Robert Rosen
Locked
12
12
 
|
Glen wrote:
> And if you tell it that > there are only, say, 10 possible answers, it will _merely_ produce one > of those prescribed 10 possible answers. > You could say that about an employee, too, but that doesn't give much insight into what that person might actually be able to do. > (I live for the day when I ask > a computer: "Is this true or false?" And it answers: "Neither, it's > _blue_!" ;-) Computers typically don't do that, except in paraphrasing/concept extraction expert systems (e.g. Cyc), because people don't typically want them to do that. For example, it's clear when this Java program is compiled that the compiler knows what `color' really is. enum Color { Blue, Red } public class Test { static void main (String args[] ) { Color color = Color.Blue; if (color == true) { System.out.println ("true!"); } } } One easy way to let that go is to switch to a dynamically typed language, where logical inconsistencies are dealt with in a case by case basis by the programmer. (Presumably until the programmer can `see' how things should fit together.) As far as detecting (supposedly) ill-posed questions goes, if you are willing to put aside the complex matter of natural language processing, it seems to me it's a matter of similarity search against a set propositions, and then engaging in a dialog of generalization and precisification with the user to identify an unambiguous and agreeable form for the question that has appropriate answers. Marcus |
|
|
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1 Marcus G. Daniels on 01/08/2008 08:49 AM: > As far as detecting (supposedly) ill-posed questions goes, if you are > willing to put aside the complex matter of natural language processing, > it seems to me it's a matter of similarity search against a set > propositions, and then engaging in a dialog of generalization and > precisification with the user to identify an unambiguous and agreeable > form for the question that has appropriate answers. But the issue isn't about handling ill-posed questions on a case-by-case basis. In fact, the hypothesis is that ill- versus well- posed questions is an unrealistic dichotomy. It's just another form of the "excluded middle". A primary point made by RR is that living systems can handle ambiguity where "machines" cannot. Of course, it's true that if a programmer pre-scribed a method for detecting and handling some particular ambiguity, then the machine will _seem_ like it handles that ambiguity. But, programmers haven't yet found a way to handle all ambiguity a computer program may or may not come across in the far-flung future. That's in contrast to a living system, which we _presume_ can handle any ambiguity presented to it (or, in a softer sense, many many more ambiguities than a computer program can handle). - -- glen e. p. ropella, 971-219-3846, http://tempusdictum.com Almost nobody dances sober, unless they happen to be insane. -- H. P. Lovecraft -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFHg7G4ZeB+vOTnLkoRAjTtAKCu0nimkhWcQdIYDn8Uy05N6jwaUACfUzUc g6rWx3ZPlmAaayG7qqJHJ1g= =kWTj -----END PGP SIGNATURE----- |
|
|
Glen E. P. Ropella wrote:
> But, programmers haven't yet > found a way to handle all ambiguity a computer program may or may not > come across in the far-flung future. That's in contrast to a living > system, which we _presume_ can handle any ambiguity presented to it (or, > in a softer sense, many many more ambiguities than a computer program > can handle). > Perception, locomotion, and signaling are capabilities that animals have evolved for millions of years. It's not fair to compare a learning algorithm to the learning capabilities of a living system without factoring in the fact that robots aren't disposable for the sake of realizing evolutionary selection and search. And even if they weren't, do you want drive over robots on the highway to make it so? Anything that requires significant short term memory and integration of broad but scare evidence is probably something a computer will be better at than a human. It may be that a `programmer' implements a self-organized neural net, or an kernel eigensystem solver but that only concerns the large classes of signals that can be extracted. It's not like some giant if/then statement for all possible cases that a programmer would keep tweaking. My assertion remains that the things computers do are primarily limited by the desire of humans to 1) understand what was learned, and then 2) use it. If those two conditions are removed, then we are talking about a very different scenario. There's little incentive to develop control systems for robots to keep them stumbling around as long as possible, with no limits on the actions they can take. Marcus |
|
|
In reply to this post by glen ep ropella
I thought the implication was that the organization of life is an
inherently ill-posed question from an observer's perspective. To me that either means you accept 'bad answers' or 'better and better answers', and the difference is methodological. Phil Henshaw ????.?? ? `?.???? ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 680 Ft. Washington Ave NY NY 10040 tel: 212-795-4844 e-mail: pfh at synapse9.com explorations: www.synapse9.com -- "it's not finding what people say interesting, but finding what's interesting in what they say" -- > -----Original Message----- > From: friam-bounces at redfish.com > [mailto:friam-bounces at redfish.com] On Behalf Of Glen E. P. Ropella > Sent: Tuesday, January 08, 2008 12:24 PM > To: The Friday Morning Applied Complexity Coffee Group > Subject: Re: [FRIAM] Robert Rosen > > > -----BEGIN PGP SIGNED MESSAGE----- > Hash: SHA1 > > Marcus G. Daniels on 01/08/2008 08:49 AM: > > As far as detecting (supposedly) ill-posed questions goes, > if you are > > willing to put aside the complex matter of natural language > processing, > > it seems to me it's a matter of similarity search against a set > > propositions, and then engaging in a dialog of generalization and > > precisification with the user to identify an unambiguous > and agreeable > > form for the question that has appropriate answers. > > But the issue isn't about handling ill-posed questions on a > case-by-case basis. In fact, the hypothesis is that ill- > versus well- posed questions is an unrealistic dichotomy. > It's just another form of the "excluded middle". > > A primary point made by RR is that living systems can handle > ambiguity where "machines" cannot. > > Of course, it's true that if a programmer pre-scribed a > method for detecting and handling some particular ambiguity, > then the machine will _seem_ like it handles that ambiguity. > But, programmers haven't yet found a way to handle all > ambiguity a computer program may or may not come across in > the far-flung future. That's in contrast to a living system, > which we _presume_ can handle any ambiguity presented to it > (or, in a softer sense, many many more ambiguities than a > computer program can handle). > > - -- > glen e. p. ropella, 971-219-3846, http://tempusdictum.com > Almost nobody dances sober, unless they happen to be insane. > -- H. P. Lovecraft > > -----BEGIN PGP SIGNATURE----- > Version: GnuPG v1.4.6 (GNU/Linux) > Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org > > iD8DBQFHg7G4ZeB+vOTnLkoRAjTtAKCu0nimkhWcQdIYDn8Uy05N6jwaUACfUzUc > g6rWx3ZPlmAaayG7qqJHJ1g= > =kWTj > -----END PGP SIGNATURE----- > > ============================================================ > FRIAM Applied Complexity Group listserv > Meets Fridays 9a-11:30 at cafe at St. John's College > lectures, archives, unsubscribe, maps at http://www.friam.org > > |
|
|
In reply to this post by Marcus G. Daniels
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1 Marcus G. Daniels on 01/08/2008 10:44 AM: > Perception, locomotion, and signaling are capabilities that animals > have evolved for millions of years. It's not fair to compare a > learning algorithm to the learning capabilities of a living system > without factoring in the fact that robots aren't disposable for the > sake of realizing evolutionary selection and search. Well, first off, you're not arguing with me. [grin] I'm doing my best to explain what I understand of RR's ideas. That's all. Feel free to go over to one of the rosen mailing lists to see how they mostly dislike me because I disagree with them (... or in their opinion because I'm a stubborn jerk who doesn't listen to them or because I'm just too lazy to dig really deep into RR's ideas ... or whatever ;-). > Anything that requires significant short > term memory and integration of broad but scare evidence is probably > something a computer will be better at than a human. That's just plain silly in terms of RR's ideas because _humans_ program the computer. Until/unless we come up with a computer that programs itself, or a computer that programs another computer, or something of that sort, computers will _never_ be better at any task than humans. I.e. in RR terms, humans are THE canalizing "efficient cause" for any computer system. > My assertion remains that the things computers do are primarily > limited by the desire of humans to 1) understand what was learned, > and then 2) use it. If those two conditions are removed, then we > are talking about a very different scenario. There's little > incentive to develop control systems for robots to keep them > stumbling around as long as possible, with no limits on the actions > they can take. Computers don't _do_ anything. Humans _do_ things using computers as tools. I believe that will change. But for the time being, it's the case. Anything else is speculation. RR speculates (albeit with significant rhetoric and build-up) that computers, as we now know them, are incapable of doing everything living systems do. I would speculate that computers, as we now know them (but with manymanymanymany more of them and radically different software), can do everything living systems do. But, just because my opinion differs from RR's doesn't mean both opinions are anything more than speculation. - -- glen e. p. ropella, 971-219-3846, http://tempusdictum.com The Government of the United States is in no sense founded on the Christian religion. -- John Adams -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFHg8xOZeB+vOTnLkoRAuAEAKCDue88MCLn77MZv/riWMkqE6l0cwCgn50L izuRo5hXA/ySB2u83GdBUWA= =YuBQ -----END PGP SIGNATURE----- |
|
|
Glen E. P. Ropella wrote:
>> Anything that requires significant short >> term memory and integration of broad but scare evidence is probably >> something a computer will be better at than a human. >> > > That's just plain silly in terms of RR's ideas because _humans_ program > the computer. Until/unless we come up with a computer that programs > itself, or a computer that programs another computer, or something of > that sort, computers will _never_ be better at any task than humans. > > I.e. in RR terms, humans are THE canalizing "efficient cause" for any > computer system. > Fine, so let's move on from RR terms. It seems to be a dead end! Marcus |
|
|
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1 Marcus G. Daniels on 01/08/2008 11:44 AM: > Fine, so let's move on from RR terms. It seems to be a dead end! No, it's not a dead-end. It's just a body of theoretical work that we may or may not need as yet. I fully support the development of theory prior to needing that theory. What if we plug along in the "computationalist" paradigm for the next 100 years and _finally_ realize that, hey! we could have used that gobbledygook Robert Rosen generated? Or, worse yet, what if _forget_ about it completely and end up reinventing it? No, I don't think we should categorize RR terms as a dead-end... not yet, anyway. However, if you don't like speculative theoretical discussions, then feel free to avoid them! [grin] - -- glen e. p. ropella, 971-219-3846, http://tempusdictum.com If people never did silly things, nothing intelligent would ever get done. -- Ludwig Wittgenstein -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFHg9k+ZeB+vOTnLkoRAgo8AJ48T7cnRvaK7aDoOEMYsBgBHynYpgCg2p9/ 7jpoMl79OW5SuwfoGQqNUVI= =zuRw -----END PGP SIGNATURE----- |
|
|
In reply to this post by Phil Henshaw-2
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1 Phil Henshaw on 01/08/2008 11:14 AM: > I thought the implication was that the organization of life is an > inherently ill-posed question from an observer's perspective. To me > that either means you accept 'bad answers' or 'better and better > answers', and the difference is methodological. Sure, if by "ill-posed" you mean "our formalisms can't handle the truth". If you must use a binary categorization of accepting bad answers or better and better answers, then RR's work would fall into the latter category. And he was trying to help us rigorously determine if and how our formalisms are inadequate.... i.e. if we get to the point where we can't accept the poor expressiveness of our formalisms, then what do we do? Well, develop a more powerful formalism ... hence all that hoo-ha about category theory in "Life Itself". - -- glen e. p. ropella, 971-219-3846, http://tempusdictum.com Moderation in temper is always a virtue; but moderation in principle is always a vice. -- Thomas Paine -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFHg9pCZeB+vOTnLkoRAkuaAKC92EpmCOuX7YGG03aPOaAC+h1GawCgtsq9 Z1ggeFcamjQAM3cwKLxElPo= =CQXD -----END PGP SIGNATURE----- |
|
|
In reply to this post by glen ep ropella
Glen E. P. Ropella wrote:
> It's just a body of theoretical work that we > may or may not need as yet. I fully support the development of theory > prior to needing that theory. Fine, and I fully support the deconstruction of theory prior to using it! In what way does Genetic Programming not provide an efficient cause? Having a stochastic aspect, and the possibility to define new instructions, it seems to me to provide an escape from anything a human might have intended. This learning algorithm could escape the constraints of being a `tool' by being used in a robot with similar senses as ours and interacting with the conditions of the `real' world. Marcus |
|
|
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1 Marcus G. Daniels on 01/08/2008 12:46 PM: > Fine, and I fully support the deconstruction of theory prior to using it! That's the spirit! > In what way does Genetic Programming not provide an efficient cause? > Having a stochastic aspect, and the possibility to define new > instructions, it seems to me to provide an escape from anything a human > might have intended. This learning algorithm could escape the > constraints of being a `tool' by being used in a robot with similar > senses as ours and interacting with the conditions of the `real' world. Well, correct me if I'm wrong, but GP currently requires a human to set up the objective function. And even in the cases where a system is created so that the objective function is dynamically (and/or implicitly) evolved, my suspicion is that the GP would soon find a computational exploit that would result in either an infinite loop (and/or deadlock), crash, or some sort of "exception". As for the robot, you're just begging the question. A robot is a tool built and programmed by us. Or, positing a regression to where we are currently, a robot_N that is built by robot_(N-1), that is built by robot_(N-2), ..., is built by a living system. RR's position might be that such a chain from 1 .. N is more fragile than a lineage of living systems. Namely, the efficient cause (humans in this case) cannot be removed even with a large but finite N _because_ machines are not closed to efficient cause. Whether or not RR's rhetoric is _sound_ is one thing. We can prove his rhetoric unsound by creating such a robot lineage. But to prove his rhetoric invalid, we'll have to show that computation is not fragile to ambiguity. And as far as I can tell, such a proof (that RR's rhetoric is invalid) would involve a constructive proof that sets up a holarchy of formal systems that, together, are not fragile in the way GP systems are fragile. Somehow we would have to build a set of (sufficiently complicated, as in modern mathematics) formal systems and prove ([meta-]mathematically) that this set is robust to ambiguity. I.e. it will never go into an infinite (null) loop, crash, or trigger some exception. Of course, we could take the _easier_ tack and point out a technical flaw in RR's rhetoric (as the largely ineffective criticism of Penrose's argument does). My choice for such a cheap shot criticism lies at the heart of "closure to efficient cause". And my criticism is basically that nothing is really closed to efficient cause. Everything is embedded in a dynamically generated and evolving goo that is holistically dependent on everything else in the goo. But even if such cheap shots are successful in getting people to ignore RR, it still doesn't make any progress on RR's main question: "can we devise better formalisms that more accurately describe living systems?" - -- glen e. p. ropella, 971-219-3846, http://tempusdictum.com We are drowning in information, while starving for wisdom. -- E.O. Wilson -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFHg+xkZeB+vOTnLkoRApPWAKCgotysX3Ooh36zeYj7Ipg4Mm59hACdFX+x krJqxKFwyGGc8q99ePPb9X8= =c1fa -----END PGP SIGNATURE----- |
enough of Robert Rosen
|
|
In reply to this post by Marcus G. Daniels
-----Original Message----- From: [hidden email] [mailto:[hidden email]] On Behalf Of Marcus G. Daniels Sent: Tuesday, January 08, 2008 1:47 PM To: The Friday Morning Applied Complexity Coffee Group Subject: Re: [FRIAM] enough of Robert Rosen Glen E. P. Ropella wrote: > It's just a body of theoretical work that we > may or may not need as yet. I fully support the development of theory > prior to needing that theory. Fine, and I fully support the deconstruction of theory prior to using it! In what way does Genetic Programming not provide an efficient cause? Because you need to look beyond the physical plane of existence into the spiritual to see that the soul is the real administrator of the life's path. Genetics is simply the hardware. Genetics is not actually causal and nothing physical can ever be causal. See The Kybalion, Principles of Hermes. Having a stochastic aspect, and the possibility to define new instructions, it seems to me to provide an escape from anything a human might have intended. This learning algorithm could escape the constraints of being a `tool' by being used in a robot with similar senses as ours and interacting with the conditions of the `real' world. Marcus ============================================================ FRIAM Applied Complexity Group listserv Meets Fridays 9a-11:30 at cafe at St. John's College lectures, archives, unsubscribe, maps at http://www.friam.org -- No virus found in this incoming message. Checked by AVG Free Edition. Version: 7.5.516 / Virus Database: 269.17.13/1214 - Release Date: 1/8/2008 1:38 PM _____ I am using the free version of SPAMfighter for private users. It has removed 5220 spam emails to date. Paying users do not have this message in their emails. Try SPAMfighter <http://www.spamfighter.com/len> for free now! -------------- next part -------------- An HTML attachment was scrubbed... URL: http://redfish.com/pipermail/friam_redfish.com/attachments/20080108/2c780d34/attachment.html |
|
|
In reply to this post by glen ep ropella
On Jan 8, 2008, at 10:34 PM, Glen E. P. Ropella wrote: > >> In what way does Genetic Programming not provide an efficient cause? >> Having a stochastic aspect, and the possibility to define new >> instructions, it seems to me to provide an escape from anything a >> human >> might have intended. This learning algorithm could escape the >> constraints of being a `tool' by being used in a robot with similar >> senses as ours and interacting with the conditions of the `real' >> world. > > Well, correct me if I'm wrong, but GP currently requires a human to > set > up the objective function. And even in the cases where a system is > created so that the objective function is dynamically (and/or > implicitly) evolved, my suspicion is that the GP would soon find a > computational exploit that would result in either an infinite loop > (and/or deadlock), crash, or some sort of "exception". This is certainly a good point, but from what I understand of Rosen's theories another limitation of GP has to do with the fact that the language in which the programming is done can not evolve. The syntax will always be circumscribed by a subset of the programming language that is used to set up the GP, and the semantics of what the symbols represent in terms of real-world measurements or actions will be fixed by the robot's senses and actuators. The only novelty possible in such a system are new arrangements of already existing primitives (which in a computationalist view could be enough, actually, if you think like Zuse/Toffoli etc. that the universe really is a big computer anyway, which means that an isomorphism could be possible), but from what I understand from Rosen, Pattee, Pask and Cariani is that novelty in a real, non- platonic (let's say Aristotelic ?) world has to do with the appearance of new primitives: new symbols with new meanings in a new syntax. The construction of symbols in the real world is an open- ended process, which is why no isomorphism with a closed, formal system is possible. right, my two cents worth and no doubt garbled version of what I've been reading lately. (I'm reading the thesis by Peter Cariani right now, who was a student of Pattee and sort of a follower of Rosen). I'm not schooled in these matters in any way, so criticism, yes please, and I will probably have no other reply than pointing to the writings of Cariani and Rosen. ciao, Joost. ------------------------------------------- Joost Rekveld ----------- http://www.lumen.nu/rekveld ------------------------------------------- ?This alone I ask you, O reader, that when you peruse the account of these marvels that you do not set up for yourself as a standard human intellectual pride, but rather the great size and vastness of earth and sky; and, comparing with that Infinity these slender shadows in which miserably and anxiously we are enveloped, you will easily know that I have related nothing which is beyond belief.? (Girolamo Cardano) ------------------------------------------- |
|
|
In reply to this post by glen ep ropella
Glen E. P. Ropella wrote:
>> In what way does Genetic Programming not provide an efficient cause? >> Having a stochastic aspect, and the possibility to define new >> instructions, it seems to me to provide an escape from anything a human >> might have intended. This learning algorithm could escape the >> constraints of being a `tool' by being used in a robot with similar >> senses as ours and interacting with the conditions of the `real' world. >> > > Well, correct me if I'm wrong, but GP currently requires a human to set > up the objective function. And even in the cases where a system is > created so that the objective function is dynamically (and/or > implicitly) evolved, my suspicion is that the GP would soon find a > computational exploit that would result in either an infinite loop > (and/or deadlock), crash, or some sort of "exception". > the agent, but there must be a large implicit emphasis on avoiding death. In a simulated world, a way to deal with exceptions is to trap them, and then reflect that in the objective function. Existing memory management hardware, operating systems and programming languages have good facilities for trapping exceptions. Imagine you have some program evolving in a process on a Linux system. Yes, a program could [try] to allocate so much memory that system would crash, or find a stack exploit (e.g. to get root and compromise the kernel), but by in large the way a broken program would die because the memory management hardware trapped illegal memory requests. If a process actually succeeds in killing the whole system, it's a security bug in the operating system (or secure programming language, etc.). As for infinite loops or deadlocks, these are things that a management process can readily detect. For the worst case, satellites typically have independent monitoring hardware that reboots the main operating system should it become unresponsive. But normally could you just have one software process monitoring the performance of the others. And here I mean performance in the sense of CPU utilization (e.g. is it cycling through the same program counter range over and over) and wall clock runtime. This is all in the context of a simulated environment, of course. In the robot example, the robots would just slump on the ground or jump up and down or whatever until its energy supplies were exhausted. Marcus |
|
|
In reply to this post by glen ep ropella
Glen E. P. Ropella wrote:
> As for the robot, you're just begging the question. A robot is a tool > built and programmed by us. Or, positing a regression to where we are > currently, a robot_N that is built by robot_(N-1), that is built by > robot_(N-2), ..., is built by a living system. > I'm imagining that the program that the robot executes is also genetic program, but one that benefits from richer and more dynamic perceptual data than in purely simulated world. The genetic code can be inherited by the robot (a rusty old robot transfers its instructions to a new shiny robot), or the robot can evolve its own programs during its lifetime using simulation or experiment. The GP candidates are random perturbations against things that sort-of work, so the random noise eventually gets it or at one of its millions of peers out of local minima (relative to its objectives). There's also effectively noise in addition to the signals from the dynamically generated and evolving goo of the environment. Marcus |
|
|
In reply to this post by Joost Rekveld
Joost Rekveld wrote:
> This is certainly a good point, but from what I understand of Rosen's > theories another limitation of GP has to do with the fact that the > language in which the programming is done can not evolve. I don't see why this must be so. One could imagine that a robot had a field programmable gate array that could, in effect, burn an all new processor and bring it online. But, usually when new computer architectures are being developed, the developers just write a software simulator for it in initial stages (that mimics the intended physics of the hardware design). Even the adiabatic quantum computer people at DWave are using existing silicon process technologies to design circuits.. > The syntax > will always be circumscribed by a subset of the programming language > that is used to set up the GP, and the semantics of what the symbols > represent in terms of real-world measurements or actions will be > fixed by the robot's senses and actuators. Biotech, nanotech... ? |
|
|
In reply to this post by Joost Rekveld
Joost Rekveld wrote:
> This is certainly a good point, but from what I understand of Rosen's > theories another limitation of GP has to do with the fact that the > language in which the programming is done can not evolve. 20 amino acids seem to go a long way... :-) |
|
|
In reply to this post by Marcus G. Daniels
On Jan 8, 2008, at 11:52 PM, Marcus G. Daniels wrote: > Joost Rekveld wrote: >> This is certainly a good point, but from what I understand of Rosen's >> theories another limitation of GP has to do with the fact that the >> language in which the programming is done can not evolve. > I don't see why this must be so. One could imagine that a robot > had a > field programmable gate array that could, in effect, burn an all new > processor and bring it online. sure, but can a robot develop representations for other operations than those already in its specifications ? can it design a processor that has some novel feature that is not already possible in the robots current architecture ? > But, usually when new computer > architectures are being developed, the developers just write a > software > simulator for it in initial stages (that mimics the intended > physics of > the hardware design). > Even the adiabatic quantum computer people at DWave are using existing > silicon process technologies to design circuits.. I guess the main creative factor in these examples are the people involved in designing new specifications and defining symbols representing aspects of the new hardware they are developing... > >> The syntax >> will always be circumscribed by a subset of the programming language >> that is used to set up the GP, and the semantics of what the symbols >> represent in terms of real-world measurements or actions will be >> fixed by the robot's senses and actuators. > Biotech, nanotech... ? yes, I guess so. In this Cariani thesis I mentioned he posits some kind of real-world assembly process enabling the construction of new senses and actuators. ( see "On the design of devices with emergent semantic functions", <http://homepage.mac.com/cariani/CarianiWebsite/Cariani89.pdf> ) I guess the crucial difference is that such a self-constructing robot would be grounded in the real world and not in a prespecified computed universe. It would be able to evolve its own computed universe. I'm not sure what to think of all this, but I like Cariani's ideas a lot and so far I haven't found any basic flaw in them. But, as said, being non-schooled in these matters that doesn't necessarily mean very much. On Jan 8, 2008, at 11:56 PM, Marcus G. Daniels wrote: > Joost Rekveld wrote: >> This is certainly a good point, but from what I understand of Rosen's >> theories another limitation of GP has to do with the fact that the >> language in which the programming is done can not evolve. > 20 amino acids seem to go a long way... :-) > characters make no language... cheers, Joost. ------------------------------------------- Joost Rekveld ----------- http://www.lumen.nu/rekveld ------------------------------------------- ?This alone I ask you, O reader, that when you peruse the account of these marvels that you do not set up for yourself as a standard human intellectual pride, but rather the great size and vastness of earth and sky; and, comparing with that Infinity these slender shadows in which miserably and anxiously we are enveloped, you will easily know that I have related nothing which is beyond belief.? (Girolamo Cardano) ------------------------------------------- |
|
|
Joost Rekveld wrote:
> sure, but can a robot develop representations for other operations > than those already in its specifications ? > can it design a processor that has some novel feature that is not > already possible in the robots current architecture ? > The main capability it would offer would be to make certain kinds of calculations more feasible, or more accurate. That could be important for a robot to specialize to new kinds of physical environments, for example. Another reason might be to capture chaotic effects that would only be witnessed with parallel execution. But I'm hard pressed to think of many things that can't be simulated, at least in principle. That's to say it can't be described or modeled, which is to say that conversation about it ridiculous! If not, it's up to the modeler to say specific ways in which a set of primitives are inadequate. >> But, usually when new computer >> architectures are being developed, the developers just write a >> software >> simulator for it in initial stages (that mimics the intended >> physics of >> the hardware design). >> Even the adiabatic quantum computer people at DWave are using existing >> silicon process technologies to design circuits.. >> > > I guess the main creative factor in these examples are the people > involved in designing new specifications and defining symbols > representing aspects of the new hardware they are developing... > symbols, provided the old symbols were from a Turing complete system (and they are). The notion of introducing a symbol or verb to a computational system is no big deal. It's a primitive in programming languages like Lisp. >> 20 amino acids seem to go a long way... :-) >> > > characters make no language... > It seems to me it's the language that's important, and how suitable that language is to the environment at hand. That's not to say there aren't new useful primitives to be discovered. Marcus |
|
|
In reply to this post by Joost Rekveld
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1 I'm going to violate the bottom-post rule because all 3 of the following excerpts focus on the point I made (in response to G?nther) that there's a difference between "computation" as the software that runs on a machine and the machine, itself. When we talk about "computation", are we talking about a concrete _thing_ that exists out there in the world? Or are we talking about an abstract machine that exists only in our minds (or software as the case may be)? Marcus' comments show that he's talking about the former... computers are real machines that can avail themselves of the full machinery of reality. Hence, that type of "computation" isn't limited in the way RR suggests because that's not what "computability" refers to. A robot that can change itself based on sensory-motor interactions with the real world is not a computer in the same sense as a universal turing machine. This distinction provides plenty of fodder for long arguments and confusion between Rosenites. Some even say that an extant, concrete machine in the real world actually is complex_rr in the same sense that a rock or a mountain is (but not a tree or a cat). Others vehemently deny that. The former seem to submit to degrees of complexity_rr whereas the others seem to think it's bivalent. So, I already asked this; but, the conversation really needs a clear understanding of what we mean by "computation". Perhaps we could split it into two categories: computation_c would indicate the activities of a concrete machine and computation_a would indicate the (supposed) activities of a universal turing machine. Joost Rekveld on 01/08/2008 02:13 PM: > isomorphism could be possible), but from what I understand from > Rosen, Pattee, Pask and Cariani is that novelty in a real, non- > platonic (let's say Aristotelic ?) world has to do with the > appearance of new primitives: new symbols with new meanings in a new > syntax. The construction of symbols in the real world is an open- > ended process, which is why no isomorphism with a closed, formal > system is possible. Marcus G. Daniels on 01/08/2008 02:52 PM: > I don't see why this must be so. One could imagine that a robot had > a field programmable gate array that could, in effect, burn an all > new processor and bring it online. But, usually when new computer > architectures are being developed, the developers just write a > software simulator for it in initial stages (that mimics the intended > physics of the hardware design). Even the adiabatic quantum computer > people at DWave are using existing silicon process technologies to > design circuits.. Joost Rekveld on 01/08/2008 03:24 PM: > I guess the crucial difference is that such a self-constructing robot > would be grounded in the real world and not in a prespecified > computed universe. It would be able to evolve its own computed > universe. I'm not sure what to think of all this, but I like > Cariani's ideas a lot and so far I haven't found any basic flaw in > them. But, as said, being non-schooled in these matters that doesn't > necessarily mean very much. - -- glen e. p. ropella, 971-219-3846, http://tempusdictum.com Government never furthered any enterprise but by the alacrity with which it got out of its way. -- Henry David Thoreau -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFHhB7PZeB+vOTnLkoRAkxPAJkBFRfqeFx/UOEwqm05yJOZ8WHO9gCfTefY HYWqQsjEqLVI5D13iIW0zoc= =WGg8 -----END PGP SIGNATURE----- |
|
|
In reply to this post by Marcus G. Daniels
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1 Marcus G. Daniels on 01/08/2008 02:18 PM: > Glen E. P. Ropella wrote: >> Well, correct me if I'm wrong, but GP currently requires a human to set >> up the objective function. And even in the cases where a system is >> created so that the objective function is dynamically (and/or >> implicitly) evolved, my suspicion is that the GP would soon find a >> computational exploit that would result in either an infinite loop >> (and/or deadlock), crash, or some sort of "exception". >> > The objective function can be to an extent arbitrary and self-defined by > the agent, but there must be a large implicit emphasis on avoiding > death. In a simulated world, a way to deal with exceptions is to trap > them, and then reflect that in the objective function. Existing memory > management hardware, operating systems and programming languages have > good facilities for trapping exceptions. Aha! What you're implicitly referring to, here, is an assemblage (though not a holarchy) of formal systems ... just like I suggested. [grin] If inference within one of the formal systems (e.g. memory allocation) reaches an impasse, the machine hops out of that system and into one that has a different semantic grounding (e.g. the OS or a hardware driver) takes over and "plugs the hole". After the hole is plugged, it hops back inside the prior formal system and continues on. _Or_ the latter formal system, through its inference modifies the former formal system (new axiom, new alphabet, whatever) such that the previous exception can no longer obtain. Of course, in that case, it's probably true that the inference int he former system has to be re-run from the start rather than picking up where it left off... but, hey, c'est la vie. The reason I suggested a holarchy rather than just an adhoc assemblage of systems, however, is important because it's unlikely we'd be able to design an assemblage of formal systems to handle every exception. (Sorry for repeating myself...) So, what's necessary is either the on-the-fly generation of new systems along with on-the-fly re-architecting of the assemblage OR a holarchy where every sub-system, regardless of what level it's at, is further composed of sub-sub-systems. > runtime. This is all in the context of a simulated environment, of > course. In the robot example, the robots would just slump on the ground > or jump up and down or whatever until its energy supplies were exhausted. Well, this is another example of fragility to ambiguity and, to some extent is the heart of my cheap shot criticism of RR's concept. The robot should fail gracefully (like living systems do). A robot endowed with the holarchy of formal systems would do everything in its power to avoid slumping on the ground or doing something over and over with no discernible effect. I.e. it would _explore_ not only its own repertoire (determined by the formal systems) but also the repertoire of its environment. Hence, a robot would find ways to harness things in its environment to plug any holes (resolve any ambiguities) it couldn't otherwise plug. E.g. a troubled robot may well find itself replacing its aging transistor-based "computer" with, say, a bag of wet fat/meat it harvests from that annoying human who lives in the apartment next door. - -- glen e. p. ropella, 971-219-3846, http://tempusdictum.com Do not fear to be eccentric in opinion, for every opinion now accepted was once eccentric. -- Bertrand Russell -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFHhCONZeB+vOTnLkoRAh5RAJ9fqcffe75m7axl9b1u8z1Rvbq/gACgkaTS FTBfh0LyX/ibYot7lIgitN8= =cUMT -----END PGP SIGNATURE----- |
«
Return to Friam
|
1 view|%1 views
| Free forum by Nabble | Edit this page |