Dynamics Part II | Philosophy of Cognitive Science | Dr. Josh Redstone

hi everyone and welcome back to philosophy of cognitive science with me dr josh redstone in today's somewhat overdue lecture we will be wrapping up our discussion of dynamics so this is dynamics part two and uh well let's start with a little bit of recap so last time when we covered part one of chapter seven from andy clark's mindware we looked at dynamic systems theory predominantly uh whoops we looked at dynamic systems theory for the most part and a little bit um a little bit of dynamic field theory of course we didn't get into any of the mathematics because those are well beyond the scope of the class and beyond my training but we got a sense of how dynamic systems theory may help us to explain the mind at least in terms of low level perception and action and problem solving related to perception and action we also learned a little bit about the type of complexity that dynamic systems theory is suited for explaining or modeling this kind of complexity has certain features right we talked about four key features low level um you know any kind of complexity where uh powerful but low level descriptions of systemic unfolding any kind of complexity like that lends itself to a dynamic analysis for example and here we're talking about things like void flocking or convection rolls in frying pans of course intuitive geometric images of the state space of the system we saw this a little bit last time when we talked about attractors and repellers and bifurcations within the state space indeed the state space itself or if you like the dynamic field is one of those geometric um one of those geometric images right uh kind of like if you're familiar with physics and you want to talk about uh albert einstein's work right relativity gravity all of that cool stuff you can talk about it just in terms of math or you can envision space time as a sheet with gravity wells in it and so on and so forth i'm sure you're familiar with what i mean we can do something similar here in dynamic systems theory where we try to use intuitive geometric uh visualizations to capture the complexity that we're trying to describe um the kind of complexity where if we use dynamic analysis we can isolate control parameters define collective variables that kind of complexity is suited to the dynamic analysis we saw this last time as well particularly with the walk governor uh walking as well uh the swinging pendulums on the wall and finally the use of the notion of coupling um this is especially pertinent when it comes to the walk governor because as we saw the the arm angle and the throttle valve the extent to which the throttle valve is open or closed uh those are coupled in the case of the watt governor similarly with the synchronizing uh self-synchronizing pendulums on the wall that synchronize due to vibrations running along the wall we can analyze that as a coupled system so that's those are the sorts of features that uh the kinds of complexity that we can understand with dynamic systems theory uh has and just to go over um or rather the kinds of complexity that we can study using the tools dynamic systems theory i should say and just to go over um uh just to recap a little bit more remember we saw all of this kind of play out in different cases like rhythmic finger motion uh i briefly mentioned treadmill stepping the watt governor of course preservative reaching and the a not b error so last time we kind of uh saw all of those features of complexity at various points in our discussion of of those four examples where researchers have actually brought the tools of dynamic systems theory to bear on low level uh perception action problem solving right today what we're going to do is briefly take up some of the concerns with dynamics that are raised in the later half of chapter seven um and of course uh fair warning uh the slides are a little bit less dense today so i might be riffing a little bit more than usual be sure to take notes if you find that helpful alright so let's get started with um the hidden premise that is um that is point of discussion a in clarks chapter seven let's take a look now so if you remember from the start of last week's lecture we kicked things off with a little bit of discussion uh concerning this uh inner symbol flight what clark calls inner symbol flight as we move from the classical traditional symbolic picture of minds and mentation to connectionism to dynamics we lose the emphasis or the importance that's placed on neat discrete inner items mental representations right this kind of rears its head again at this point in the chapter where clark introduces something called the radical embodied cognition thesis just before i explain what this is explain uh better explain embodied cognition and embody cognition i've kind of sort of gestured at before mentioned in passing a lot of what we've been talking about lately would count as embodied cognition cognition that does not just involve some kind of symbolic control system but that you know takes advantage of the musculoskeletal properties of the body and the environmental features in order to problem solve so think of uh you know when we talked about robots briefly think about asimo versus a passive dynamic walker right a passive dynamic walker takes advantage of its bodily form and the environment in order to save energy and it's also representation sparse right so that's embodied cognition when um not just the mind but the body and the environment are taken into account when it comes to uh problem solving intelligence understanding these things and so forth so the radical embodied cognition thesis is this and you can find this on page 151 of clark structured and symbolic representational and computational views of cognition are mistaken embodied cognition is best suited uh or sorry embodied cognition is best studied using non-computational and non-representational ideas and explanatory schemes and especially the tools of dynamic systems theory okay that's the radical embodied cognition thesis and many dynamicists seem to support an idea if not this exact idea at least something close to it but clark thinks it's surprising that we find such sweeping conclusions given the state of the art of the research after all low-level sensory motor engagements are suited for explanation by dynamic systems theory we've seen this and i think this chapter very nicely lays this out along with the previous chapter so and heck even the chapter before that actually so chapters 5 6 and 7 all highlight this idea that low level so we're talking rudimentary representation sparse kind of direct sense act cycles the kinds of things we're talking about when we're talking about passive walkers or robot crickets or an infant uh treadmill stepping you know it's it's its leg acts like a spring and it just has to make contact with the treadmill in the right way for walking behavior to emerge these are suited to explanation with dynamic systems theory of course but the standard framework that is the traditional framework symbolic representations rules mental representations or nice discrete inner items that clark thinks may still be the best for understanding higher level cognition and we'll see this kind of unfold further as we discuss all of these different points of discussion even at low levels of analysis clark maintains some aspects of the traditional approach can still provide rewarding analysis right so even if we can explain something non-representationally non-computationally using the tools of dynamics maybe we don't have to do that necessarily we don't have to forget that it's also possible to explain such low-level interactions symbolically and it remains an empirical question whether such low-level systems in nature are realized non-symbolically or not right so none nonetheless a lot of the thinkers that we've covered in this chapter like felon and smith or kelso are sympathetic to the radical thesis but what do we need to do with all of this i mean we're not really going to try to adjudicate between the radical thesis and the standard view but clark does think that there are some things that we need to keep in mind as we make our way into this new and interesting area in the cognitive sciences what clark thinks needs to happen here is that we need to do a better job of connecting the dots uh between the empirical work and the radical conclusions of the radical thesis so if you're a supporter of the radical thesis you know there's a hidden premise in there or rather actually that's not a good way to put it the hidden premise is really the radical thesis kind of inserted between uh the observations that dynamicists have made uh and um you know this idea that uh we can capture all of this i suppose non-symbolically and non-computationally um in any case we need to connect the dots between the conclusions of the radical thesis and the empirical work that has been done the best way to do this clark proposes might be to use an idea that we've already encountered and that idea is the continuity thesis the idea that there is a continuity between life and between mind um you know perhaps perhaps the way we look at this uh is um perhaps the way that the traditional view guides the way we look at mentation and cognition and the mental is wrong even though the idea of computation representation so on and so forth even though those ideas may not actually be wrong i'll try to clarify this with an analogy that you can find in this chapter from uh pollock in a 1994 publication and he makes the analogy with flight so think back in the day when humans were trying to achieve flight they would look at birds in nature flying you know and they see that they flap their wings um insects too and bats for that matter right the salient feature of flight seemed to be wing flapping for you know early um early pioneers of flight you know we're talking pre wright brothers here but this isn't what's really important for flying at least not when it comes to artificial flight like with an airplane or a glider what's more important is the aileron principle the aileron principle is this idea that we need to use a control system of some kind to raise and lower wing flaps on the airplane not to flap like a bird but just so that the pilot can control the aircraft as it is flying through the air or gliding through the air now according to pollock the flapping is like symbolic cognition it's an obvious piece of the puzzle but it's kind of like the last piece of the puzzle what's more important for birds and airplanes are probably a lot of representation sparse passive dynamics interactions between bodily form and the environment these are what are more important for flight and these similar kinds of dynamics may be more important for cognition too at least explaining how it works but flapping is still important and symbolic cognition may still be important but it's kind of like later on that continuity spectrum right so uh the saliency of symbols and symbolic cognition uh obscures cognition um it obscures what's actually what's going on right cognition is a control system that doesn't you know work like some kind of homunculus in someone's head or some kind of um character in a in a giant robot controlling a giant robot it's really uh more like a system that governs complex real-time interactions uh within the system itself and between the system and the world a bit like a watt governor right so anyway um symbol systems may still be useful here they just might not be the whole story the dynamic approach might illustrate the fundamentals of the story higher level cognition perhaps is still best explained using the tools of the traditional approach and that brings us back to the idea of strong and weak continuity and this is point of discussion b so let's take a look at point of discussion b now the radical thesis is rooted in an observation that we've already made in this class and that is that higher level cognitive systems are probably built upon older more evolutionarily basic cognitive systems now both connectionists and dynamicists make this point but connectionists do it without calling into question the ideas of computation and representation at least they do question them they question the traditional notions of computation and representation but they don't deny their existence they just say that well computation and representation are different than we thought they were but dynamicists kind of question whether we even need these notions in the first place because we're not dealing with sense think act cycles they deal more with direct sense act cycles kind of like the new roboticists do of course in doing so dynamicists are emphasizing uh interaction reciprocal causation coupling all of this stuff you know things like ongoing uh couplings between the um the environment uh the the body the mind different perceptual and cognitive systems in the mind and all of these influence one another and are influenced by each other right but none of this uh even if all of this is true actually establishes the radical thesis according to clark you know and to see why you can just consider a far less radical claim of solving problems using you know something like an offline model an offline stand-in um for uh for environmental stimuli right if cognition and perception and problem-solving all boils down to these mind body world couplings that are influencing and being influenced all at the same time then how do we explain offline reasoning reasoning when i'm not um directly engaged with the environment you know perhaps a dynamicist can explain how you know um an expert tetris player plays tetris in an embodied sort of way or how how a clerk at the grocery store packs a grocery bag uh in the most uh you know efficient kind of way using some kind of embodied approach but what if you were doing a problem like that offline um you know a great example uh i i just finished watching that netflix series the queen's gambit uh you know offline reasoning is kind of like when she plays chess in her mind right and chess grand masters really can do this by the way they can use their fusiform face area to recognize chess patterns and have offline chess games chess grand masters can play entire games with one another without a chessboard okay very cool stuff how do we explain that if cognition is all about these dynamic interactions how do we reason offline well um we just take those systems that we use when we are engaged with the environment and we just use them offline we use them like a model and that would seem to me and i think to clark as well to count as a mental representation even if it's built upon this more basic um much more dynamic system right so um you know this is continuous in a certain way continuous and discontinuous in other ways it's representational in some ways and non-representational in some ways it depends how you slice it up and really a lot of these are open questions right how best ought we tackle or how oft we tackle the question of offline reasoning it seems like we need some kind of representations are these representations just um you know using these systems these cognitive and perceptual systems that we normally couple with the body and the world are we just using them offline and is that how we get our mental representations uh who knows these are open questions good stuff to think about and good stuff to perhaps consider for a critical response or an essay if you like very briefly to go over the next point of discussion which is representation and computation again um clark points out that what dynamicists are skeptical of or what they tend to be skeptical of where representations are concerned are what you might call objectivist representations or viewer independent finely detailed models of the world you know kind of um you know the snapshot connection of reality the sort of picture of the world in your head those kinds of finely detailed models but clark points out that internal representations don't have to be of this kind they could be of a kind that's geared towards um you know supporting the kinds of real-world real-time actions that dynamicists study so again there might be representations they just might not look like representations look on a sort of traditional or classical uh picture so much of the skepticism in dynamics at least where dynamics studies the mind is aimed at objectivist representations you know the kind of semantically transparent representations we talked about uh back when we were looking at physical symbol systems but these are not the only kinds of representations there are right we've already encountered lots of different kinds of representations from the sub-symbolic distributed superpositional representations of um connectionism and perhaps there are other uh kinds of representations that are quite different from the objectivist semantically transparent sort that we find in the classical uh symbolic paradigm but which nonetheless support the kind of dynamic interactions that dynamicists are interested in studying right so um i'm not going to go into much more detail here but um that is just something i thought that i should leave you with before we move on to the next point which is the space of reasons so here clark points out that one of the biggest problems with anti-representationalism you know the kind of anti-representationalism we find in dynamics or in the new robotics um is the treatment of the brain as just another factor in this kind of uh larger larger causal web you know that's part of the body and the world of the environment as well in one sense this is true to be sure but in another sense it's clearly false humans with our brains and indeed other thinking creatures in the world non-human animals and so forth are not like swinging pendulums or convection roles in important ways and the ways in which we're different from these kinds of phenomena that we can model with the tools of dynamic systems theory seem to all have something to do with brains as a kind of control system so what do we do what do we do here so how do we do justice to this distinction between uh physical causal systems which dynamicists are interested in and knowledge-based systems which um fit more within the traditional picture within the cognitive scientists or within the cognitive sciences well some dynamicists um would deny that there is any such difference at all um but if we do this there's a worry that will fail to do justice to many of the different kinds of behaviors that we as thinking agents express you know for example uh and this is my example this is not clark's if we if we just think of everything as a physical causal system that would uh you know we could still explain something like goal oriented behavior like the behavior of gray walters tortoises right but what about the selection of a particular goal over another if you're a tortoise uh even the selection between the goal of moving toward a light source and going towards your charging station is quite uh simple and we could explain that with a dynamic approach but how would i explain my decision to record this lecture versus practice the guitar versus going to the park to take some pictures of the birds right that seems less amenable to a dynamic explanation if you're one of those dynamicists that think that there is no real difference between physical causal systems and knowledge based systems maybe a solution here clark says is dynamic computationalism so here details of the flow of information would be just as important as the dynamics and that flow could perhaps be understood with the tools of the computational approach that is very much an open question for clark all right let's turn to the next point of discussion richer dynamics for preservative reaching all right let's bring it back to preservative reaching and the a not b error if you remember from last time uh esther thalin and colleagues created a dynamic field model of this error that the infants make you know they reach for something several times that's been placed in location a the experimenter moves it into location b in full view of the infant the infant sees this yet reaches for location a even though the infant knows the toy is in location b that's the a naught b error that's the name of the error the phenomenon is called preservative reaching now spencer and colleagues in a 2009 publication report that schuner and colleagues in an earlier publication attempted to implement this dynamic field model of preservative reaching in a robot but something interesting happened the behavior of the robot was actually different than what we observe when we um test uh human infants in these preservative reaching experiments the robot didn't mimic the a naught b error like human infants do instead the robot just kind of jiggled back and forth unable to decide where to reach whether for the toy's original location or for its new location so what was going on here well neither of the reaching signals clark thinks was strong enough within the dynamic field um to win out so we needed a way of boosting the signal turning up the gain on these signals so they enriched the dynamic model that they were trying to implement into this robot by raising the resting level of the dynamic field during the response phase kind of turning up the volume of the field or turning up the gain on the field during the response phase of the task when the robot has to reach for either location a or b this way even a very small increase in the a or b response would cause the appropriate activation level to cross the threshold and initiate the right reaching response this is kind of like turning up attention in the same way you might turn up the volume on your music player right that's what we mean by gain here we're just making the signal stronger by turning up the gain now um if we think about this in terms of the human brain um we're talking about something that sounds very much like attention and perhaps in the human brain this is realized or implemented by the selective um increasing or decreasing of the gain of certain populations of neurons in the prefrontal cortex and that could be how attention is mediated to generate this or that response in human inhuman infants now uh mediating attention in this way uh anything to do with the prefrontal cortex in fact is fairly high level cognitive representation heavy stuff so clark is saying that maybe we can have the best both uh best of both worlds here clark seems to be optimistic that um there is still room or rather there is room for both the dynamic and the more traditional notions of representation and computation here of course i am passing over a lot of the details uh that clark discusses at the end of this point of discussion and that's because i want to move on to the final point nonetheless i think clark is correct and i think throughout this whole chapter if you haven't already noticed by now clark does a very good job of a very good job of taking the strengths of both approaches of focusing in where different approaches towards studying the mind complement one another rather than contradict one another this is certainly one of those uh cases and indeed he does this quite successfully successfully throughout the entire book um in any case if you have any of the any questions about some of the material that i've kind of glossed over let me know but now we're going to come to the final point of discussion which is cognitive incrementalism and whether this is the biggest issue concerning the material that we've discussed in this chapter and the uh previous chapters so let's take a look at that now all right so what is cognitive incrementalism all about well basically cognitive incrementalism concerns the relationship between the low level strategies for perception and action and higher level more cognitive strategies what is the connection between the kinds of problem solving the dynamicists study the low-level stuff versus the kind of things that more traditional cognitive scientists study like the more cognitive stuff well um the the the view right now in embodied cognition circles and this is where we come to this question of cognitive incrementalism is that the human mind something complicated like the human mind uh is built upon an evolutionarily older um set of basic embodied embedded strategies basically evolution has added these uh bells and whistles so that we get more and more complicated uh complicated cognitive systems uh daniel dennett refers to this as a sort of ratcheting effect right where we uh successively build upon more fundamental more fundamental mental components if you like um so there's a principle of continuity here like we saw earlier this issue of incrementalism raises the idea of continuity again and of course um you know some dynamicists at probably all dynamicists or a great many of them don't think that there is any real difference between these low-level strategies and high-level cognition the key term is uh no difference what kind of difference is there no difference at all or is there some difference that perhaps dynamicists are overlooking um because it is true uh that not all cognitive systems are like this it's true that even outside of embodied cognition there is a general consensus that the more complicated um high level reasoning problem-solving abilities that humans have are built upon older evolutionarily speaking and more fundamental cognitive and perceptual machinery but there are cognitive processes that don't seem to depend on these kinds of basic tweaks to our sensory motor capabilities they seem to involve novel and independent processing systems and one of these which we've already briefly talked about in this class is the dual visual system you can take a look at an interesting a bit of discussion about the dual visual system in box 7.3 in clark if you like and i'm not going to go into that uh box in detail and i'm not even going to go into the dual visual system hypothesis in great detail um other than to provide a brief recap of what it is now uh in the literature you usually see this as attributed to milner and goodale of course they were building on the earlier work of unger leader and mishkin and basically the consensus on the visual pathway in the human brain nowadays is that once information reaches the occipital cortex it can take one of two routes depending on what the information is being used for one route goes to the parietal cortex and one route goes to the temporal cortex uh if we're talking about conscious visual experience uh or offline visual reasoning which is like using your imagination verbally reporting what you are seeing you know i see this lamp i see this coffee cup i see this chair over here that i'm resting my laptop on that you cannot see um we are using the ventral visual processing stream that is uh we are using what uh unger leader and mishkin and milner and goodale have called the what pathway however online visual motor reasoning depends upon the door dorsal visual swing stream the wear pathway this is not consciously accessible it's what we use as i discussed in an earlier video to coordinate our actions when reaching out to pick up objects or open doors right it seems to you perhaps that if you want to reach out and pick up a coffee cup or open a door let's go with a coffee cup since i have one here that you see the coffee cup you consciously identify it and you consciously direct your arm and your hand to grasp the cup and that what you are consciously doing is affecting your action or causing your action but that's not quite true um the uh actual reaching for the cup is handled by the dorsal stream which is um operates online but is not accessible to conscious awareness and we know this from studies of blind sight so if you are curious about learning more about that i will provide a link in the video description to an interesting video on blindsight um it's an interesting phenomenon where uh there is some damage to the um to the eventual visual stream perhaps in one hemisphere such that you're blind in one eye you have no phenomenal visual experience but you still do better beyond chance of reaching for the correct object that an experimenter tells you to reach for say a coffee cup versus a pencil in the visual side of your visual field where you do not have any phenomenal uh visual uh experience going on so uh these people are seeing without seeing uh very strange these people who suffer from blind sight so that's an interesting piece of evidence for this lots of other evidence but blindside is the most interesting one i'll link you to a cool video that you can watch about blind sight but here excuse me it seems that what has happened is that nature has not in fact added uh successive bells and whistles onto um onto a basic sensory motor capacity nature has added a functionally novel uh and functionally independent visual pathway the ventral pathway the dorsal pathway is probably older um but the ventral pathway is functionally novel it does something different than what the dorsal pathway does and it's functionally independent um it is not subserved the ventral pathway is not subserved by the dorsal pathway it can operate independently of the pathway we see that kind of dissociation in blind site where it doesn't operate but the dorsal pathway works fine and there are probably other cases where the dorsal pathway uh is perhaps damaged uh so that you cannot reach out and um you know grasp the correct object but you could still look at it and visually identify it right so because of this double dissociation we know that these are functionally uh dissociated they're independent um and that they the the ventral pathway is functionally novel it does something different it's it's it's a system that we have in addition to the the other pathway that's not built on top of that pathway now why is clark bringing all of this up well because it raises the issue of cognitive incrementalism namely an issue concerning cognitive intermental incrementalism is it the right way to look at everything that's going on in the mind um probably not in all cases we don't know because as clark is arguing here there is a lot of empirical uncertainty as to how uh how much uh cognitive incrementalism can do for us it clearly uh clearly there's cognitive incrementalism going on in some cognitive systems but probably not in all of them like the dual visual system so perhaps we can still make sense of this question uh by doing further empirical work um i mean uh this is all still very muddy water um it's possible that's inputs offline or offline features can be repurposed by another system so um you know there's a kind of perceptual game of tag going on even though these systems operate independently from one another the dual visual systems there is a sort of game of tag that goes on between them when you do reach out for a coffee cup or reach out to open a door and we have other capabilities that are clearly novel that may or may not be built upon other systems like our linguistic faculties there's a question of cognitive incrementalism there um and there's a question there's a question uh of the extent to which those faculties allow us to repurpose yet other systems so the uh the question of cognitive incrementalism i guess what i'm trying to say in a very roundabout way is not just empirical it certainly is empirical but it's um perhaps also also conceptual uh it depends on how we uh conceive of not only what particular neural systems are what counts as a neural system but what it's doing for other systems and how it's doing it so this is an open question and i think that i'm going to leave it there for today so let's wrap it up i actually have a few announcements before we wrap it up today so let's cover those before i say goodbye all right so if you've already taken a look at the slides before watching this lecture then you know that i have planned to change things up for the remainder of the term uh for my next lecture which i expect i will have delivered either tomorrow or saturday it's not going to be a lecture on any of the material what i've actually decided to do is do a lecture on uh how to approach your essay assignment for this class because i realized i did not include that in the syllabus uh in the course outline and i probably should have so what i'm going to do is cut a chapter from the clark textbook and the next lecture will technically be this week's lecture but there will only be one lecture and it will all be about your essay how to choose a topic how to approach writing it what the grading scheme will be so on and so forth i will also be getting your second critical responses done over the weekend and i will assign your third response uh of course uh i'm doing i'm placing so much emphasis on the essay because i'm changing up a few other things um i was initially have uh planning to have you write um a final exam that would be a little bit like the midterm i've decided not to do that um i've decided rather that the essay will count as your final exam so the final exam will be dropped from the syllabus the final take-home and the essay assignment will replace that so it'll be like a final take-home essay um the chapter that i will be dropping is chapter 11 so we're not going to read chapter 11. we will still read the conclusion uh but i'm dropping a chapter and i'm having i'm simplifying the final exam by having you just do the essay as your final because you've been working on critical responses and i feel like you're probably all more suited to producing an essay and i think you probably all get a little bit more out of it since there's so much in this book to cover i don't know how useful it would be just to test you on a a very broad uh you know selection of the material i think you'll get more out of um doing your own kind of deep dive in an essay on a question or a topic that has stood out to you as you've read this book so that's how we will do that of course i will re-weight uh other assignments and i will share all these details in an updated syllabus which you will have for monday i may also drop one of the critical responses you know you were initially supposed to write five of them i think i'll probably just have you do four and then if anybody wants to uh if they did you know say badly on one of their responses they could still submit a fifth one as a kind of makeup critical response to try and boost the grades a little bit so um i want to hear from you guys do these changes sound good to you if they do let me know but this is how i'm planning on doing things so um hopefully this is just going to make everyone's lives easier i know it will make my life easier and i'm fairly certain that all of you are probably as busy or possibly more busy than i am so i imagine this will make your lives easier as well but uh that's the plan that's what we'll do um so just to reiterate uh the next lecture is on the essay assignment which will replace your final exam we're going to drop uh chapter 11. um after the essay lecture we'll get back onto our uh regular lecture lecture schedule with chapter eight um you'll have your grades for critical response due soon and i am about to assign critical response number three for those of you who have been wondering about that all right that's it for now again i hope you guys are all doing well i hope everyone's taking care of themselves and taking care of each other um if you have any questions get in touch let me know otherwise i will see you next time for my lecture on your essay assignment which is also now your your take home final okay all right take care everybody and i'll see you next time bye for now

Channel: Mind & Metal

Views: 79

Rating: 5 out of 5

Keywords: #mindandmetal, #cognitivescience, #dynamics, #philosophy

Id: OSqdkXaTUy4

Length: 44min 9sec (2649 seconds)

Published: Thu Nov 12 2020