FRIAM and causality
Locked
12
12
 
|
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1 Sorry for breaking the threading. Phil wrote: > But isn't the shape of our varying ability to fit our models a direct > image of 'nature itself', in fact, and our main mistake to discard > them all but the 'best' one and so loose the shape of what they are > all unable to describe? That's why I like to go back and forth > studying alternate models for their discrepancies and their fit, > using models as learning tools rather than answers. I think the > notable thing you find that way is independent whole systems...i Yes! Sheesh, your prose is so hard to parse it feels good when I finally do parse it. [grin] Anyway, I definitely agree that it's a "mistake" in some sense to discard all but the best projections. However, in cases where a limit _exists_ (and it is reasonable to believe it exists), then it's not a mistake at all. Preserving an erroneous model when much more accurate models are at hand would be perverse (or evidence that one should be a historian rather than a scientist). I'm not talking about the type of preservation that allows us to think back and learn from previous events. I'm talking about someone _sticking_ to and/or regularly relying on a "bad" model even when they know it's wrong. However, in most cases, we have no idea if the limit even exists and it is often just psychological bias or delusion that makes us believe in such a limit. And in _those_ cases (MOST cases) it is definitely a mistake to discard any model that is reasonably effective. (Notice my shift from "erroneous" or "accurate" to "effective".) Personally, I believe this is the fundamental point of critical rationalism and _open_ science where we allow and seriously consider _any_ hypothesis, no matter how bizarre or offensive. Only when a hypothesis is falsified should it be demoted to secondary consideration or the history books. - -- glen e. p. ropella, 971-219-3846, http://tempusdictum.com We must respect the other fellow's religion, but only in the sense and to the extent that we respect his theory that his wife is beautiful and his children smart. -- H.L. Mencken -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFHTfzbZeB+vOTnLkoRAon3AJwLpmeuuW86PeKLEjj9Raw+erP23ACgtOcM UPMukBlumR6ywMMkAb9TF0M= =5vqn -----END PGP SIGNATURE----- |
|
|
All scientific models/theories tend to lie on a plane with the axes
"accuracy" and "ease of use". Explicability is also there, roughly aligned with "ease of use". Basically we should only keep those theories/models that lie on the Pareto front, and discard those that are dominated. This is why we still keep Newtonian gravity, even though it is less accurate than GR (ie falsified), but discard the Ptolomaic system. Cheers. On Wed, Nov 28, 2007 at 03:42:19PM -0800, Glen E. P. Ropella wrote: > > Anyway, I definitely agree that it's a "mistake" in some sense to > discard all but the best projections. However, in cases where a limit > _exists_ (and it is reasonable to believe it exists), then it's not a > mistake at all. Preserving an erroneous model when much more accurate > models are at hand would be perverse (or evidence that one should be a > historian rather than a scientist). I'm not talking about the type of > preservation that allows us to think back and learn from previous > events. I'm talking about someone _sticking_ to and/or regularly > relying on a "bad" model even when they know it's wrong. > -- ---------------------------------------------------------------------------- A/Prof Russell Standish Phone 0425 253119 (mobile) Mathematics UNSW SYDNEY 2052 hpcoder at hpcoders.com.au Australia http://www.hpcoders.com.au ---------------------------------------------------------------------------- |
|
|
Russell,
That's a sound way to choose the most valuable model of the moment, but it won't help you with what models can't show. You need to study the space between the models. If you use optimal models and study the discrepancy between them and the continually changing systems they imperfectly reflect, you have a chance of seeing and engaging with the real thing. Models are inherently lifeless, and quite unlike the inventive independent networks we find in the complex physical world. Using the 'best' model to represent nature is like putting a high resolution picture of a frog in your son's terrarium. Very nice, but not the real thing. Assuming that all behavior is deterministic, just waiting for us to find the formula, still lingers. It blocks learning about what we can't write formulas for, though, so I think it should be among the first things to go. Phil Henshaw ????.?? ? `?.???? ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 680 Ft. Washington Ave NY NY 10040 tel: 212-795-4844 e-mail: pfh at synapse9.com explorations: www.synapse9.com > -----Original Message----- > From: friam-bounces at redfish.com > [mailto:friam-bounces at redfish.com] On Behalf Of Russell Standish > Sent: Wednesday, November 28, 2007 8:11 PM > To: The Friday Morning Applied Complexity Coffee Group > Subject: Re: [FRIAM] FRIAM and causality > > > All scientific models/theories tend to lie on a plane with > the axes "accuracy" and "ease of use". Explicability is also > there, roughly aligned with "ease of use". > > Basically we should only keep those theories/models that lie > on the Pareto front, and discard those that are dominated. > This is why we still keep Newtonian gravity, even though it > is less accurate than GR (ie falsified), but discard the > Ptolomaic system. > > Cheers. > > On Wed, Nov 28, 2007 at 03:42:19PM -0800, Glen E. P. Ropella wrote: > > > > > Anyway, I definitely agree that it's a "mistake" in some sense to > > discard all but the best projections. However, in cases > where a limit > > _exists_ (and it is reasonable to believe it exists), then > it's not a > > mistake at all. Preserving an erroneous model when much > more accurate > > models are at hand would be perverse (or evidence that one > should be a > > historian rather than a scientist). I'm not talking about > the type of > > preservation that allows us to think back and learn from previous > > events. I'm talking about someone _sticking_ to and/or regularly > > relying on a "bad" model even when they know it's wrong. > > > > -- > > -------------------------------------------------------------- > -------------- > A/Prof Russell Standish Phone 0425 253119 (mobile) > Mathematics > UNSW SYDNEY 2052 hpcoder at hpcoders.com.au > Australia http://www.hpcoders.com.au > -------------------------------------------------------------- > -------------- > > ============================================================ > 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 glen ep ropella
Glen
> > Sorry for breaking the threading. > > Phil wrote: > > But isn't the shape of our varying ability to fit our > models a direct > > image of 'nature itself', in fact, and our main mistake to discard > > them all but the 'best' one and so loose the shape of what they are > > all unable to describe? That's why I like to go back and forth > > studying alternate models for their discrepancies and their > fit, using > > models as learning tools rather than answers. I think the notable > > thing you find that way is independent whole systems...i > > Yes! > > Sheesh, your prose is so hard to parse it feels good when I > finally do parse it. [grin] Well, partly that's from my mad editing scheme... :-,) repeated word substitution looking for ways to suggest difficult ideas. I too find on rereading that it can get disjointed... Glad it occasionally works! > > Anyway, I definitely agree that it's a "mistake" in some > sense to discard all but the best projections. However, in > cases where a limit _exists_ (and it is reasonable to believe > it exists), then it's not a mistake at all. Yes, like the data showing that economies are approaching a thermodynamic limit in using energy to create wealth. I use approaching limits to narrow down my definitions of natural structures and categories all the time. My preference is using them like an envelope with a space in-between, though. When you use upper and lower bounds to home in on a natural subject, fitting to it with a matching shape, like a ball in a catcher's mitt, your image of any prominence in the natural structure is self-centering. By 'pointing around' it rather than 'pointing at' it you're also both more likely to capture where the natural structure is located and less likely to represent it as being your model. > Preserving an > erroneous model when much more accurate models are at hand > would be perverse (or evidence that one should be a historian > rather than a scientist). I'm not talking about the type of > preservation that allows us to think back and learn from > previous events. I'm talking about someone _sticking_ to > and/or regularly relying on a "bad" model even when they know > it's wrong. A historical view is a fine way to gain more perspective on how our images really fit with nature. The complex world is far too complicated and full of independently behaving things, and any way to begin to appreciate that seems fine. Then as we begin using models as learning tools rather than representational tools, looking for which models are the 'most help for learning' rather than the 'least wrong for representation', I think the focus moves toward modeling as a learning process. > > However, in most cases, we have no idea if the limit even > exists and it is often just psychological bias or delusion > that makes us believe in such a limit. And in _those_ cases > (MOST cases) it is definitely a mistake to discard any model > that is reasonably effective. (Notice my shift from > "erroneous" or "accurate" to "effective".) Yes! They help you recall your own thought processes and it's branchings. Another way to make discarded models more useful can be to break them up. Turning well made things into a clutter of probably useless parts might seem confusing, but like compost they may contain very useful parts for some unforeseen purpose. That can even be a specific strategy for evolving complex systems sometimes. Economies use that method of creative reinvention fairly often, capitalizing the fortuitous design of waste products and byproducts. > > Personally, I believe this is the fundamental point of > critical rationalism and _open_ science where we allow and > seriously consider _any_ hypothesis, no matter how bizarre or > offensive. Only when a hypothesis is falsified should it be > demoted to secondary consideration or the history books. Sure, while not discarding too much, and we should still keep the word 'falsified'. False theories, say like those of Freud or Lamarck, the flat earth or idealized determinism, can offer fruitful ground for asking what made them so compelling. ...does this go anywhere you think? Best, Phil > - -- > glen e. p. ropella, 971-219-3846, http://tempusdictum.com > We must respect the other fellow's religion, but only in the > sense and to the extent that we respect his theory that his > wife is beautiful and his children smart. -- H.L. Mencken > |
|
|
In reply to this post by Phil Henshaw-2
This is a very "Phil Henshaw" response - its a bit hard to know how to
respond to this. On Thu, Nov 29, 2007 at 10:14:41AM -0500, Phil Henshaw wrote: > Russell, > That's a sound way to choose the most valuable model of the moment, but > it won't help you with what models can't show. You need to study the > space between the models. If you use optimal models and study the > discrepancy between them and the continually changing systems they > imperfectly reflect, you have a chance of seeing and engaging with the > real thing. > So you're just saying we should be performing crossover operations between successful models? But this is exactly what happens when multidisciplinary teams form leading to cross-polination of ideas. The results are often quite interesting and advance the field. > Models are inherently lifeless, and quite unlike the inventive > independent networks we find in the complex physical world. As a long time ALife practitioner, I don't really believe this at all. I have often been surprised at the behaviour of my models, even lifelike behaviour. > Using the > 'best' model to represent nature is like putting a high resolution > picture of a frog in your son's terrarium. Very nice, but not the real > thing. Nice metaphor, but I don't understand how it relates... What about replacing the frog with a detailed robotic imitation that has been evolved to imitate frog behaviour using artificial life techniques? > Assuming that all behavior is deterministic, just waiting for us > to find the formula, still lingers. What do you think of stochastic descriptions of nature then (starting with Boltzmann's statistical physics)? > It blocks learning about what we > can't write formulas for, though, so I think it should be among the > first things to go. > What we cannot "write formulas for" (by which I mean "find compressible descriptions for"), we cannot learn. For that is the very nature of learning - being able to generalise from the specific. -- ---------------------------------------------------------------------------- A/Prof Russell Standish Phone 0425 253119 (mobile) Mathematics UNSW SYDNEY 2052 hpcoder at hpcoders.com.au Australia http://www.hpcoders.com.au ---------------------------------------------------------------------------- |
|
|
Biting my lip over here. [Don't respond...DON'T RESPOND!]
-- Doug Roberts, RTI International droberts at rti.org doug at parrot-farm.net 505-455-7333 - Office 505-670-8195 - Cell On Dec 5, 2007 5:14 PM, Russell Standish <r.standish at unsw.edu.au> wrote: > This is a very "Phil Henshaw" response - its a bit hard to know how to > respond to this. > > On Thu, Nov 29, 2007 at 10:14:41AM -0500, Phil Henshaw wrote: > > Russell, > > That's a sound way to choose the most valuable model of the moment, but > > it won't help you with what models can't show. You need to study the > > space between the models. If you use optimal models and study the > > discrepancy between them and the continually changing systems they > > imperfectly reflect, you have a chance of seeing and engaging with the > > real thing. > > > > So you're just saying we should be performing crossover operations > between successful models? But this is exactly what happens when > multidisciplinary teams form leading to cross-polination of ideas. The > results are often quite interesting and advance the field. > > > Models are inherently lifeless, and quite unlike the inventive > > independent networks we find in the complex physical world. > > As a long time ALife practitioner, I don't really believe this at > all. I have often been surprised at the behaviour of my models, even > lifelike behaviour. > > > Using the > > 'best' model to represent nature is like putting a high resolution > > picture of a frog in your son's terrarium. Very nice, but not the real > > thing. > > Nice metaphor, but I don't understand how it relates... What about > replacing the frog with a detailed robotic imitation that has been > evolved to imitate frog behaviour using artificial life techniques? > > > Assuming that all behavior is deterministic, just waiting for us > > to find the formula, still lingers. > > What do you think of stochastic descriptions of nature then (starting with > Boltzmann's statistical physics)? > > > It blocks learning about what we > > can't write formulas for, though, so I think it should be among the > > first things to go. > > > > What we cannot "write formulas for" (by which I mean "find > compressible descriptions for"), we cannot learn. For that is the very > nature of learning - being able to generalise from the specific. > > > -- > > > ---------------------------------------------------------------------------- > A/Prof Russell Standish Phone 0425 253119 (mobile) > Mathematics > UNSW SYDNEY 2052 hpcoder at hpcoders.com.au > Australia http://www.hpcoders.com.au > > ---------------------------------------------------------------------------- > > ============================================================ > 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 > An HTML attachment was scrubbed... URL: http://redfish.com/pipermail/friam_redfish.com/attachments/20071205/cca57354/attachment.html |
|
|
In reply to this post by glen ep ropella
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1 OK. I hope this is the last time I have to break the threading. My upgrade is stalled; so my exim4 should work fine for now. [grin] Phil on Thu Nov 29 at 11:46:09 EST 2007 wrote: > Sure, while not discarding too much, and we should still keep the word > 'falsified'. False theories, say like those of Freud or Lamarck, the > flat earth or idealized determinism, can offer fruitful ground for > asking what made them so compelling. > > ...does this go anywhere you think? Well, going back to your original objection to my use of the word "any", I think it does go somewhere. My statement was that any actual (a.k.a. realized, "real"... whatever word you use) system can be projected onto any ordering (or any measure in general). The resulting projection may be a gross distortion of the system or a relatively accurate representation. My point was simply that multiple models are necessary. But, taking your point that even gross distortions are useful for learning, we might posit that not only are multiple models necessary, but the _distribution_ of those models must have a certain character. E.g. perhaps really "bad" models _must_ be included in order to understand the system. I'd say that "goes somewhere". - -- glen e. p. ropella, 971-219-3846, http://tempusdictum.com See, in my line of work you got to keep repeating things over and over and over again for the truth to sink in, to kind of catapult the propaganda. -- George W. Bush -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFHV0cgZeB+vOTnLkoRAujNAJ9RsNyeitlPOxKtq4fYL+2CqtHL5wCfehg9 gZMEtsP8SvOR9pMFYD64774= =0UeV -----END PGP SIGNATURE----- |
|
|
Glen,
Excellent! If they're honestly derived from physical things, like network maps, say, every model is going to be both a 'bad' model and a helpful one. The principle comes to this complex statement, yes, but I think also to a simple one that to understand anything you need multiple measures. A measure is a sort of simplistic model. You could say, "He's 6'1"." and have a model of a person that has some helpful and some bad features. Measure is informative, but it's also a reductive projection of only one dimension or set of relationships from the subject. It's like taking Lincoln's quip "A tree is best measured when it is down" and turning it backwards, to say 'it takes many kinds of measure to begin getting the whole picture'... The hard part seems to be to take the first dark step to accepting there might be a shape of another form that the measures are missing (like the whole tree or person). It means looking for how to best extend and complete your image based on the limited cast of the measures at hand. Interpolation gone wild?? Free form projection perhaps?? Sort of... You just gotta do something to make sense of the larger continuities that develop in natural complex systems. What I think we can see clearly is that our measures and models are highly incomplete. Phil > > OK. I hope this is the last time I have to break the > threading. My upgrade is stalled; so my exim4 should work > fine for now. [grin] > > Phil on Thu Nov 29 at 11:46:09 EST 2007 wrote: > > Sure, while not discarding too much, and we should still > keep the word > > 'falsified'. False theories, say like those of Freud or > Lamarck, the > > flat earth or idealized determinism, can offer fruitful ground for > > asking what made them so compelling. > > > > ...does this go anywhere you think? > > Well, going back to your original objection to my use of the > word "any", I think it does go somewhere. My statement was > that any actual (a.k.a. realized, "real"... whatever word you > use) system can be projected onto any ordering (or any > measure in general). The resulting projection may be a gross > distortion of the system or a relatively accurate representation. > > My point was simply that multiple models are necessary. But, > taking your point that even gross distortions are useful for > learning, we might posit that not only are multiple models > necessary, but the _distribution_ of those models must have a > certain character. E.g. perhaps really "bad" models _must_ > be included in order to understand the system. I'd say that > "goes somewhere". > > - -- > glen e. p. ropella, 971-219-3846, http://tempusdictum.com > See, in my line of work you got to keep repeating things over > and over and over again for the truth to sink in, to kind of > catapult the propaganda. -- George W. Bush > > -----BEGIN PGP SIGNATURE----- > Version: GnuPG v1.4.6 (GNU/Linux) > Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org > > iD8DBQFHV0cgZeB+vOTnLkoRAujNAJ9RsNyeitlPOxKtq4fYL+2CqtHL5wCfehg9 > gZMEtsP8SvOR9pMFYD64774= > =0UeV > -----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 Russell Standish
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1 Russell Standish on 12/05/2007 04:14 PM: >> It blocks learning about what we >> can't write formulas for, though, so I think it should be among the >> first things to go. >> > > What we cannot "write formulas for" (by which I mean "find > compressible descriptions for"), we cannot learn. For that is the very > nature of learning - being able to generalise from the specific. I don't really disagree with you. But, given Douglas' self-restraint, someone must jump in... and since I'm not afraid of making an ass out of myself, it may as well be me. [grin] I don't think it's completely true that we cannot learn something that won't submit to a "good" compressible description. The difference between tacit experience and explicit knowledge highlights that some things (which seem to resist "good" compressed description) are _learnable_. The catch lies in the definition of "learning". Granted, personal (particular) experience can't be completely transpersonal. Hence, all tacit experience has some element(s) that cannot be formulated as explicit knowledge and transmitted. But, we can _appeal_ to (or rely upon) some psychologically projected (imputed) commonalities between us. For example, if I see you wearing shoes with laces, I can assume that you know what it's like to tie your shoes. (No smart-alack remarks about being dressed by one's mother!) And in that sense, even if I can't write a formula for "tying one's shoes", I can still _learn_ how to tie shoes. Further, I can use the inaccurate ("bad") formulas for how to tie one's shoes as a way to actually learn how to tie shoes. Even further, I can _teach_ others how to tie their shoes based on these "bad" models. Hence, we can use "bad" models to learn something that has no "good" model. One might even go so far as to say _that's_ the very nature of learning, not as you characterize it above. Note, however, that this is pre-scientific. Science (and all externalized, transpersonal methods) relies wholeheartedly on making as much knowledge as explicit as possible. If one went that far, it would be reasonable to assume that the more autistic and rational of us learn best through the development of compressed descriptions, whereas those of us addicted to getting our hands dirty learn best through the use of "bad" models and direct experience of applying those "bad" models. - -- glen e. p. ropella, 971-219-3846, http://tempusdictum.com Morality cannot exist one minute without freedom... Only a free man can possibly be moral. Unless a good deed is voluntary, it has no moral significance. -- Everett Martin -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFHWE24ZeB+vOTnLkoRAn7NAKCk56nz+lA1qPEG5yzySQnIqKutvwCgx90G /bGBq9jNrFcyaX10b3HuTUs= =yNb8 -----END PGP SIGNATURE----- |
|
|
>
> And in that > sense, even if I can't write a formula for "tying one's shoes", I can > still _learn_ how to tie shoes. Further, I can use the inaccurate > ("bad") formulas for how to tie one's shoes as a way to actually learn > how to tie shoes. Even further, I can _teach_ others how to tie their > shoes based on these "bad" models. What's the metric you're using for good and bad here? That one person looked it up on Wikipedia and another person learned it from their mom, i.e. formal vs. informal description? Or ability to stay tied vs. ease in shoe removal, or?? Or some mixture of these features? Who decides the relative weights for goodness? |
|
|
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1 Marcus G. Daniels on 12/06/2007 12:23 PM: >> And in that >> sense, even if I can't write a formula for "tying one's shoes", I can >> still _learn_ how to tie shoes. Further, I can use the inaccurate >> ("bad") formulas for how to tie one's shoes as a way to actually learn >> how to tie shoes. Even further, I can _teach_ others how to tie their >> shoes based on these "bad" models. > > What's the metric you're using for good and bad here? That one person > looked it up on Wikipedia and another person learned it from their mom, > i.e. formal vs. informal description? Or ability to stay tied vs. ease > in shoe removal, or?? Or some mixture of these features? Who decides > the relative weights for goodness? I'm not using a measure ("metric" is the wrong word) at all. My statements are measure-independent. _All_ measures provide an incomplete description of any system. "The map is not the territory." If one can find a measure that is complete, then that measure _is_ the system. How one determines whether a given measure is better than another depends entirely on their purpose at the time. - -- glen e. p. ropella, 971-219-3846, http://tempusdictum.com With or without religion, you would have good people doing good things and evil people doing evil things. But for good people to do evil things, that takes religion. -- Steven Weinberg -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFHWGDaZeB+vOTnLkoRAtrKAJkBf7YZx94ctNK44bd6oJ4LnS6sVwCgjSjH sQgbSJ+cilF6Ig33+GHjHoY= =Y34k -----END PGP SIGNATURE----- |
|
|
In reply to this post by Marcus G. Daniels
On Thu, Dec 06, 2007 at 01:23:00PM -0700, Marcus G. Daniels wrote:
> > > > And in that > > sense, even if I can't write a formula for "tying one's shoes", I can > > still _learn_ how to tie shoes. Further, I can use the inaccurate > > ("bad") formulas for how to tie one's shoes as a way to actually learn > > how to tie shoes. Even further, I can _teach_ others how to tie their > > shoes based on these "bad" models. > What's the metric you're using for good and bad here? That one person > looked it up on Wikipedia and another person learned it from their mom, > i.e. formal vs. informal description? Or ability to stay tied vs. ease > in shoe removal, or?? Or some mixture of these features? Who decides > the relative weights for goodness? > There is no formalised metric, and it may not even be formalisable. However, a good model is one that has utility, it can predict stuff, or explain stuff, or a mixture of the two. A better model is one that can both explain and predict stuff better than the other model. Otherwise other models that can better predict or explain stuff are just other good models. Obviously there is a certain amount of subjectivity here - some folk think that a model "God did it" explains stuff, but explaining stuff in terms of a mysterious, unexplainable, all powerful entity doesn't work for me, nor for most scientists. Cheers -- ---------------------------------------------------------------------------- A/Prof Russell Standish Phone 0425 253119 (mobile) Mathematics UNSW SYDNEY 2052 hpcoder at hpcoders.com.au Australia http://www.hpcoders.com.au ---------------------------------------------------------------------------- |
|
|
In reply to this post by glen ep ropella
On Thu, Dec 06, 2007 at 11:30:00AM -0800, Glen E. P. Ropella wrote:
> -----BEGIN PGP SIGNED MESSAGE----- > Hash: SHA1 > > Russell Standish on 12/05/2007 04:14 PM: > >> It blocks learning about what we > >> can't write formulas for, though, so I think it should be among the > >> first things to go. > >> > > > > What we cannot "write formulas for" (by which I mean "find > > compressible descriptions for"), we cannot learn. For that is the very > > nature of learning - being able to generalise from the specific. > > I don't really disagree with you. But, given Douglas' self-restraint, > someone must jump in... and since I'm not afraid of making an ass out of > myself, it may as well be me. [grin] > > I don't think it's completely true that we cannot learn something that > won't submit to a "good" compressible description. The difference > between tacit experience and explicit knowledge highlights that some > things (which seem to resist "good" compressed description) are > _learnable_. The catch lies in the definition of "learning". > > Granted, personal (particular) experience can't be completely > transpersonal. Hence, all tacit experience has some element(s) that > cannot be formulated as explicit knowledge and transmitted. But, we can > _appeal_ to (or rely upon) some psychologically projected (imputed) > commonalities between us. For example, if I see you wearing shoes with > laces, I can assume that you know what it's like to tie your shoes. (No > smart-alack remarks about being dressed by one's mother!) And in that > sense, even if I can't write a formula for "tying one's shoes", I can > still _learn_ how to tie shoes. Further, I can use the inaccurate > ("bad") formulas for how to tie one's shoes as a way to actually learn > how to tie shoes. Even further, I can _teach_ others how to tie their > shoes based on these "bad" models. Well I didn't have in mind cerebellum type learning, but rather cerebral learning. I'm not sure whether the cerebellum involves compressible descriptions or not. I had never come across this notion of tacit knowledge before - I've just read the Wikipedia article on it. It certainly wasn't what I had in mind when discussing learning, but I wasn't explicit about that (excuse the pun!). Note that inaccurate is not necessarily bad. Newtonian gravity is inaccurate (it gets Mercury's orbit wrong in a detectable way), but it is still a good model (you can fly spacecraft using it). Bad just means lack of any utility. > > Hence, we can use "bad" models to learn something that has no "good" model. > > One might even go so far as to say _that's_ the very nature of learning, > not as you characterize it above. Note, however, that this is > pre-scientific. Science (and all externalized, transpersonal methods) > relies wholeheartedly on making as much knowledge as explicit as > possible. If one went that far, it would be reasonable to assume that > the more autistic and rational of us learn best through the development > of compressed descriptions, whereas those of us addicted to getting our > hands dirty learn best through the use of "bad" models and direct > experience of applying those "bad" models. > If you are learning, the models you use must be getting better, and don't remain bad. And yes, these models are compressed descriptions of reality. However, whether tacit learning involves models, or not I simply don't know. It is possible that I have some form of model of how my bicycle behaves buried in cerebellum which is used as part of a predictor-corrector loop. It is also possible that this is just an interpretation of what is a tangle of neurons and synaptic weights, just as we might say an artificial neural network is a predictive model (but not at all explanatory), whereas a fuzzy logic controller is (at least partially) explanatory. > - -- > glen e. p. ropella, 971-219-3846, http://tempusdictum.com > Morality cannot exist one minute without freedom... Only a free man can > possibly be moral. Unless a good deed is voluntary, it has no moral > significance. -- Everett Martin > > -----BEGIN PGP SIGNATURE----- > Version: GnuPG v1.4.6 (GNU/Linux) > Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org > > iD8DBQFHWE24ZeB+vOTnLkoRAn7NAKCk56nz+lA1qPEG5yzySQnIqKutvwCgx90G > /bGBq9jNrFcyaX10b3HuTUs= > =yNb8 > -----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 -- ---------------------------------------------------------------------------- A/Prof Russell Standish Phone 0425 253119 (mobile) Mathematics UNSW SYDNEY 2052 hpcoder at hpcoders.com.au Australia http://www.hpcoders.com.au ---------------------------------------------------------------------------- |
|
|
In reply to this post by Russell Standish
Russell Standish wrote:
> There is no formalised metric, and it may not even be > formalisable. However, a good model is one that has utility, it can > predict stuff, or explain stuff, or a mixture of the two. Well, it seems to me once there is a question, then bad and good become grounded (if not found) and we can stop going in circles. The answer may be loafers, or it may be a square knot, or it may involve some funny looking lambda calculus stuff that compresses the how-to-tie-shoes story to some minimal number of characters. |
|
|
In reply to this post by Phil Henshaw-2
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1 Phil Henshaw on 12/06/2007 10:53 AM: > The hard part seems to be to take the first dark step to accepting there > might be a shape of another form that the measures are missing (like the > whole tree or person). It means looking for how to best extend and > complete your image based on the limited cast of the measures at hand. > Interpolation gone wild?? Free form projection perhaps?? Sort of... > You just gotta do something to make sense of the larger continuities > that develop in natural complex systems. What I think we can see > clearly is that our measures and models are highly incomplete. I think we agree, which normally means there's nothing to talk about! [grin] But, I thought I'd throw out my term for what you're describing: "triangulation". It's not really triangulation, of course. But it's certainly more like triangulation than, say, population sampling. Perhaps we could call it "tuple-angulation"??? [grin] Here's a paper in which "we" (i.e. my outrageous rhetoric is reigned in and made coherent by the authors of the paper ;-) try to describe it: http://www.biomedcentral.com/1752-0509/1/14/abstract See Figure 1. This particular example is just one sub-type of the general method we're talking about, here, though. - -- glen e. p. ropella, 971-219-3846, http://tempusdictum.com If this were a dictatorship, it would be a heck of a lot easier, just so long as I'm the dictator. -- George W. Bush -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFHWX6LZeB+vOTnLkoRAhfiAJ4ldUf3p2wtlih3736TIp28uVtEZACfWyMf Pi/MX4iy1xD4PrqQNyNvbYo= =9GWs -----END PGP SIGNATURE----- |
|
|
Phil Henshaw wrote:
> The hard part seems to be to take the first dark step to accepting there > might be a shape of another form that the measures are missing (like the > whole tree or person). Glen E. P. Ropella wrote: > See Figure 1. This particular example is just one sub-type of the > general method we're talking about, here, though. > Figure 1 concerns using behavioral distributions estimated from in vitro data to constrain the choice of parameters/tuples/object composition/etc. in an agent model -- model fitting. Phil seems to be talking about the situation where it isn't yet clear what to measure -- theory driving experiment, e.g. the development of general relativity preceding experiments to find gravitational waves. |
|
|
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1 Marcus G. Daniels on 12/07/2007 10:37 AM: > Figure 1 concerns using behavioral distributions estimated from in vitro > data to constrain the choice of parameters/tuples/object > composition/etc. in an agent model -- model fitting. No. It concerns the iterative construction of new models (and new measures) which behave (according to the chosen measures) more like the reference model. It is not model fitting in the sense of merely tuning parameters so that a model reproduces the data. In this sense, it is not about model fitting. It is about triangulating around the actual system in an effort to gain a better understanding (and way to characterize) the behavior of the actual system. The fact that the in vitro model is not iterated is just a consequence of practical matters. Both the in vitro model and the in silico model are modified to generate the vague phenotype of the actual system. > Phil seems to be > talking about the situation where it isn't yet clear what to measure -- > theory driving experiment, e.g. the development of general relativity > preceding experiments to find gravitational waves. It's the same thing. It's just the case that our models are filled with nitty gritty (and obfuscating) particular details. It may be tempting to think that we already _know_ precisely what to measure. But we do NOT. We use prior models (including measures) in order to triangulate toward better models. - -- glen e. p. ropella, 971-219-3846, http://tempusdictum.com Think not those faithful who praise all thy words and actions; but those who kindly reprove thy faults. -- Socrates -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFHWaPbZeB+vOTnLkoRAqo0AKCpwKG+5Qhdx9J7CduLqeXU4QnVowCdGvvb s//LKWhufSQKWnlHslu1A2w= =Zvww -----END PGP SIGNATURE----- |
|
|
In reply to this post by glen ep ropella
Glen,
I think I missed some replies there somehow, glad you picked it up. > Russell Standish on 12/05/2007 04:14 PM: > >> It blocks learning about what we > >> can't write formulas for, though, so I think it should be > among the > >> first things to go. > >> > > > > What we cannot "write formulas for" (by which I mean "find > > compressible descriptions for"), we cannot learn. For that > is the very > > nature of learning - being able to generalise from the specific. > > I don't really disagree with you. But, given Douglas' > self-restraint, someone must jump in... and since I'm not > afraid of making an ass out of myself, it may as well be me. [grin] > > I don't think it's completely true that we cannot learn > something that won't submit to a "good" compressible > description. The difference between tacit experience and > explicit knowledge highlights that some things (which seem to > resist "good" compressed description) are _learnable_. The > catch lies in the definition of "learning". PH: I wouldn't disagree that there are some kinds of generalities that are so solid that they seem like they're real things. They're certainly real for us, but I think, they are still physically located in our minds. The conservation laws, for example, are highly reliable, but don't seem to have a physical presence except in our minds. Some learning tasks are sort of the opposite, and can't result in capturing what is outside our minds in a reliable form to hold inside our minds. Complex systems are one of those, just not reducible. I don't see it as an all-or-nothing kind of thing, though. When I see a growth curve I usually find confirming evidence that what produced it contained an evolving network of relationships. I can usually extract some descriptors from the records of it for analysis. That analysis usually directs my attention to hidden features of the evolving network that I hadn't thought of looking for. I'm also usually able to confirm those things are there and learn new things about them with a little more study. I don't require a picture in my mind of what I don't see, and accept I may never understand the whole system but. I still learn useful things by looking for what gives it continuity. From experience I learn that a directed search for continuities not yet seen will keep revealing more. One good example of exploring continuities to discover new structures are the discoveries of physics that came from equations in which the denominators tended to zero. Those revealed parts of nature that violated our equations, and that was very useful. > > Granted, personal (particular) experience can't be completely > transpersonal. Hence, all tacit experience has some > element(s) that cannot be formulated as explicit knowledge > and transmitted. But, we can _appeal_ to (or rely upon) some > psychologically projected (imputed) commonalities between us. > For example, if I see you wearing shoes with laces, I can > assume that you know what it's like to tie your shoes. (No > smart-alack remarks about being dressed by one's mother!) > And in that sense, even if I can't write a formula for "tying > one's shoes", I can still _learn_ how to tie shoes. Further, > I can use the inaccurate > ("bad") formulas for how to tie one's shoes as a way to > actually learn how to tie shoes. Even further, I can _teach_ > others how to tie their shoes based on these "bad" models. > > Hence, we can use "bad" models to learn something that has no > "good" model. > > One might even go so far as to say _that's_ the very nature > of learning, not as you characterize it above. Note, > however, that this is pre-scientific. Science (and all > externalized, transpersonal methods) relies wholeheartedly on > making as much knowledge as explicit as possible. If one > went that far, it would be reasonable to assume that the more > autistic and rational of us learn best through the > development of compressed descriptions, whereas those of us > addicted to getting our hands dirty learn best through the > use of "bad" models and direct experience of applying those > "bad" models. There are lots of ways 'bad' models can be used. I think the one I was thinking of first is that they might be good for sort of 'reverse engineering' the thought process from which they came. I often look at a 'bad' model and ask what was the valid part of the question that led to it. What we then got to was an image of considering a set of bad models, built from various generating questions, and the complementary spaces they describe and shapes of realities they hint at beyond the models. I'm not sure how to use that in a direct analytical way with any real models of things at all. I just like it as an image for responding to the dilemma that all models are simplistic and we need to look how they misrepresent reality for other hints. Perhaps the simplest practical interpretation is how models need to change over time, why we need a process of examining the divergence between models and reality to change the models over and over to keep up. That leaves a trail of models as a record of either a thought process or a natural system process, or both. Best, Phil > > - -- > glen e. p. ropella, 971-219-3846, http://tempusdictum.com > Morality cannot exist one minute without freedom... Only a > free man can possibly be moral. Unless a good deed is > voluntary, it has no moral significance. -- Everett Martin > > -----BEGIN PGP SIGNATURE----- > Version: GnuPG v1.4.6 (GNU/Linux) > Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org > > iD8DBQFHWE24ZeB+vOTnLkoRAn7NAKCk56nz+lA1qPEG5yzySQnIqKutvwCgx90G > /bGBq9jNrFcyaX10b3HuTUs= > =yNb8 > -----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 glen ep ropella
Glen,
> > Phil Henshaw on 12/06/2007 10:53 AM: > > The hard part seems to be to take the first dark step to accepting > > there might be a shape of another form that the measures > are missing > > (like the whole tree or person). It means looking for how to best > > extend and complete your image based on the limited cast of the > > measures at hand. Interpolation gone wild?? Free form projection > > perhaps?? Sort of... You just gotta do something to make > sense of the > > larger continuities that develop in natural complex > systems. What I > > think we can see clearly is that our measures and models are highly > > incomplete. > > I think we agree, which normally means there's nothing to > talk about! [grin] But, I thought I'd throw out my term for > what you're describing: "triangulation". > > It's not really triangulation, of course. But it's certainly > more like triangulation than, say, population sampling. > Perhaps we could call it "tuple-angulation"??? [grin] PH: I guess I just call it filling in the gaps, understanding that as a combination of analysis and synthesis. So, if 'gaps' then become a raw material for systems science part of what makes a model 'good' is if you can see how it is also interestingly 'bad', since without having some interest in the 'bad' you can't be tracking the usually moving and significantly misrepresented targets of the physical system.. :-,) I do come close to 'triangulation' in my derivative reconstruction method, except I use 4 points to find a 5th rather than 2 points to find a 3rd. Given 5 points in time sequence it imputes a new value for the middle one, based on the making the implied 3rd derivatives from right and left the same (going forward and back in separate passes and averaging). If each point is considered a separate "bad" model for the system one could impute an average value and a system having a single fixed average state. Using derivative reconstruction imputes a continuous complex process without fixed definition instead. That seems to be a less distorting way of data smoothing, and more useful for raising questions about the turning points within the changing mechanisms producing it. > Here's a paper in which "we" (i.e. my outrageous rhetoric is > reigned in and made coherent by the authors of the paper ;-) > try to describe it: > http://www.biomedcentral.com/1752-0509/1/14/abstract >See Figure 1. This particular example is just one sub-type of the general method we're talking >about, here, though. PH: I was impressed with the clarity of the abstract and their not confusing biology, lab chemistry and computer model references. Figure 1 puzzles me though. I get your suggestion that this shows a way people are using new visualization techniques to compare models. I don't understand how highly complex comparisons of test tube and computer based things would make them look so very much alike unless both are parametric data displays of a sort not described, though. Comparing hugely complicated systems does need visualization help, certainly, but if that's what makes the images look so much alike it should be mentioned. Still, what I get from the picture is that they give themselves an A+. I don't see how their model recreates some features of the natural process and interestingly leaves others out. It's importantly that art of making what you've failed to account for interesting, rather than hiding it, that I find missing in lots of studies. So, here's to all 'bad' models...! may we survive them...:-) Best, Phil - -- glen e. p. ropella, 971-219-3846, http://tempusdictum.com If this were a dictatorship, it would be a heck of a lot easier, just so long as I'm the dictator. -- George W. Bush -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFHWX6LZeB+vOTnLkoRAhfiAJ4ldUf3p2wtlih3736TIp28uVtEZACfWyMf Pi/MX4iy1xD4PrqQNyNvbYo= =9GWs -----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 |
|
|
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1 Phil Henshaw on 12/07/2007 01:42 PM: > PH: I was impressed with the clarity of the abstract and their not > confusing biology, lab chemistry and computer model references. Figure > 1 puzzles me though. I get your suggestion that this shows a way people > are using new visualization techniques to compare models. I don't > understand how highly complex comparisons of test tube and computer > based things would make them look so very much alike unless both are > parametric data displays of a sort not described, though. Comparing > hugely complicated systems does need visualization help, certainly, but > if that's what makes the images look so much alike it should be > mentioned. Still, what I get from the picture is that they give > themselves an A+. I don't see how their model recreates some features > of the natural process and interestingly leaves others out. It's > importantly that art of making what you've failed to account for > interesting, rather than hiding it, that I find missing in lots of > studies. Just for clarity, it's a cartoon and not a visualization. The diagram is merely intended to give a visual impression of the iterative process being used. The gray smudges and spots representing targeted attributes do not map to particular behaviors of the in vitro or in silico models. So, it's not that they're giving themselves an A+, they're just trying to say that the first model (the gray circle in A) is falsified because it doesn't exhibit the behavior indicated by the spot labeled "a" even though it exhibits the behaviors labeled "t". The second model (not just the same model with different parameter values), pointed to by "2" in B is _also_ falsified because it does not exhibit "a". However, there are indications that model 2 is "better" than model 1 because it exhibits those two behaviors indicated by the spots that are closer in the behavior space to "a". The subsequent model 3 is _validated_ because it exhibits behaviors "t" and "a". > So, here's to all 'bad' models...! may we survive them...:-) Perfect! I'll make that toast over my next pint. - -- glen e. p. ropella, 971-219-3846, http://tempusdictum.com The only good is knowledge and the only evil is ignorance. -- Socrates -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFHWcv/ZeB+vOTnLkoRAsDcAJ97VJWKqW1O7XZjfvRqJccektNC3QCgn1fV TJh+giOWVLF9kvPtmpfVoi0= =sxCj -----END PGP SIGNATURE----- |
«
Return to Friam
|
1 view|%1 views
| Free forum by Nabble | Edit this page |