta are finally going to be ready to *say* something. The funny thing is that their patents are quite revealing about what they're up to - a speeded up version of the self-modifying FPGA technology that has occasionally spawned 'new era' claims. I'm not saying that their chip is just an FPGA, but that the effect is meant to be much the same: a metamicrocode that can be optimised in near-real time by a JIT-like (or is dynamic compiler a better term than JIT?) compiler and scheduler.
Please though, don't beleive all the speed hype. Remember, it was a year ago or so when 1GHz sounded astonishing, but now it's almost boring for those chiller guys. The thing is going to be *flexible* not *necessarily* fast.
Curiosity killed the cat, but who ever saw a cat reading a patent application?
.sig thingy
Just look at the site - the whole thing is an obvious troll. I can see that much even having been up since 07:00 GMT Wednesday (now 09:10 GMT Thursday). Rob must really be pushing the edge these days - he doesn't usually fall for this kind of thing.
Sorry to seem harsh, but the first level filter seems to have failed bigtime.
.sigish thing
Hi, I would have emailed this but you don't give an address, so I hope you're the kind of person who looks for replies to their comments ;-)
Good question, why recommend a book on three such aparently unrelated topics? Well, Kurt Godel is mainly famous for his "incompleteness theorem", in which he showed c. 1931 that there will always be valid statements of a formal system that cannot be proved wither true or false within that system. For some time "Fermats Last Theorem" was thought to be of this type. The relevance of this is that it can be shown to be equivalet to the "Halting Problem" for universal turing machines.
A Turing Machine is a very generally specified computer with infinite memory and infinite time to carry out its calculations. a Universal Turing Machine is a TM that emulates all other TMs (where a TM is the machine + a program + data). So the haltng problem is the question, "is there a TM that, when fed the specs of another TM, says whether the TM it has been given will produce an answer?" To which Turing showed that the answer is "No".
We can see that the question about TMs (the halting problem) is equivalent to the question about mathematical statements (the "decidability question") that Godel posed and solved.
Escher comes into this by virtue of the fact that his art poses deep questions about the relationships between different levels of understanding. What exactly is the foreground and what is the background of a picture like "Swans"?
Bach comes into it because the fundamentally mthematical nature of much of his music is fascinating for the author, and provides the source for some very complex and original analogies in the text - try the chapter "Crab Canon" in which the structure of a Bach Canon is translated into a dialogue (with one interjection).
Try it and see.
M
Now really you should have put the disclaimer as the title. If you're religious then the original question is meaningless, and if you're not, then the phrase 'turing machine' would be a lot more helpful than "lossless accurate finite state machine/automaton".
Can a turing machine emulate brain states? Most everyone in the field thinks so except Roger Penrose ("the emperor's new mind"), who has a bee in his bonnet about wanting to recover free will by invoking quantum gravity (LOL IMO, but then writing off the first guy to tile the plane without repetitions is perhaps getting above station).
So next time you start a post with "I dont know dick but...", how about stopping just there?
Oh yeah, I had a point: You make many comments stating the noncomputability of brain functions ("you can't represent X as bytes"), and then go and concur with the idea that the question has an answer.
suggested reading:
Godel, escher, Bach, an Eternal Golden Braid, by Douglas Hofstadter
First off, although I don't do this stuff day to day anymore, I did do my degree in cognitive science, so this is the kind of stuff we were expected to think about.
/. analogy. Consider a minimal working Linux installation (basic hardware, kernel, shell). Let us say that the system has a 10GB hard disk. It is obvious that most of the HD is almost completely devoid of information, since it has no context, but is just blank. However the portion of the HD that contains the base software is very information rich, but only in the conext of the surrounding hardware (bios etc), and perhaps more importantly, in the context of the wider environment of /.ers and others who know what Linux is and what it can do, and can interact with it.
The question as asked is ambiguous, since a 'hard disk' is an object whose contents have no semantics, just syntax, that is they have no meaning except when interpreted by another entity (you, me, sendmail, etc). OTOH the information that is stored in the brain is a mixture of semantics and syntax that we do not yet understand.
To clarify this I would like to give a very
Now here we have drawn three levels fairly clearly - the HD, the bios and other hard/firmware, and the rest of the world. But in the case of a human brain it is not at all clear where (or if) one can draw these distinctions, so only a complete description can suffice (i.e. we are not able to summarise the state by means of external references)
In this event we can rephrase the question as follows:
Given the required processing capacity, what amount of storage would be necessary to provide the same information processing capacity as a human brain?
Now here we hit astronomical numbers. The question is equivalent (check Turing, Church, etc) to asking how many bits it would require to store a complete description of a human brain at a given instant. This is certainly a smaller number than a precise description of the state of every subatomic particle in a brain (i.e. less than the memory required for a Star Trek transporter), but is still pretty big.
Back of an envelope? A very conservative envelope? If a brain state could be described by the states of each neuron and each connection in the brain, and each of those took 16 bits (which is almost certainly a gross underestimate), and there are ~10^12 neurons and ~10^3 connections each on average (Churchland and Sejnovki, 1992), then that is 16*10^15.
or 16 peta bits
HTH
Matt
I don't know where you get that idea. There is no proof one way or the other as to what the result of perfect play might be