"Categories are kinds, and categorization occurs when the same output occurs with the same kind of input, rather than the exact same input. And a different output occurs with a different kind of input."
Aren't outputs kinds as well? It seems to me that the passage is suggesting a type-token relationship between inputs and outputs. It doesn't seem to make sense that a category would result in the same exact output action every time—the outputs a) might differ at least slightly for one kind and b) change over time.
The definition of categories given in this article holds up well when it comes to sensorimotor categories, but I am worried that delimiting categories based on output it is less intuitive for abstract concepts and categories that might have “fuzzy” borders.
“Learning occurs when a system samples inputs and generates outputs in response to them on the basis of trial and error, its performance guided by corrective feedback.”
I’m also curious about whether outputs have to be behavioral, or whether they can be mental states. In the instance that someone learns a category by observing what happens in the world (i.e. someone plays an instrument they haven’t seen before), and the learner has not interacted or produced any behavioral/causal output relevant to this new category, have they learned the category? I can also see that it may be reasonable to say that in order to learn a category, one must try out certain outputs and recieved feedback about them--otherwise, they are in imminent danger of being wrong about what the category is.
Initially, I was inclined to see that the nativist perspective comes from a reasonable place. Our senses are tuned to particular physical properties (RGB light, air compression-rarefaction frequency, etc), and separating the physical world into streams based on those tunings is a sort of categorization (even as simply as this is in the category of “sound” as opposed to “smell”) in of itself, and as such our feature driven categories are limited to being combinations of the definite set of sense categories, leaving a potentially infinite, but entirely predeterminable set of categories which we can experience.
I think that where this notion begins to run into difficulty is when we start to talk about the “hearsay” method of categorization, and categories that seem to be learned mostly through hearsay and supervision. “Primeness” does not really come from our senses at all, and though it will usually be learned through one specific symbol system, the category is not bound to that symbol system (we don’t have to use those symbols which invoke similar sensory responses to the ones which we learned about primeness using to re-invoke the notion of primeness or to categorize something as prime). In fact, it may be that we need no symbols at all to invoke “primeness” once it is learned. This then lacks a basis in those finite, implicit categorizations which occur in the specific A/D conversion mechanisms that our body uses, and therefore we don’t necessarily have to define categories in terms of that finite set, which makes for a much less (un?) determined set of categories which we can use.
In the 25th section Harnad mentions that “when I am sorting things as instances of roundness and non-roundness, I am sorting features of things… And features themselves are things too: roundness is a feature, an apple is not (although any thing… can also be a [feature]... of another thing).” I am curious about these features. I am tentative in believing that features are things. Things have properties (features) but properties do not really have properties… Roundness, redness, etc., are qualities that we find difficult to explain, yet somehow they make complete sense to us - they are just names for aspects that differentiate everything in our environments.
Otherwise, for something like beauty, where do these supposed tastes come from if they are not influenced by the “hearsay” of the people and society we are surrounded by? Are these things biologically encoded for some evolutionary advantage? Or is this something much less primal and much more cerebral? All I can say is that my definition for beauty is completely tied in with how I feel when I observe something. A beautiful landscape or actually a beautiful feeling - like peacefulness and centeredness - can bring an overwhelming amount of tearful joy. Sometimes, there are feelings we get, that we cannot name, less describe to another human being. This differs from the inability to describe redness because there is no name for thing we would like the other person to at least feel in their own way (like how if I say red, I know what it feels like for me to see red, the other person knows what it is like for them to see red, and while we cannot tell what the other feels, there is at least the capability to find a common connection that we have agreed upon - this thing is red.) Do we learn this “thing” that we feel/experience/etc then though? It seems an answer can be thought up for an innateness that we just put labels on. Perhaps I am confusing feeling with kinds, but the feeling component will always be there and I don’t think we can really ever separate it.
In the paper To Cognize is to Categorize: Cognition is Categorization, Harnad explains what categorization is and why it is learned instead of innate. I pretty much agree with what Harnad says in this paper. Unlike Universal Grammar, the categorization or absolute discrimination cannot be innate. It is either learned by trial and error or by being told through language by others. I also agree with the part of abstraction and selective neglect. It is hard to imagine how we can categorize things (put several things in one category) if we perceive each object as unique and different in every aspect from other objects because we cannot see any similar characteristics that enable us to group them together.
However, despite my agreement with most of this paper, my question arises at the same time: is cognition fully categorization? And can artificial intelligence categorize?
For the second question, Harnad talks about it a little in his paper when he compares unsupervised learning and supervised learning. Is it possible for artificial intelligence to learn categorization just like we do? Or Is it possible we tell artificial intelligence how to categorize or how things different from things by using the language they can understand? If we give it feedback on what is right and wrong, can it learn? In my personal opinion, I believe artificial intelligence has the ability to learn. It is possible for us to teach them which is right and which is wrong and gradually they generate their own rules and do some kind of categorization. However, I wonder if artificial intelligence can do everything about categorization that a human being is able to do. I wonder this because I think it is hard for computers to do selective neglect. I think for computers, everything is distinct and is stored in a single separate space and they seem to be lack of the ability to group things together naturally unless we tell them to do so.
For the first question, from Harnad's paper I think I can get why categorization is largely involved in cognition and that we use categorization in our lives all the time, but is cognition only and fully categorization? I can hardly see if from Harnad's paper. My personal opinion is still unclear on this question, but I do think that cognition is very complicated and it is hard to say it is only one thing; instead I think cognition is a mixture of several abilities.
I kind of agree with the points argued in this paper especially what was said in paragraph 22 learned categorical perception and the whorf hypothesis. "how objects look to us depends on how we sort and name them"
This really got to me because i realized that everything that we do in this life which mainly learning has to do with categorisation. Wether it be a relationship, problems with the self, the way we act in social settings, everything has to do with categorization. But most importantly it means that change, seeing new perspective, broadening our horizons, opening up our minds so to speak has to do with rearrangements in categorization. Wether it be creating a new category for a specific "thing" that was in another category, creating a new category or even linking different "things" to the same category. This shows the way we are constantly learning in life, from one novel experience to another, we are making new associations new connections, getting rid of old rigid constructs and seeing the world in a different manner.
As stated in this paper the way we categorize is through abstraction of certain priviliged features. The learning and reallocation talked about above, might be related to a change in the priviliged features to abstract. Meaning that if we start shifting our attention to different features and abstracting those instead of the old ones this enables us to form those new categorizations and therefore from this new view.
If categorization is doing the right thing with the right kind of thing, which in turn may be summarized by a theoretical, lengthy Boolean string, is it not still possible to categorize without having the feeling of understanding? This was illustrated with the modus ponens example brought up in class. Someone can accept a formal statement (say “all swans are white”), while still not feeling like they understand it’s implications (“here’s a swan, but I can’t say whether or not it is white”). In this sense, a T3 robot could play a part in a similar story, and accept all there is to accept about every content phrase, without actually “understanding” or “believing” any of it.
Of course Harnad isn’t saying that there can’t be categorization without consciousness, he’s saying that cognition is categorization, “…categorization is any systematic differential interaction between an autonomous, adaptive sensorimotor system and its world.” But does this mean that it is impossible to build a T3 robot that interacts without systematically differentiating objects? That is, an autonomous sensorimotor system that sees the world as a continuum, without any abstraction, but can still interacts as a human would. Of course, this robot would need an unlimited memory capacity. But it’s tempting to think that you could ask this T3 robot to bring up a picture of any moment in it’s lifetime, and have him(?) describe it to you in infinite detail – as in “Funes the Memorious,” but without the lack of function. It seems like Miller’s magic number is essential to our cognition, but if that number is increased - from 7+/-2 to something unimaginably high due to added computational power - would we lose our ability to cognize at a regular human level?
This paper opened up a new perspective for me in how i define categorization. In class we have said that it consists of doing the right thing with the right kind of thing but all this ever meant was feature searching and defining. How are the items alike? By which feature am I grouping them?
The idea of infinite possibilities of what a thing can be (i.e. bigger than a breadbox, not blue, not cerulean) then leads to items (features) that must not be included. Unlike Funes, we learn to ignore certain things so that we can make mental maps of where things should go and so categorization actually consists of infinite decisions that are quickly made.
"Everything in nature is a dynamical system, of course, but some things are not only dynamical systems, and categorization refers to a special kind of dynamical system. Sand also interacts "differentially" with wind: Blow it this way and it goes this way; blow it that way and it goes that way. But that is neither the right kind of systematicity nor the right kind of differentiality."
Why not? What is it about humans that isn't just blowing in the wind? This view of humans as having the right kind of systematicity/differentiality has to do with what harnad calls autonomous systems. In what way is a human an autonomous system, maintaining purely internal states. And how does the internal state of a grain of sand (its chemical structure, for example) differ in type from the internal state of a human being (our orientation towards things, our stomach acid etc.) If it were that a grain of sand does not categorize, or lacks the systematicity of a human, the grain of sand would immediately disassociate, all its chemical bonds broken. Even an inanimate object categorizes as its physical structure, in relation to the external world, has certain affordances from it, just as the human body does with its senses. The sand is sensing with its structure just as the human is sensing with its bodily structure albeit at a much higher order of complexity. While the grain of sand is much more susceptible to the whims of the external world than a human body, the human body too can be at the whim of the external world and is not by any means totally autonomous.Then, could it be said that what separates the human from the sand is not some essential attribute but instead many orders of complexity?
From my (limited) understanding of the paper, categorization can be described from many aspects:
1. Usability: categorization is to do the right thing with the right kind of things. 2. Actors: categorization is "any systematic differential interaction between an autonomous, adaptive sensorimotor system and its world" 3. Abstraction: "invariant under these sensorimotor transformations" 4. Learnable: Any content word in the dictionary is a category, learning a new word is learning a new category. 5. Discrete: There are no continuous categories, just discrete ones. You can, however, make a chunk out of a continuous action (by taking specific features from that continuous action).
I was surprised from the definition Prof. Harnad gave in class: to do the right thing with the right kind of things. I think it's because this definition stresses on the idea of usability: categories must be right and helpful.
The definition of categorization on Wiki is "the process in which ideas and objects are recognized, differentiated, and understood". What I don't understand is why categorization in cognitive science stresses on the idea of categories being right. What about random categories that we make up fictionally, are they not categorization?
"categorization is any systematic differential interaction between an autonomous, adaptive sensorimotor system and its world"
Categorization is doing the right thing with the right kinds of things. It is also tied into learning, where they may be learned by induction or instruction. The example given is that of mushrooms. You can either find out which mushrooms are poisonous through trial and (perhaps fatal) error or, someone who is an expert in the field may explain to you what features to look out for in order to know which mushrooms are poisonous. What cognitive scientist are interested in is how the human categorization system manages to respond differentially to different categories. The paper uses the example of Funes to illustrate that in order to categorize, we need to be able to be able to pick up on recurrences and ignore/forget anything that might detract from the recurring details. In order to categorize then, there seems to be a trade-off between rote memory and the sensory-motor/brain activity of abstracting recurring information.
"But the categorization problem is not determining what kinds of things there are, but how it is that sensorimotor systems like ourselves manage to detect those kinds that they can and do detect: how they manage to respond differentially to them."
I think this is the crucial aspect to categorization that concerns cognitive science. To what degree are categorize innate, learned or both. I think it's evident that me categories are innate and they have presumably become innate through evolution. Perhaps the ability to categorize though learning and language is a means of not burdening the brain and therefore making the process of categorization more efficient.
"what is the sensory shadow of categories like "goodness," "truth," or "beauty"?"
This is an interesting question for me. With a simple example like mushrooms, understanding how categorization works seems very straightforward and simple. How might "goodness", "truth" and "beauty" be grounded through our sensory motor aparatus? I think this ties into the discussion on 'hearsay' and how once language in in place, it form a new platform to make new categories. Perhaps these ephemeral words are grounding in a sensory motor feelings that certain experiences provide, rather than something that can be seen through feature detection.
In "To Cognize is to Categorize: Cognition is Categorization" Harnad shows what the title states. First he explains how sensorimotor contact affords certain interactions between organisms and the world. These affordances determine what an organism can do. This allows for categorization, which is " any systematic differential interaction between an autonomous, adaptive sensorimotor system and its world." Categorization is a special kind of dynamical system, changing in time. This means that different outputs could occur when there's the same input - the important thing is that the same output happens when the same kind of input happens. Because of its grounding in sensorimotor experiences, categorization is linked with learning. Harnad's emphasis in the paper is not in what kinds of categories there are (the ontic side of categorization) and instead on how categorization happens. Some categories are innate (like Universal Grammar) but most are learned. There is also supervised learning of categories, which is based in trial and error, as opposed to instrumental learning which is a primitive case of categorization based in conditioning (learning to give a certain response based on a stimulus).
I found there was a contradiction between the vanishing intersections argument, (where the intersection of all the sensory shadows will at some point be empty) and the stories of Funes and S, where "living in the world requires the capacity to detect recurrences, and that that in turn requires the capacity to forget or at least ignore what makes every instant infinitely unique, and hence incapable of exactly recurring." I feel like I side more with the Funes view of categorization, where endless commonalities could possibly exist between sensorimotor experience, from another perspective. This ties in perfectly with Watanabe's Ugly Duckling Theorem. Later though, this question is clarified when discussing ontology - many categories are acquired by hearsay and is very much contingent on culture.
Harnad strengthens the link between categorization and sensorimotor capacity in the discussion on discrimination. He also describes derivative sensorimotor grounding, based in language. "Language allows as to acquire new categories indirectly, through “hearsay,” without having to go through the time-consuming and risky process of direct trial-and- error learning." But, acquisition of new categories is done implicitly (figuring out how we do is the role of cognitive science!) Harnad concludes that cognition is categorization - all our categories are made up of how we behave differently towards different things.
"Categories are kinds, and categorization occurs when the same output occurs with the same kind of input, rather than the exact same input. And a different output occurs with a different kind of input."
ReplyDeleteAren't outputs kinds as well? It seems to me that the passage is suggesting a type-token relationship between inputs and outputs. It doesn't seem to make sense that a category would result in the same exact output action every time—the outputs a) might differ at least slightly for one kind and b) change over time.
The definition of categories given in this article holds up well when it comes to sensorimotor categories, but I am worried that delimiting categories based on output it is less intuitive for abstract concepts and categories that might have “fuzzy” borders.
“Learning occurs when a system samples inputs and generates outputs in response to them on the basis of trial and error, its performance guided by corrective feedback.”
I’m also curious about whether outputs have to be behavioral, or whether they can be mental states. In the instance that someone learns a category by observing what happens in the world (i.e. someone plays an instrument they haven’t seen before), and the learner has not interacted or produced any behavioral/causal output relevant to this new category, have they learned the category? I can also see that it may be reasonable to say that in order to learn a category, one must try out certain outputs and recieved feedback about them--otherwise, they are in imminent danger of being wrong about what the category is.
Initially, I was inclined to see that the nativist perspective comes from a reasonable place. Our senses are tuned to particular physical properties (RGB light, air compression-rarefaction frequency, etc), and separating the physical world into streams based on those tunings is a sort of categorization (even as simply as this is in the category of “sound” as opposed to “smell”) in of itself, and as such our feature driven categories are limited to being combinations of the definite set of sense categories, leaving a potentially infinite, but entirely predeterminable set of categories which we can experience.
ReplyDeleteI think that where this notion begins to run into difficulty is when we start to talk about the “hearsay” method of categorization, and categories that seem to be learned mostly through hearsay and supervision. “Primeness” does not really come from our senses at all, and though it will usually be learned through one specific symbol system, the category is not bound to that symbol system (we don’t have to use those symbols which invoke similar sensory responses to the ones which we learned about primeness using to re-invoke the notion of primeness or to categorize something as prime). In fact, it may be that we need no symbols at all to invoke “primeness” once it is learned. This then lacks a basis in those finite, implicit categorizations which occur in the specific A/D conversion mechanisms that our body uses, and therefore we don’t necessarily have to define categories in terms of that finite set, which makes for a much less (un?) determined set of categories which we can use.
In the 25th section Harnad mentions that “when I am sorting things as instances of roundness and non-roundness, I am sorting features of things… And features themselves are things too: roundness is a feature, an apple is not (although any thing… can also be a [feature]... of another thing).”
ReplyDeleteI am curious about these features. I am tentative in believing that features are things. Things have properties (features) but properties do not really have properties… Roundness, redness, etc., are qualities that we find difficult to explain, yet somehow they make complete sense to us - they are just names for aspects that differentiate everything in our environments.
Otherwise, for something like beauty, where do these supposed tastes come from if they are not influenced by the “hearsay” of the people and society we are surrounded by? Are these things biologically encoded for some evolutionary advantage? Or is this something much less primal and much more cerebral? All I can say is that my definition for beauty is completely tied in with how I feel when I observe something. A beautiful landscape or actually a beautiful feeling - like peacefulness and centeredness - can bring an overwhelming amount of tearful joy. Sometimes, there are feelings we get, that we cannot name, less describe to another human being. This differs from the inability to describe redness because there is no name for thing we would like the other person to at least feel in their own way (like how if I say red, I know what it feels like for me to see red, the other person knows what it is like for them to see red, and while we cannot tell what the other feels, there is at least the capability to find a common connection that we have agreed upon - this thing is red.) Do we learn this “thing” that we feel/experience/etc then though? It seems an answer can be thought up for an innateness that we just put labels on.
Perhaps I am confusing feeling with kinds, but the feeling component will always be there and I don’t think we can really ever separate it.
In the paper To Cognize is to Categorize: Cognition is Categorization, Harnad explains what categorization is and why it is learned instead of innate. I pretty much agree with what Harnad says in this paper. Unlike Universal Grammar, the categorization or absolute discrimination cannot be innate. It is either learned by trial and error or by being told through language by others. I also agree with the part of abstraction and selective neglect. It is hard to imagine how we can categorize things (put several things in one category) if we perceive each object as unique and different in every aspect from other objects because we cannot see any similar characteristics that enable us to group them together.
ReplyDeleteHowever, despite my agreement with most of this paper, my question arises at the same time: is cognition fully categorization? And can artificial intelligence categorize?
For the second question, Harnad talks about it a little in his paper when he compares unsupervised learning and supervised learning. Is it possible for artificial intelligence to learn categorization just like we do? Or Is it possible we tell artificial intelligence how to categorize or how things different from things by using the language they can understand? If we give it feedback on what is right and wrong, can it learn? In my personal opinion, I believe artificial intelligence has the ability to learn. It is possible for us to teach them which is right and which is wrong and gradually they generate their own rules and do some kind of categorization. However, I wonder if artificial intelligence can do everything about categorization that a human being is able to do. I wonder this because I think it is hard for computers to do selective neglect. I think for computers, everything is distinct and is stored in a single separate space and they seem to be lack of the ability to group things together naturally unless we tell them to do so.
For the first question, from Harnad's paper I think I can get why categorization is largely involved in cognition and that we use categorization in our lives all the time, but is cognition only and fully categorization? I can hardly see if from Harnad's paper. My personal opinion is still unclear on this question, but I do think that cognition is very complicated and it is hard to say it is only one thing; instead I think cognition is a mixture of several abilities.
I kind of agree with the points argued in this paper especially what was said in paragraph 22 learned categorical perception and the whorf hypothesis.
ReplyDelete"how objects look to us depends on how we sort and name them"
This really got to me because i realized that everything that we do in this life which mainly learning has to do with categorisation. Wether it be a relationship, problems with the self, the way we act in social settings, everything has to do with categorization.
But most importantly it means that change, seeing new perspective, broadening our horizons, opening up our minds so to speak has to do with rearrangements in categorization. Wether it be creating a new category for a specific "thing" that was in another category, creating a new category or even linking different "things" to the same category.
This shows the way we are constantly learning in life, from one novel experience to another, we are making new associations new connections, getting rid of old rigid constructs and seeing the world in a different manner.
As stated in this paper the way we categorize is through abstraction of certain priviliged features. The learning and reallocation talked about above, might be related to a change in the priviliged features to abstract. Meaning that if we start shifting our attention to different features and abstracting those instead of the old ones this enables us to form those new categorizations and therefore from this new view.
If categorization is doing the right thing with the right kind of thing, which in turn may be summarized by a theoretical, lengthy Boolean string, is it not still possible to categorize without having the feeling of understanding? This was illustrated with the modus ponens example brought up in class. Someone can accept a formal statement (say “all swans are white”), while still not feeling like they understand it’s implications (“here’s a swan, but I can’t say whether or not it is white”). In this sense, a T3 robot could play a part in a similar story, and accept all there is to accept about every content phrase, without actually “understanding” or “believing” any of it.
ReplyDeleteOf course Harnad isn’t saying that there can’t be categorization without consciousness, he’s saying that cognition is categorization, “…categorization is any systematic differential interaction between an autonomous, adaptive sensorimotor system and its world.” But does this mean that it is impossible to build a T3 robot that interacts without systematically differentiating objects? That is, an autonomous sensorimotor system that sees the world as a continuum, without any abstraction, but can still interacts as a human would. Of course, this robot would need an unlimited memory capacity. But it’s tempting to think that you could ask this T3 robot to bring up a picture of any moment in it’s lifetime, and have him(?) describe it to you in infinite detail – as in “Funes the Memorious,” but without the lack of function. It seems like Miller’s magic number is essential to our cognition, but if that number is increased - from 7+/-2 to something unimaginably high due to added computational power - would we lose our ability to cognize at a regular human level?
This paper opened up a new perspective for me in how i define categorization. In class we have said that it consists of doing the right thing with the right kind of thing but all this ever meant was feature searching and defining. How are the items alike? By which feature am I grouping them?
ReplyDeleteThe idea of infinite possibilities of what a thing can be (i.e. bigger than a breadbox, not blue, not cerulean) then leads to items (features) that must not be included. Unlike Funes, we learn to ignore certain things so that we can make mental maps of where things should go and so categorization actually consists of infinite decisions that are quickly made.
"Everything in nature is a dynamical system, of course, but some things are not only dynamical systems, and categorization refers to a special kind of dynamical system. Sand also interacts "differentially" with wind: Blow it this way and it goes this way; blow it that way and it goes that way. But that is neither the right kind of systematicity nor the right kind of differentiality."
ReplyDeleteWhy not? What is it about humans that isn't just blowing in the wind? This view of humans as having the right kind of systematicity/differentiality has to do with what harnad calls autonomous systems. In what way is a human an autonomous system, maintaining purely internal states. And how does the internal state of a grain of sand (its chemical structure, for example) differ in type from the internal state of a human being (our orientation towards things, our stomach acid etc.) If it were that a grain of sand does not categorize, or lacks the systematicity of a human, the grain of sand would immediately disassociate, all its chemical bonds broken. Even an inanimate object categorizes as its physical structure, in relation to the external world, has certain affordances from it, just as the human body does with its senses. The sand is sensing with its structure just as the human is sensing with its bodily structure albeit at a much higher order of complexity. While the grain of sand is much more susceptible to the whims of the external world than a human body, the human body too can be at the whim of the external world and is not by any means totally autonomous.Then, could it be said that what separates the human from the sand is not some essential attribute but instead many orders of complexity?
From my (limited) understanding of the paper, categorization can be described from many aspects:
ReplyDelete1. Usability: categorization is to do the right thing with the right kind of things.
2. Actors: categorization is "any systematic differential interaction between an autonomous, adaptive sensorimotor system and its world"
3. Abstraction: "invariant under these sensorimotor transformations"
4. Learnable: Any content word in the dictionary is a category, learning a new word is learning a new category.
5. Discrete: There are no continuous categories, just discrete ones. You can, however, make a chunk out of a continuous action (by taking specific features from that continuous action).
I was surprised from the definition Prof. Harnad gave in class: to do the right thing with the right kind of things. I think it's because this definition stresses on the idea of usability: categories must be right and helpful.
The definition of categorization on Wiki is "the process in which ideas and objects are recognized, differentiated, and understood". What I don't understand is why categorization in cognitive science stresses on the idea of categories being right. What about random categories that we make up fictionally, are they not categorization?
"categorization is any systematic differential interaction between an autonomous, adaptive sensorimotor system and its world"
ReplyDeleteCategorization is doing the right thing with the right kinds of things. It is also tied into learning, where they may be learned by induction or instruction. The example given is that of mushrooms. You can either find out which mushrooms are poisonous through trial and (perhaps fatal) error or, someone who is an expert in the field may explain to you what features to look out for in order to know which mushrooms are poisonous. What cognitive scientist are interested in is how the human categorization system manages to respond differentially to different categories. The paper uses the example of Funes to illustrate that in order to categorize, we need to be able to be able to pick up on recurrences and ignore/forget anything that might detract from the recurring details. In order to categorize then, there seems to be a trade-off between rote memory and the sensory-motor/brain activity of abstracting recurring information.
"But the categorization problem is not determining what kinds of things there are, but how it is that sensorimotor systems like ourselves manage to detect those kinds that they can and do detect: how they manage to respond differentially to them."
I think this is the crucial aspect to categorization that concerns cognitive science. To what degree are categorize innate, learned or both. I think it's evident that me categories are innate and they have presumably become innate through evolution. Perhaps the ability to categorize though learning and language is a means of not burdening the brain and therefore making the process of categorization more efficient.
"what is the sensory shadow of categories like "goodness," "truth," or "beauty"?"
This is an interesting question for me. With a simple example like mushrooms, understanding how categorization works seems very straightforward and simple. How might "goodness", "truth" and "beauty" be grounded through our sensory motor aparatus? I think this ties into the discussion on 'hearsay' and how once language in in place, it form a new platform to make new categories. Perhaps these ephemeral words are grounding in a sensory motor feelings that certain experiences provide, rather than something that can be seen through feature detection.
In "To Cognize is to Categorize: Cognition is Categorization" Harnad shows what the title states. First he explains how sensorimotor contact affords certain interactions between organisms and the world. These affordances determine what an organism can do. This allows for categorization, which is " any systematic differential interaction between an autonomous, adaptive sensorimotor system and its world." Categorization is a special kind of dynamical system, changing in time. This means that different outputs could occur when there's the same input - the important thing is that the same output happens when the same kind of input happens. Because of its grounding in sensorimotor experiences, categorization is linked with learning. Harnad's emphasis in the paper is not in what kinds of categories there are (the ontic side of categorization) and instead on how categorization happens. Some categories are innate (like Universal Grammar) but most are learned. There is also supervised learning of categories, which is based in trial and error, as opposed to instrumental learning which is a primitive case of categorization based in conditioning (learning to give a certain response based on a stimulus).
ReplyDeleteI found there was a contradiction between the vanishing intersections argument, (where the intersection of all the sensory shadows will at some point be empty) and the stories of Funes and S, where "living in the world requires the capacity to detect recurrences, and that that in turn requires the capacity to forget or at least ignore what makes every instant infinitely unique, and hence incapable of exactly recurring." I feel like I side more with the Funes view of categorization, where endless commonalities could possibly exist between sensorimotor experience, from another perspective. This ties in perfectly with Watanabe's Ugly Duckling Theorem. Later though, this question is clarified when discussing ontology - many categories are acquired by hearsay and is very much contingent on culture.
Harnad strengthens the link between categorization and sensorimotor capacity in the discussion on discrimination. He also describes derivative sensorimotor grounding, based in language. "Language allows as to acquire new categories indirectly, through “hearsay,” without having to go through the time-consuming and risky process of direct trial-and- error learning." But, acquisition of new categories is done implicitly (figuring out how we do is the role of cognitive science!) Harnad concludes that cognition is categorization - all our categories are made up of how we behave differently towards different things.