CatComCon2015: 6a. Harnad, S. (2005) To Cognize is to Categorize: Cognition is Categorization

Saturday 11 January 2014

6a. Harnad, S. (2005) To Cognize is to Categorize: Cognition is Categorization

Harnad, S. (2005) To Cognize is to Categorize: Cognition is Categorization, in Lefebvre, C. and Cohen, H., Eds. Handbook of Categorization. Elsevier.

We organisms are sensorimotor systems. The things in the world come in contact with our sensory surfaces, and we interact with them based on what that sensorimotor contact “affords”. All of our categories consist in ways we behave differently toward different kinds of things -- things we do or don’t eat, mate-with, or flee-from, or the things that we describe, through our language, as prime numbers, affordances, absolute discriminables, or truths. That is all that cognition is for, and about.

55 comments:

Unknown11 February 2015 at 19:07
After reading this article, the main claim is that cognition is categorization and that categorization is “any systematic differential interaction between an autonomous, adaptive sensorimotor system and its world” (2). In other words (and to mimic what was covered in class), categorization is doing the right things to the right kinds of things. It was also claimed in class that continuous motor skills, such as playing tennis, and imitation are not categorization. The trouble I have with this statement is that continuous motor skills and imitation are aspects of information processing, and therefore they are aspects of cognition. We are perceptually aware of the movements we make when doing these actions and are thinking and understanding these movements. So to say that cognition is categorization but that continuous motor movements and imitation are not is a little unclear to me.
The other thing that came to mind when reading this article is colour agnosia. This is a condition in which there is damage in the occipito-temporal region of the brain that causes patients to be unable to recognize colours. They have full perceptual ability to see colours but are not able to categorize them. It seems to me that this supports the view that cognition is categorization because they are not able to cognize the colour information and are therefore not able to do the right thing to the right kinds of things; if asked which colour is green they would not be able to recognize it. They are therefore showing a categorical problem because they are not able to process the information from the colour.
In the paper, Harnad argues “if we did not have colour categorical perception then the continuum from red to violet would look very much like shades of grey”(29). I am having trouble understanding how colour agnosia comes into place. They do not have colour categorical perception, but they still are able to see the full spectrum and colours do not look like shades of grey. Just like the Whorf hypothesis, cultures that do not have words for certain colours does not mean they are not able to perceive different colours, it is simply a naming difference to which they have broader categories than other languages with more colour terms. This does not mean that they do not see them, simply that they categorize them differently. I guess I just don’t understand how language can have such an effect on the classification of items to different categories if categorization is suppose to be innate as Fodor argues. I find it much more plausible that categorization is a learned process. Otherwise if categories are suppose to be innate, how does he explain how some people categorize blue and green together, but that others do not?
ReplyDelete
Replies
Reginald Oey12 February 2015 at 17:00
During the last seminar, I was having trouble grasping Dr. Harnad’s that doing something continuous (such as swimming) is not categorization. However, this article has cleared up some of my confusion. Specifically, the explanation that “categorization occurs when the same output occurs with the same kind of input” has helped me comprehend the differences between categorization and dynamical systems. From what I understand, a dynamical system will act the same way with the same input, but categorization (as an autonomous system that changes over time) will not produce the same output with the exact same input.

Thus, if I am swimming I am doing the exact same thing when the exact same thing happens to me, but will react differently to a new stimuli. For example, when I come to a wall, I will turn and push off every time. But when I reach a different wall, my reaction will be different than when I turned at the first wall; my reaction will be different because this is a different input even if it was the same kind of input as the first wall.

My question is, can we still categorize and label these actions? For example, if I see someone swimming on their back rotating one arm after the other, I can say that they are swimming backstroke. But if I see someone swimming on their front rotating one arm after the other, I can say that they are swimming freestyle. Am I categorizing that person’s actions? I understand that their actions are not exactly the same all the time (for example, the person may be swimming slower if they are more tired) and their actions might not look the same as if I swim backstroke or freestyle. But, would I still be able to extract the common features (e.g. swimming on one’s back in backstroke) and categorize their actions as a kind of backstroke (versus not a kind of backstroke)? Or am I simply distinguishing the two actions based on their relative differences (freestyle is done more on one’s front and backstroke is done more on one’s back)?
ReplyDelete
Replies
Unknown13 February 2015 at 11:08
“Categorization, it seems, is a sensorimotor skill, though most of the weight is on the sensory part (and the output is usually categorical, i.e., discrete, rather than continuous); and, like all skills, it must be learned.”

Like some of the comments above have mentioned, this really tied in with the discussion about how swimming is not categorization, since it is not continuous. I am still having difficulty understanding this. Can’t we categorize each movement in swimming, and, thus, the swimming motion is just a series of categorized motions that make a bigger category (like how apples are red, oranges are orange, but apples and oranges are in the ‘fruit’ category)?

Swimming combines pull, kick, and body position. The pull of the stroke in freestyle can even be further broken down – the ‘catch’ (the ‘catching’ of water when your arm first enters the water, but is still extended in front of you), the pull (from the catch to point of exit), and the recovery (when the arm travels through the air to return to the forward position). If swimming can be broken down like this, are these aspects not discrete?
ReplyDelete
Replies
Unknown13 February 2015 at 13:39
On the question of: Is all cognition categorization?

The way I understood it from class is that there are (in a sense) two streams of cognition: 1)Categorizing: doing the right things to the right kinds of things
2)Embodying (this is the word I will be using for this idea): doing the right things with one’s body to the right kinds of things [without the need of categorizing]
Therefore, even if embodying isn’t a part of categorizing, it is still a form of cognition (how we do what we do).

My first question is: Does categorization preceed embodiment?

Examples:
Imitation
There is some sort of natural process in which allows for one to do what the other is doing. However doesn’t this need a base from which to work from?
Ex: A baby imitates a person sticking out their tongue
Is this instance of imitation not working through noticing the innate category of facial expressions and replicating it? Therefore embodying (imitation) occurs only after having categorized that something (in this case the facial expression).

Riding a bicycle
You can’t teach someone how to ride a bicycle through words; it is implicit learning and needs some sort of motor practice. In the process of learning how to ride a bike, and riding a bike (even if we are unable to explain it), can we not say that once we have achieved the motion and motor skills of riding we have achieved it through a categorization of learning how to move our muscles and becoming in-tune with our balance? I have achieved riding the bicycle through categorizing the way I needed to move the muscles in my legs, feet arms, and hands, and the way I needed to move my body from side to side in order to achieve balance. So even if it became implicit in the process, it is still something I needed to learn through categorization of my movements. Therefore embodying (riding the bicycle) occurs only after having categorized that something (my body’s motions).

Funes the Memorious
“Every successive instant of his experience was stored forever.” Professor Harnad explains how if Funes as a person actually existed he would have been unable to speak at all, and instead he would just be an experiencing mush.
Is this a proof that categorizing must come before embodying? Without grasping categories there would be no sense of being able to do with the body (whether it be walking, eating talking, or anything!)?
Or would it be possible to argue that these two streams of cognition are independent of each other? Ex: That he would still be able to suck on a bottle of water if fed [(because this is an innate reflex)-Question: would innate reflexes even be considered as embodiment?]. Therefore this later example showing embodiment without the need to categorize.

My second question is: If cognition does occur prior to embodiment, does this not have implications in the way we understand cognition? (example: (I mean this in the metaphorical sense) instead of having different streams of cognition, we have layers of cognition)
ReplyDelete
Replies
Unknown13 February 2015 at 13:57
The gist of the paper, which is explicit in the title, really resonates with my own attempt at understanding cognition. I figured I’d add my 2 cents or ask questions where discrepancies emerged.

3. Categorization

“Categorization is accordingly not about exactly the same output occurring whenever there is exactly the same input. 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. So that's where the "differential" comes from.” (Harnad)

So, in a sense, a range of different inputs that have something in common along some dimension or dimensions would give rise to the same output. This seems like categorical perception applied to an extremely large number of dimensions. Does that make sense? I’m essentially positing that objects, like cars and trucks for example, have categorical compression within the category and expansion between the categories, but instead of the output being “that’s blue” or “that’s green”, it’s “that’s a car” or “that’s a truck”.

For this to occur, wouldn’t a word (the output when presented with the referent) be activated by types of stimuli that consistently are present (the mostly invariant properties/features) when encountering referents to the word? Do we need to posit something beyond Hebbian Learning to account for this connection? Is the Poverty of Stimulus argument what’s standing in the way?

4. Learning.

“UG-compliance is the underlying invariant in question, and, according to Chomsky, our capacity to detect and generate UG-compliant strings of words is shaped neither by learning nor by evolution; it is instead somehow inherent in the structure of our brains as a matter of structural inevitability, directly from the Big Bang.”

This passage confuses me. Maybe someone could clear it up for me. I understand the distinction between the innate programmed by evolution and the acquired programmed through experience/culture. Is Chomsky positing that UG is a necessary mechanism of any and every structure in the universe that has language, as if more innate than the innateness we refer to when speaking of our evolved biology? I always thought Chomsky meant hard-wired in human brains, with evolution doing all the design work. Does he mean evolution couldn’t have designed us without something like a UG device, or that the UG isn’t even designed?

(Continued in reply)
ReplyDelete
Replies
Unknown13 February 2015 at 14:51
Regarding categorization and non-categorization: although the difference seemed clear to me in the previous lecture, I'm now a bit confused on the distinction.

The article states :
Categorization is accordingly not about exactly the same output occurring whenever there is exactly the same input. 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. So that's where the "differential" comes from.

But, the wind blowing the sand is a dynamical system, and not categorization. Is the wind not blowing the sand in the same kind of way? To me this seems like an example of the same kind of input, leading to the same output. With reference to what was said in the linked youtube video (Steven Harnad: Categorization (07/13)) about apples, that all instances of apples have some common features, I don't see why this isn't true for wind blowing sand. All apples are not identical, and not all wind blowing sand is identical. I don't see the difference, as to why the former may be categorized, and the latter may not.

Or if we look at the example taken up last class, of an individual swimming. It seems reasonable to categorize, and distinguish between swimming and running, or activities in water, and out of water. Or, as another student brought up, different kinds of swimming (back stroke versus breast stroke). Once again, the same kind of input (the swimming action) leads to the same output (movement in the water, or physical exertion etc..).

If we apply the definition from the video, and from the last class, of categorization as doing what needs to be done, with the right thing, nothing seems to preclude swimming from being categorization.

What am I missing here? If people can be categorized (according to the lecture), who are clearly dynamic, why can activities/actions, or events not be categorized?
ReplyDelete
Replies
Dia (Deborah) Sztein Baremberg14 February 2015 at 07:18
"Look for the 'invariance' shared by all the sensory shadows of just about any of things things designated by those words, you will find that there is none: their 'intersection' is empty."

The process of schema-making is referred to as “abstraction” in this paper, but I prefer using the word “schema” to refer to the end result of this process of abstraction because it gives a clear idea of what our mental “representation” (if something so inaccessible to the conscious mind can be called that) is. A common example is “writing.” What is the definition of writing? We have pencil-and-paper note-takers, computer typists, and skywriters all participating in this activity, but is there an invariance shared by the sensory shadows of all these actions? Not really, but nor does there need to be – when Junior English Learner grows up in our world and hears “writing” being described, not only to refer to every time daddy creates a shopping list, fills out a form, or teaches him how to spell C-A-T, the child’s representation of “writing” abstracts from “that thing daddy was doing the particular time that mommy said ‘daddy is writing’ to something like “writing is someone handling an implement that leaves a trace on a surface.” All content words, from simple animal words to complex concepts such as justice, are schemas like this – a sort of configuration with open slots that can be filled in with stuff. As is noted in the paper, features can be differentially weighted; though the schema for writing is quite rigid (someone scratching a dry pen to a surface might be attempting to write, but is not writing), the schema for “game” or “beauty” has a lot more leeway and are a lot harder to put into words. Constructing a schema is what allows us to chunk so effectively – we can abstract away a lot of detail about our daily routine by saying “I had breakfast and then walked to school”, each action being recalled as an ‘average’ of what usually happens when we undertake such activities. Whether we walk ten steps or eleven to the first turn, the experience is nevertheless chunked as an ordinary walk to school. If we are mugged on the way to school, our schema no longer applies and the event is recalled separately.
ReplyDelete
Replies
Unknown14 February 2015 at 09:02
“So if we accept that all categorization, great and small, depends on
selectively abstracting some features and ignoring others, then all
categories are abstract”

I am slightly ambivalent with this sentence. At first read, I entirely agreed with Harnard’s statement that categorization is abstraction. A dog, for instance, could be an animal, or a mammal, or a friend, or even food. Same thing with mushrooms, which could belong in the categories of “things that grow in a forest” or “food” or “poison”.

Categorization is just a fleeting label that we impose on the things (whether concrete or abstract) in our world. We live with all these concrete/abstract things, and categorization is the process that enables us to navigate. I would assume that the “kind of categorization” one imposes would depend on the adaptability of the situation. Again, as an example, if one is foraging for food, a horse can enter in the category of “edible meat”. Yet, if one needs transportation, the same horse would then fall in the category of “mode of transportation”.

However, the statement that “all categories are abstract” made me re-think when it come to animals, insects, and plant. Since animals also know what to eat, what to mate or what to flea away from, does that mean that animals are capable of abstraction too?

(Of course, I guess this depends on whether we are discussing about supervised or unsupervised categorization. When Harnard claimed that categorization is abstraction, was he referring to supervised learning or unsupervised learning (or perhaps even both)? Perhaps the “kind” of categorization that animals do have nothing to do with abstraction)
ReplyDelete
Replies
Jessica Magonet14 February 2015 at 14:19
In class, we learned that categorization is doing the right thing with the right kind of thing. Harnad argues that “most of our categories are learned.” Sure, we might be born with the ability to categorize a few things (ex: we are probably born with the ability to tell the difference between red and blue), but even those “innate” categories are the result of adaptation (read: natural selection – those humans who could tell the difference between colours at birth passed on their genes because this was a useful skill for survival).

Most categories, however, must be learned, either through supervised learning (ex: kids learning to tell the difference between fruits and vegetables by their parents teaching them at the supermarket and correcting them when they misidentify something) or unsupervised learning (ex: you probably learn to distinguish the figure from the ground on your own). In both cases, we need to be exposed to a lot of examples of things and have the opportunity to try (and sometimes fail) at doing the right thing with the right kind of thing.

Harnad illuminates the power of language in allowing us to learn how to categorize indirectly when he writes: ”Language allows us to acquire new categories indirectly, through ‘hearsay,’ without having to go through the time-consuming and risky process of direct trial-and-error learning. Someone who already knows can just tell me the features of an X that will allow me to recognize it as an X.”

Take the example of mushroom identification. You can learn to tell the difference between safe and dangerous mushrooms through trial and (potentially deadly) error, or you could have someone with a lot of mushroom identification experience tell you the features to look for to tell if a mushroom is safe (ex: safe mushrooms are brown, prickly and spotted). The first case is an example of trial and error learning (less safe, more time consuming) whereas the second case is an example of learning indirectly through language (much safer and quicker!).

I think the most striking point in the paper is this: Harnad points out that “[i]t can’t be hearsay all the way down, though, I will have to have had to learn some ground-level things the hard, sensorimotor way.” Here is an example to help illustrate what Harnad means: I can learn to tell the difference between safe and dangerous mushrooms indirectly through language (someone tells me that the safe mushrooms are brown, prickly and spotted), but maybe I have to learn the difference between soft and prickly things, or spotted and non-spotted things, the sensorimotor way (ex: I am given different surfaces as a kid and taught through trial and error what is soft and what is prickly).

I believe this claim parallels Harnad’s symbol-grounding problem: just as we need to be sensorimotor systems to learn categories, so must we be sensorimotor systems for words to have meaning for us. In both cases, it can’t be language all the way down.

Harnad defines categorization at the start of his paper as a specific kind of interaction between a sensorimotor system and its world, but just in case we had any doubts, the claim that “it can’t be language all the way down” means that computers cannot categorize. Whatever a computer is doing when it sorts numbers into prime and non-prime categories, it isn’t categorization. And therefore, claim Harnad, it can’t be cognition.
ReplyDelete
Replies
Unknown14 February 2015 at 17:57
“The evidence for this is that people who are perfectly capable of sorting and naming things correctly usually cannot tell you how they do it. They may try to tell you what features and rules they are using, but as often as not their explanation is incomplete, or even just plain wrong. This is what makes cognitive science a science; for if we could all make it explicit, merely by introspecting, how it is that we are able to do all that we can do, then our introspection would have done all of cognitive science’s work for it.”

Interestingly, this inability to verbalize what we know is still a major problem across other scientific domains. Many students learn through stages to develop skills that we are unable to verbalize in the classroom. One example is bedside manner. Over time, doctors (hopefully) develop the ability to read situations and know when it is best to speak up or to stay quiet and allow a patient to process new information. They need to learn how to take a patient’s body and verbal signals, and integrate them to decide how to act. This is a very important skill, but even doctors who are revered for their bedside manner cannot provide formulaic instructions about how to act. An interesting possibility is that just because a skill is “unteachable” does not mean that the skill necessarily needs to be learned through sensorimotor learning—perhaps we have simply not developed an effective way to categorize the different aspects of the skill. These limitations are another argument in support of cognitive science research, for if a model to explain how we can do all we can do is developed, then all of these inneffable skills will be come much more accessible through proper categorization.
ReplyDelete
Replies
Unknown15 February 2015 at 14:02
This week’s question: whether cognition is categorization; on Harnad’s paper seems to follow very logically but I am going to try to break it down just in case this is a false sense of overconfidence.

The conclusion is: “all of our categories consist in ways we behave differently toward different kinds of things, whether it be things we do or don’t eat, mate with, or flee from, or the things that we describe, through our language, as prime numbers, affordances, absolute discriminables, or thrust. And isn’t that all that cognition is for – and about?”

Here I don’t think Harnad is implying that this (cognize by categorizing) is HOW we do it (or is he?). As he showed in his paper even people who are perfectly capable of categorizing may not be able to tell you how they do it (point 27) and this is why introspection has lost the battle to cognitive science. Furthermore the ability to categorize has some innate aspects and some learned aspects. Learned categorization capacity comes in two flavors: supervised and unsupervised. Supervised learning done through “error corrective feedback” (and in our case verbally, because we are compulsively verbal) allows us to speed the learning process through abstraction and hearsay. Unsupervised learning can be more time consuming, it uses trial-and-error learning, and I get from the paper that this type of learning is closely tied to our innate sensorimotor capacities trying to find the “winning features.” We are sensorimotor systems and our relationship to everything that surrounds is determined by the motor interactions that occur with sensory input from everything that surrounds us and their affordances. To categorize we need to be able to abstract and ignore some features (otherwise it would be impossible for us to categorize, everything would be unique) while also being able to weight more heavily other features of similarity (that is to identify recurrence).
I found very interesting the distinction (also made in the video) between categorization which is “any systematic differential interaction between an autonomous, adaptive, sensorimotor system and its world” and categorical perception which is “how well we can identify and how well we can discriminate between and within categories,” the case being that compression occurs within categories and expansion occurs between categories. I will try to stablish this distinction with an example: suppose I have a glass (a standard glass destined to hold drinking liquids) and a soup plate (a standard soup plate destined to hold liquid food) in the case of mere categorization, I would drink directly from the glass but I would use a spoon to eat the liquid in the soup plate (forget about cultural differences). In the case of categorical perception this would simply define my own ability to distinguish that even though both objects are designed to hold liquids, the differences in length, diameter and height of the objects suppose different affordances, such that I will not pour soup into a water glass. I can compress water glass and soup plate into the category ‘crockery’, but I expand the differential if the categories are ‘iced-tea holder’ and ‘tomato soup holders’.
ReplyDelete
Replies
Unknown16 February 2015 at 15:39
“Miller pointed out that if the differences are all along only one sensory dimension, such as size, then the number of JNDs [just-noticeable-differences] we can discriminate is very large, and the size of the JND is very small, and depends on the dimension in question. In contrast, the number of regions along the dimension for which we can categorize the object in isolation is approximately seven”

The difference between discrimination and categorization interests me. Discrimination involves comparison and is the ability to distinguish things from each other. Categorization is a form of absolute discrimination, or the ability to abstract characteristic features that distinguish a stand-alone thing. Harnad discusses the role of memory in each of these functions. Discrimination is facilitated by memory. A larger memory will allow a person to identify more unique aspects of an item. For example, the mental picture of two sunsets will be indistinguishable in one’s memory unless one can remember the exact gradient of colors that characterized each. Alternatively, a larger memory will hinder one’s categorization ability. We see this in the case of “Funes the Memorious” and can also apply it to the sunset example. If you remember every detail of the sunset, the specific gradient of colors and hue, you will be less able to abstract the invariant features shared by the two sunsets and categorize both mental pictures as sunsets. Memory clearly plays different roles in these two tasks. If cognition is categorization, our average human memories make our cognitive abilities possible. That said, would we consider Funes as being incapable of cognition? Additionally, would a computer need to be designed with memory limits that are on par to a human memory in order for it to have cognitive abilities?
ReplyDelete
Replies
ab17 February 2015 at 05:09

"Although basic color CP is inborn rather than a result of learning, it still meets our definition of categorization because the real-time trial-and-error process that “shaped” CP through error-corrective feedback from adaptive consequences was Darwinian evolution. Those of our ancestors who could make rapid, accurate distinctions based on color out-survived and out-reproduced those who could not. That natural selection served as the “error-correcting” feedback on the genetic trial-and-error variation. There are probably more lessons to be learned from the analogy between categories acquired through learning and through evolution as well as from the specific features of the mechanism underlying color CP -- but this brings us back to the “how” question raised earlier, to which we promised to return. […] In the case of ambiguous figures such as Escher drawings there may be more than one way to do this, but in general, there is a default way to do it that works, and our visual systems usually manage to find it quickly and reliably for most scenes. It is unlikely that they learned to do this on the basis of having had error-corrective feedback resulting from sensorimotor interactions with samples of the endless possible combinations of scenes and their shadows.

I am not so clear on what exactly differs between the case of black/white and figure/ground. Why is it that only the first is hypothesized to be the result of “error-correcting” during "natural selection"?
ReplyDelete
Replies
Unknown17 February 2015 at 08:49
There must be innate categories. Instinctive behaviors that guide living beings on how they are supposed to interact with food for instance are a good example of that. And there probably must be at least one first category we use to build the subsequent categories.
Plus, if there were no innate categories, how could we make sure that people have similar enough categories in their mind? Different individuals do not all have exactly the same experience even with the same object. How can we be sure that they retain the same features from one object, that they weigh those features similarly, and consequently that they have a similar enough abstract representations of the world’s objects? There must some universality in categories, at least for language, because when communicating, two persons in a conversation when using the same word, assume they are referring to the same object.
Also, could we say that mirror neurons actually help forming similar categories in different individuals because the neural net will show the same activation pattern with respect to the same object in both the doer and the observer? So both will actually be doing the same kind of thing with the same kind of object.
ReplyDelete
Replies
Unknown17 February 2015 at 08:51
This article proved very interesting to read however, it raised a few concerns. Harnad states that 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 am a bit confused by this. I would assume that in order to categorize something, you would need to know what it is. I believe I understand the distinction between "what" and "how", but what I do not understand is how you can achieve "how" without knowing "what". In other words, in order to understand how we have categorized something (or how we have cognized it) don’t we have to know what we have categorized (or cognized)?
My next question has to do with the explanation of categorical perception. In particular, I’m having trouble understanding the argument regarding color. Harnad argues, in the absence of colour CP, we would only see shades of grey when looking at the continuum of red to violet. So would we be unable to perceive or unable to categorize? What I mean is, how do we know what we would see if we did not have CP? Moreover, without categorical perception, would we really only see shades of grey? But isn't grey in itself another form of categorical perception?

ReplyDelete
Replies
Unknown17 February 2015 at 13:09
In section 16 - Abstraction and Hearsay - Harnad discusses the 5 sensorimotor ways we can interact differentially with things: “We can see them, recognize them, manipulate them, name them or describe them.” He then describes later that “recognizing is special, because it is not just a passive sensory event. when we recognize something, we see it as a kind of thing (or an individual) that we have seen before.

I found this interesting, because in the familiarity sense of the work “recognize”, I feel that recognizing in itself is categorizing. When you recognize something you are categorizing it as being familiar, and this is one of the most important categorizations, at least evolutionarily, and also one of the most innate. To be able to recognize something/someone, or even certain characteristics or elements of something/someone, is of vital importance to the survival of all animals, who are wired to fear the unknown and feel more comfortable with the familiar.

Moreover, when you “recognize” something as being familiar you are categorizing it as being familiar, so in this case the words “recognize” and “categorize” could even be considered synonymous. Therefore, in this situation, I have a hard time separating “recognize” from “categorize”, and an even harder time thinking of “recognize” as being comparable to “see”, “manipulate”, “name”, and “describe”; I feel as though it should be on a list of its own.
ReplyDelete
Replies
tom17 February 2015 at 14:22
“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” (Harnad, 2005).
Using a computational approach, Harnad explains how learning occurs, stating that when we do something, we receive feedback, and based on that feedback, our performance should change if we are indeed learning. More specifically, if our performance improves because we are generating more correct outputs and fewer errors, we are learning. Harnad then explains behaviorism’s theory of learning, which is similar to the one suggested in the aforementioned quote, although it is not computational in nature. Also, cognitive science has yet to discover what neural mechanisms are taking place during the learning process; all we have so far are brain images showing activity in certain areas, which don’t explain much. It seems then that this quote is explaining that our learning is really dependent on our interpretation of the input, or at least our ability to accurately assess or categorize the input. This idea seems to suggest that we as humans are always dependent on our environment to guide our behaviour, but should there not be a feature of our cognition that would discriminate between proper inputs and generate the correct outputs? What about in the case where we produce erroneous outputs to certain inputs? Is this an error in properly assessing the input or in generating the proper output? I think that until we have a better ontogenetic understanding of the brain and cognition as a faculty of mind, it will be difficult to fully understand categorization’s role in cognition.
ReplyDelete
Replies
Unknown17 February 2015 at 14:35
(Vegetative consciousness?? The reality, the physical, the “sensory”. What about the stuff we can’t explain?) I would like to think that in cognitive science we are better than forgetting or simply ignoring things that haven’t been explained. Openness in cognitive science is something I’m looking for not narrowing. For example, pigeons have tetra colour vision. We THINK we understand how this works, but we cannot experience it. Especially colour blind people (wow that sucks!). Let’s move to discuss the idea of Universal Grammar, which is becoming a less and less universal idea in linguistic psychology. (http://aeon.co/magazine/culture/there-is-no-language-instinct/) In more recent years (2013~?) even Chomsky himself has begun to reformulate his previous ideas about how a relatively small set of sounds are able to give us an audible image of the mind. Not wanting to rewatch the entire UQAM special from October 2013, I will simply reference this talk (https://www.youtube.com/watch?v=Rgd8BnZ2-iw). If universal grammar is not so universal, than I find it easily follows that categories are not something that is inherent to cognitive selves. I do not believe that there is no place for categorization in certain techniques of learning about cognition with our goal of robot-Turing test in mind. Nevertheless, I struggle to understand the “whys/why nots” about categorization as a strong idea in cognitive science. The problem I present is vegetative consciousness; what is there to be understood in terms of a categorization perspective for a person that is simply a sensorimotor system has extremely reduced capacities in some ways but completely functional or semi-functional cognitive ability?
ReplyDelete
Replies
Unknown17 February 2015 at 15:42
"The kinds of things there in the world are, if you like, the sum total of the world’s potential affordances to sensorimotor systems like ourselves. 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'd like to ask a follow-up question regarding this quote. Does the ability to detect and categorize kinds of things (using sensorimotor systems) necessarily entail the ability to differentially respond to them? In other words, is categorizing a sufficient condition for systematic differential interaction? To answer this, we probably have to define "categorization" first. Would one define an instance of successful categorization as the moment from which the sensorimotor system detects a thing in the world to the moment it decides which systematic differential action to take, or would categorization only be successful if the subsequent interaction is successful? Clearly, one doesn't have to be able to kick a ball in order to categorize (one of) the ball's affordance as "being kicked". It follows that categorization only concerns the mental/cognitive process aspect of an output, not the motor capabilities. Bear this in mind when we move on to the issue of motor theory. If sensory input plus correct categorization is insufficient to generate successful motor output, then what about the motor theory? Would it be true that successful motor output is necessary for our sensorimotor systems to make correct categorization decisions based on sensory input? Seems unlikely. One doesn't have to be able to kick a ball to know that it can be kicked, and one doesn't have to be able to produce the in-between consonant sounds to know that they can be continuously categorized.
ReplyDelete
Replies
Unknown17 February 2015 at 18:22
“Funes’s infinite rote memory was hence a handicap, not an advantage. He was unable to forget -- yet selective forgetting, or at least selective ignoring, is what is required in order to recognize and name things.” (12)
To a certain extent, categorization is functional and efficient. At the same time, attention to detail is necessary but too much information is inefficient. Sensorimotor processing takes times.
So how do we attune our ability to categorize? How do we discern between 'kinds' that require broad or narrow categories? Does each 'kind' of thing require a new process of trial and error, and categorical learning?
ReplyDelete
Replies
Unknown17 February 2015 at 18:30
This article made me think a lot about how categorization occurs. Is it at the perceptual level or at a higher level? If we were to accept that categorization is cognition then it would be at the higher level. In this case, our sensory inputs would encode everything on the continuous spectrum and send it to higher structures to be separated into categories. But in the other case, if categorization is just part of perception then the sensory neurons are "generalized" in a sense--they respond to a range of stimulus. Based on the paper, it seems like categorization is based on cognition, that something other than the bare sensory inputs divides different 'kinds' of stimulus. But could it be that we have both? Innate categorical perception appears to be a result of hard-wired categories in the sensory neurons. And the learned categorical perceptions would be dependent on cognition. Is it possible that some categorical perception is cognition but others are not?
ReplyDelete
Replies
Daria Khadir17 February 2015 at 18:37
This comment has been removed by the author.
ReplyDelete
Replies
Daria Khadir17 February 2015 at 18:40
The article claims that categorization is what cognition is about. Categorization is our capacity to break down the world in categories and classify its objects in different kinds. This feature is crucial in evolutionary means as it allows us to react effectively to the events of our surrounding world. As was shown with Borges’ character, its infallible memory did not allow him to assimilate things based on shared features, therefore each new occurrence is for him something totally unknown that he has to rediscover. The impossibility for this character to forget details means that he is not able to selectively retain features that would be more meaningful to him and classify objects based on those features. To imagine oneself in that position is tiring and terrifying, as it makes one incapable of reacting to things since everything in always new and unknown.
Categorization, however, requires some level of memory and learning as it is adaptive. Through experience (with or without error-corrective feedback) we first create basic categories based on morphological and geometrical features. These categories are then developed and multiplied according to what features are relevant to us for an adapted interaction with the world. For example, when detecting danger represented by animals, we give more importance to shape than to size differences. This capacity relies on our sensory motor system - and this is why robotics is crucial for reverse engineering of the brain and human – and its detection of stimuli (sensors) as well as its detection of constancy (invariance-detectors). Although some categorical perception is innate (colour perception), most of it is acquired through evolution and learning. The case of colour is actually an interesting example where innate categorization can then be sophisticated by learning, for example when communities invent new colour words in order to discriminate things in their environment based on the different features that suggested by these colour variations and the differential response they require.
ReplyDelete
Replies
Unknown17 February 2015 at 18:46
I do think that that there are problems with the “classical theory” of categorization. True, the boomerang will remain recognizable and its perceived size will remain the same as I move around it. But before I have identified it as a boomerang I will need to move it around until it is presented to me in some kind of canonical presentation. Until I have achieved this presentation, uncertainty will remain as to the nature of the object.

Also, the exact same object projecting the exact same shadow will be perceived as two different objects depending on the context. The very same piece of wood is cognized as a broomstick in the janitor’s hands whereas it is cognized as a weapon in a ninja’s hands. We can try and patch up the theory by saying that the context is first categorized, then the elements inside are categorized accordingly. But on what basis is the context cognized if not on the basis of the elements inside?

--

“Where does this leave goodness, truth and beauty, and their sensorimotor invariants? [...] these categories are acquired largely by hearsay.”

Goodness, truth and beauty are paradigms of ineffability. All of philosophy is dedicated to trying to come to terms with these concepts so how can we say that they were learned through language? These are not like chicken-sexing, there is no one story as to what is good/bad, true/false, or beautiful/ugly. Yet we all embody an implict story as to which is which.

--

“Cognitive Science is Not Ontology…” (#29)
I agree that scientists should not spend time debating over the ultimate nature of things, or if they want to do so, they should switch their science hat for a philosophy hat. That being said, scientists should at least know that they are working according to a presupposed ontic story (physical realism). This is particularly true of cognitive science, since its subject matter includes the entire field of experience. Knowing that they are assuming an ontic state of affairs, they should be open to the possibility that some persisting or seemingly intractable problems may be predetermined by this assumption. Perhaps stepping outside of this assumption is to give up on science altogether, but I wouldn’t give up so fast.
ReplyDelete
Replies
Unknown17 February 2015 at 18:50
“What the stories of Funes and S show is that 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 am often weary of being average in any sense – grades, looks, intelligence etc. – because I grew up in a generation where every one of us was a special flower that could become anything it wanted to be, and so being average was a waste of potential. After reading this article, I am happy to be average, at least when it comes to categorization.

We at once admire those with incredible memory capacities and those who can see patterns in massive amounts of data. The opposite of Funes’ problem would be a super-chunker (pardon the atrocious word) – somebody who forms categories and finds patterns where the rest of us see randomness. That too would be an unpleasant life. If the goal at the extreme end of this hypothetical condition is to minimize things into categories, and then categorize those categories, then further chunk them and so on, then I where would that leave us? Maybe social interactions would lose their nuances so that the main intention for the utterance is distilled to one motive. Maybe the exponential chunking ability would come at the cost of missing details that were not weighted heavily enough at the time of chunking, but that ultimately lead to miscalculated categories. In any case, I am glad my memory and my categorization abilities aren’t special at all.
ReplyDelete
Replies
Unknown17 February 2015 at 18:52
I've been thinking quite a bit about the statement that "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."
Most of this article focused on our capacity to recognize, abstract, and categorize, because these are affirmative actions, but are we also to believe that the opposite - not recognizing, or forgetting - is also a type of ability that we possess? Or is it an inability, something that we may involve to improve? If we currently have the ability to categorize and memorize 7 +/- 2 objects, was that number lower several centuries / millennia ago, and will it perhaps increase in the future? Obviously we don't want to end up like S or Funes, but it seems to me that we wouldn't be worse off if we could augment our capacity to categorize and shed some of our inability to forget. I read once in a book on the history of linguistics, and the Sapir-Whorf hypothesis in particular, of the theory that the ancient Greeks could discriminate fewer colors than humans nowadays can, as supposedly evidenced by Homer's famous use of the adjective "wine-dark sea," among other examples. I recognize that this theory doesn't have quite as much proof as one would like, but if we are to believe it for a moment, and accept that humans seem to be evolving towards being able to distinguish more and more categories, then will this trend continue, and to what extent? Are we heading in the direction of Funes, and if so, will our ability to 'forget' evolve simultaneously, in order to hold us back from experiencing every moment to infinite detail in a debilitating manner?
ReplyDelete
Replies
Unknown17 February 2015 at 19:16

In the video, Harnand said that categorization is to do what needs to be done with that kind of thing.

I understand that categories are based upon sensory-motor inputs that a person gets from their surrounding stimuli. Harnand also mentioned Gibson's statement on categorization-Gibson gave the example of a chair and said that a chair is anything that you can sit on. Categorizing a chair as a four-legged structure made out of wood isn't enough because lots of things can be chairs: rocks, logs, cushions, etc.

I wonder how categories for ideological things fit into this definition. For example, one could categorize people as conservatives, or communists. I understand that the information needed to create these categories is still gathered through the sensory-motor system -one reads about conservatives, hears about them in the news etc. I'm not sure how one can act upon conservatism, though, in the way that one could act to sit upon an object that was categorized as a chair. Would acting upon conservatism simply be following a conservative policy? I somehow feel like that's not the whole picture, because one can have categories for things that the person itself will never act upon. Categories such as Neptune's moons, for example. Is interacting with the object, or thinking about the object enough to count as "doing the thing you need to do?"
ReplyDelete
Replies
Laura McDougall17 February 2015 at 19:27
In “To Cognize is to Categorize: Cognition is Categorization” Stevan Harnad posits that all cognition is categorization. A cognizing system is a sensorimotor system. It uses its sensorimotor skills to interact with all kinds of things in the world around it: eat them, play with them, run from them, mate with them. How does it know what to do with each thing? It categorizes them: it does the right thing with the right kind of thing. Some categories are innate, human faces, for example. Others are learned through trial and error (a kid may watch Sesames Street’s “Which of These is Not Like the Others” and learn that a tuba is different from an apple and an orange.)

Cognition-as-categorization has taken us a far way from cognition-as-computation. For one, categorization is very dependent on the “hardware” that computation theory deemed irrelevant. Each individual cognizer will have his own set of categories under his belt which he uses to tackle the world. I might go outside and see a “bird,” while an experienced bird-watcher might see a “striated pardalote.” Even categories that one would think are universal, such as colour, according the Whorf hypothesis can actually be perceived differently according to how one is shaped by language and colour. I find this to be a much richer view of cognition than computation as if accounts for all individual differences, not just the bare minimum required to pass a pen-pal Turing test.

On the other hand, computation’s symbol manipulation is not dissimilar to categorization: they both demand that one “do the right thing” when presented with another thing. The difference is that in computation the item-to-be-reacted-to is a symbol. There we run into the symbol-grounding problem. Categorization gets around this problem. A cognizing sensorimotor system is not reacting to symbols, but to objects themselves.
ReplyDelete
Replies
Nicolas.Botero17 February 2015 at 19:42
Harnad rejects Fodor's view that the categorization is completely innate, pointing out how the any "content word" in a dictionary, of which there are tens of thousands, contains a category. It would be absurd to presume that we come pre-loaded with every possible category, some of which may not exist when we are born. However, let's consider a softer position to Fodors: the set of all categories a system can conceive of is determined by innate "primitive categories" that are themselves determined by the physical structure of the system (it's sensorimotor capabilities, its functional equivalent to a brain, etc..)

The concept of primitive categories I think ties in with Harnad's idea of category grounding through sensorimotor experience. Take for example, a duck. We can categorize a duck as a bird that swims, has two legs, quacks, etc. Each one of those properties can be further decomposed to something simpler, if one considers how things are defined starting from a perceptual standpoint. Following the bird route, a bird is an animal with wings, an animal (to the perceiver who knows nothing of biology), is a thing that moves autonomously.

The hypothesis would be that everything a system can make a category for can be decomposed into a finite set of primitive categories, grounded in the system's sensorimotor capabilities, that act as the bedrock from which more abstract categories form.

Returning to the duck a.ka "thing that moves autonomously", we have conceptualized it into a category that could be account for basic parts of a nervous system. To find a thing that moves autonomously, the nervous system needs have a way of identifying movement relative to a background. This is a task already accomplished by neurons at early levels of visual processing, so can be a viable way to implement this "primitive" category. Just as objects afford different uses, the internal structure of a system affords it's ability to perceive an external object's affordances

Under this view, individual categories can still be learned, but the set of all categories that can be learned is innate. For example, we can never learnt the category of "things that sound like a wall" because we don't have echolocation.
ReplyDelete
Replies
Unknown17 February 2015 at 20:12
Being able to correctly categorize involves doing the right thing with the right kind of thing. This article makes it clear that the categorization problem is determining how we (sensorimotor systems) detect kinds of things and respond differentially to them. The story of Funes the Memorious tells us that in order to be able to function we have to be able to selectively ignore within category details and pay extra attention to (or preferentially weight) details that distinguish between categories. But how do we know which details are relevant? Some categories, such as colour, are innate (in the Darwinian sense not the Cartesian sense). But ‘all evidence suggests most of our categories are learned.’ Unsupervised learning is basically learning by manipulating things on your own with no feedback. It’s sufficient for learning which features are relevant to the categorization of a set of items/objects/things when there’s only one correct way of ordering that set. However, there can by multiple equally correct ways of categorizing a set and in that case we need supervised learning (learning with feedback) in order to figure out which categorizations are appropriate for certain contexts. What makes individuals with language unique is the fact that once we have learned a certain number of base categories through our sensorimotor capacities and supervised learning we can acquire additional categories through hearsay (someone describing the relevant features used to identify things that fit in that category). If most categories are learned and context dependant, does this mean that groups of people who live completely differently would have different ways of categorizing their world? I don’t mean this in the Whorf-hypothesis-7-billion-words-for-snow sense. I’m thinking independently from language, if you’ve always used ashes to wash dishes, you’re going to put ashes in a different category than someone who washes their dishes with Dawn and thinks of ashes as something to get rid of after a fire. Obviously in this case the Dawn person just hasn’t been in the context where they’d learn they can add ashes to the category of things to wash dishes with, but I’m wondering if there would be examples like this for more immaterial categories like emotions etc.
ReplyDelete
Replies
Unknown17 February 2015 at 20:13
As Harnad notes, “language allows us to acquire new categories indirectly through “hearsay”, without having to go through the time-consuming and risky process of direct trial-and-error learning”. (Harnad, 13)

The hierarchical separation between language as a “hearsay” categorization tool and the underlying learning mechanisms, by which categorization formally occurs, seems reminiscent of the hierarchical ordering of knowledge: math governs physics, physics governs chemistry, chemistry governs biology and so on. However, in the instance of categorization by language, certain mechanistic confounds seem to arise. First, if language really does function to acquire new categories, then its mechanism of action must follow transitively from those underlying mechanisms which govern it. Assessing this from a bottom-up approach, the initial categorization must occur at the symbol grounding level at which point sensorimotor transformations give symbols meaningful semantic content. At this level, unsupervised learning acts as an internal competitive mechanism to correlate sampled input into similar and dissimilar material. In parallel, supervised learning clusters sensory input according to error-corrective feedback. This accounts for contextualization of the input with respect to one’s sensorimotor interactions. In this manner, the invariants of objects are detected selectively relative to their variations, weighted with respect to the invariants, and recoded as abstraction of the invariants. Until now, it is theoretically clear how dynamic, continuous input transforms into static, discrete perceptual abstractions, related to the “kind” of input. However, in my opinion, the mechanism by which more abstract things (probably “symbols rather than direct sensorimotor interactions”) (Harnad. 13) network in order to initialize higher order categories is unclear. Its seems like a mechanism of this nature is required for language to occur via experience alone, as opposed to mere trial and error (although it does seem to me like trial and error might govern the language process as well). It is in integrating and separating higher order abstractions where I speculate that computational aspects of cognition occur. Language categorization seems to require a built in logic, presumably one reduced to Boolean truth values, residing in 1’s and 0’s. But, it is the question of how the initial chunking and separation of categorical precepts (at the sensorimotor level) translates into logical categorization of higher order abstract invariant features that continues to confuse me. Where do the sensorimotor and computational components of cognition converge?
ReplyDelete
Replies
Audrey Vong27 March 2015 at 10:35
“The affordances are not imposed by me; they are "external" constraints, properties of the outside world, if you like, governing its sensorimotor interactions with me. And what I do know of the outside world is only through what it affords (to my senses, and to any sensory prostheses I can use to augment them).”

I find it difficult to think of affordances as constraints imposed by external things upon us, that "govern"—rather, to met they seem more like constraints imposed by properties of our own systems upon external information—after all, how can things be anything but what they are? Regardless of how we perceive them, and the consistency and universality with which we do so, how can the outside world be said to govern its sensorimotor interactions with us anymore than we with it? For example, the average traffic light would be perceived as having lights of three different colours by a non-colour blind person, while a colour-blind individual could perceive the red and green lights as being one and the same shade. How is this governance on part of the traffic light, or a question of what it affords—other than the fact that it is red and green? Would it not then be a question of what our senses afford us? After all: “Mere cognitive scientists (sensorimotor roboticists, really) should not presume to do ontology at all, or should at least restrict their ontic claims to their own variables and terms of art -- in this case, sensorimotor systems and their inputs and outputs. By this token, whatever it is that "subtends" absolute discriminations -- whatever distal objects, events or states are the sources of the proximal projections on our sensory surfaces that afford us our capacity to see, recognize, manipulate, name and describe them -- are all on an ontological par; and subtler discriminations are unaffordable."—unaffordable properties are undeniably very present in the world, and it is just that we are unaware of them (and in terms of cognitive science, unconcerned with these unknowable things since they clearly lie outside the realm of cognition, as per the above). Even if we take affordance's substrate to be properties of the outside world, how are they grounded in anything but the limits of our own mechanisms?
ReplyDelete
Replies
Jacob Spitzer4 April 2015 at 11:58
“But I am sure I could find out, either through direct trial and error experience, my guesses corrected by feedback from the masters, and my internal neural nets busily and implicitly solving the credit-assignment problem for me, converging eventually on the winning invariants; or, if the grandmasters are willing and able to make the invariants explicit for me in words, I could find out what primroses are through hearsay.”
I think Harnad slips in here an answer to questions of ineffability. Common candidates for ineffable statements, like “What is beauty, really?” are just difficult categorizations. Harnad says that if categorization is happening, the categorizer is using invariant features that can be made explicit with some effort, and are thus effable. In other words, if the categorizer can ‘think’ the difference between two things, they can express that thought. But do we know that there aren’t thoughts that cannot be expressed?
We know some categories (chicken sex) are more difficult to make explicit than other categories (red vs. blue). Perhaps difficult categorizations rely somewhat on explicit processes and somewhat on ineffable differences, and with effort, the categorization can be made with just explicit differences. Could there be differences that can allow us to categorize, but can never be made explicit? My intuition says no, any thought translated into words, but I don’t think we’ve ruled out the possibility of ineffable properties. An example could a kind of ‘animal instinct’ that happens in our reptilian brain that influences our behavior but cannot be put into words. I know that’s very wishy-washy, but I don’t think we’ve ruled it out: there may still be categorizations that cannot be made explicit.
ReplyDelete
Replies
Unknown18 April 2015 at 05:50
"Is there any way to increase our capacity to make categorizations? One way is to add more dimensions of variation"

This sentence stuck out to me as rather surprising. Is there evidence to suggest that simply adding more dimensions of variation would increase our capacity to make categorizations, or would it simply confound the capacities we already have? For example, say you showed two sets of items to people, large items and small items. If every single item in both sets also frequently and randomly varied in a wide array of features, like shape and color, I would expect categorization would become more difficult (I would expect that participants would need to undergo some sort of reinforcement or supervised learning in order to correctly categorize).

"This is rather like what Biederman did for his experimental subjects, in
telling them what features to use to sex chickens, except that his method
was not pure hearsay, but hybrid: It was show-and-tell, not just tell,
because he did not merely describe the critical features verbally; he also
pointed them out and illustrated them visually."

This hybrid format reminds me of some of the ways dance teachers instruct students. I danced for years and some of the verbal descriptions of what we should do with our bodies were particularly odd. For example, a few of my dance teachers used to tell us that during kicks we should "lift with the backs of our legs". Of course, this is physically impossible as there is no way for a muscle to push (in this case we were told to push our legs into the air with our hamstrings instead of pulling with our quadriceps). Surprisingly, just by thinking differently the dancers movement would typically change as though they were "pushing with their hamstrings" (even if they consciously knew it was impossible to move in such a way). In this way, teachers could use language to fool dancers into using their bodies correctly. After reading this article I'm now seeing it as even more odd that humans are able to construct impossible categories to encourage particular continuous movements in other humans.
ReplyDelete
Replies
Unknown15 April 2016 at 20:53
This comment has been removed by the author.
ReplyDelete
Replies

Add comment