I'll try my best not to repeat anything my classmates have said, but it's hard to do when 50 of them have posted before I have: my main takeaway from the mirror neurons article by R+C is the way in which mirror neurons may explain action understanding. There are several questions that necessarily must be asked as followup: first, what IS understanding, actually? Are we to believe it's just the firing of neurons? I know that some others have already discussed why this is an insufficient answer that doesn't provide a causal mechanism for how we can understand anything. I feel like this is an argument we get into a lot - either we end up with a homonculus, or we end up with neurons firing; either way, someone is unsatisfied with the explanation of how we understand/know/think. This makes me wonder how will we know when we've actually found the right answer? If we're not satisfied with neurons firing, must we turn to dualism? Is cognitive science the space between neuroscience and philosophy which can once and for all answer the question of 'what is thinking?' or will it just attempt to bridge the gap between what two disciplines have already established, in their own fields, that thinking is? Another thought I had was that, if we have these exciting mirror neurons that can supposedly indicate action understanding, then might not we also have even more interesting neurons elsewhere in the brain, that can indicate understanding of other, more complicated things - e.g., ideas, emotions, etc.? Based on what I have read about Theory of Mind, which is becoming increasingly popular, it does seem possible that we will within the next few decades have discovered other types of neurons that code for understanding of other types of knowledge, like knowlege of others' thoughts/emotions. So when we put all these together, will that amount to thinking? And that brings me back to the first few questions I asked in this increasingly curious post - will we ever be satisfied with the neurons-are-firing explanation for thinking?
In an earlier comment, Prof Harnad asked: "Did finding that there were mirror neurons tell us anything that would help as generate (reverse-engineer) the T3 capacities that are correlated with firing by mirror neurons?"
The strongest claims that can be made on the basis of findings such as those in this article are claims like "the mirror neuron system is involved in action understanding". It is possible that mirror neurons play some role in generating T3 performance, as action understanding seems to be a critical part of T3 ability. But it would be an overstatement to say that it helps explain the causal mechanism that generates T3 performance. Correlative cognitive neuroscience findings point to possible components of a causal mechanism, but are unable to do this with confidence, since they are incapable of explaining how one possible component fits into the rest of the found possible components.
Given the relative youth of cognitive neuroscience, I'm not sure what a better method for finding new components for a causal mechanism would consist in. At this point, however shallow, the resulting insights into neural causal mechanisms might at least serve to confirm or correct our existing cognitive theories.
The fact that the size of objects in observed human-object interactions influences the aperture of the mouth during pronunciation indicates that we should place hand motion and mouth motor movements in the same category with respect to their neural mechanisms. However, this isn’t very informative about whether the cognitive mechanisms that produce these phenomena are related. Piecewise correlative findings give us as much insight into the mechanism underlying T3 capacities as laying out a billion unordered puzzle pieces uniform in shape and asking us to solve.
I think it's interesting that the article attempts a definition of understanding, or rather presupposes what it is. According to the text, one form of understanding would manifest as the cortical activation of zones of the premotor cortex corresponding to the limbs it is observing. In this process, the visual cortex recognizes the input as corresponding to its own body parts, and thus the cortical activation due to observation of an action is similar to that of the execution of that action, with some limits due to species. This means understanding would be ''making yours'', identifying in some way, with what you perceive in the world. So we say of these subjects that they understand because their actions(or activations) correspond with material reality, such as when the neuron fires when the monkey knows there is an object behind the screen because it saw it before even though it doesn't see it at the moment the test is effectuated. This understanding, effectuated ''without any cognitive mediation'', is then what allowed us to develop language by transferring the meaning of actions onto the sounds that represent them. However, when R&C say that meaning ''transferred, in evolution, from hand gestures to oro-laryngeal gestures'' I think it would be more appropriate to say that when humans developed an auditory mirror system, new words were created increasingly independent from gesture.
If the goal of cognitive scientists is to reverse engineer a cognizing system that passes as a T3 model, I cannot help but think that neuroscience is endlessly informative. The challenge is ambitious to say the least, since we are not only cognizing but also looking into our own minds and brains to design cognition outside of the body.
In terms of activation, we saw in this article that simple observation of goal-directed actions performed by closely related primates triggers the same neurons that are used when we do those actions ourselves. The development of our cognitive capacities was presumably jump-started when our species started sharing ideas through an agreed-upon system of symbols. In order for two individuals to know a symbol represents the same thing in both of their heads, there has to be some sort of shared experience. This is where the mirror neuron becomes interesting. It seems to be a mechanism that takes meaningless symbols (for example, sound wave fluctuations produced by speech), relates that set of symbols to a time in history when they existed in the receptive individual, thus providing the sound wave with meaning, so that meaning can be interpreted by the individual. If this is the case, then it could give cognitive scientists a shot at engineering a physical symbol/semantics-interpreting mechanism that could get us one step closer to understanding cognition.
“Second, the temporal characteristics of cortical excitability, during action observation, suggest that human mirror-neuron systems code also for the movements forming an action and not only for action as monkey mirror-neuron systems do. These properties of the human mirror-neuron system should play an important role in determining the humans’ capacity to imitate others’ action.”
If there are aspects of motor neurons that are uniquely human, than using neuroscience to find out what mechanism allows human motor neurons to anticipate actions has an implication for cognitive science as well. To be able to imitate others’ actions, as the article argues, not only has to do with physical mimicry but with learning as well. Learning, imitating, and anticipating actions are topics that fall under cognition as well as neuroscience. If the human motor neuron system is indeed coding for certain actions, then that is a powerful way to consider how our brain might organize information—even information that is not reflected in the present, physical world we see at a given moment (I’m talking about intuition, essentially). If our brain is coding for actions and sorting them into categories, then this certainly falls under the domain of cognitive science. While neuroscience can help discover the mechanism, it appears that cognitive science could extrapolate to describe a more general model of how the brain might function based on the human motor neuron system.
“In the case of humans, there is another faculty that depends on the observation of others’ actions: imitation learning. Unlike most species, we are able to learn by imitation, and this faculty is at the basis of human culture”
Rizolatti and Craighero’s article described mirror neurons: neurons that fire both while performing an action and while observing others perform an action. Most interestingly, they talk about the role of mirror neurons in imitation learning and their potential role in the development of language.
Reading this article and thinking about what we have read and learned in class so far, I thought about the potential for modeling neural networks of the brain, particularly those of mirror neurons, in order to create a robot that would pass the Turing Test. It also made me think more about “imitation” and how we said that the Turing Test was not about imitation at all, but that it was about finding a causal mechanism behind cognition. If we consider “imitation” as a way that we grow and learn, particularly to interact with one another in society, imitation may actually play an integral role in this causal mechanism. Now, my question is: would this be achievable at the level of T3, or would this require T4?
In the paper The Mirror-Neuron System, Rizzolatti and Craighero use different neuro-imaging methods to compare the mirror-neuron systems in humans and monkeys in neuroscientific and neuopsychological aspects. They examine which parts of the brain are activated during viewing objects and moving our bodies. They tell us what humans can do while monkeys cannot. They put forward ideas like language/speech might be derived from hand movements.
However, the experiments and examinations they run still cannot tell us what exactly happens when imitation happens. In last week's reading, I recall when Searle comments on brain's involvement in imitation. He says:"If we had to know how the brain worked to do AI, we wouldn't bother with AI. " I think this sentence can best express my feelings towards the paper by Rizzolatti and Craighero. It is true that from the experiments we can discover some relations among movements and sights, differences in brain activation between monkeys and human and then understand what brain areas are correlated with what functions; however, these findings are of little use if we want to explain how and why imitation actually works (the materials we talk in class - instead of what part of the brain is associated with imitation); otherwise, we do not need to bother AI or the imitation game. These problems are easy to solve.
Nevertheless, the experiments that Rizzolatti and Craighero conduct are influential in a general way. They give accounts for some behavioral differences between monkeys and humans. Although imitation is not only for humans, there are some things that monkeys are unable to do. They connect speech with hand movements, and I believe encourage more researches in this field.
"When we eat, we move our mouth, tongue, and lips in a specific manner. … If, while making this action, we blow air through the oro-laryngeal cavities, we produce a sound like “mnyam-mnyam,” or “mnya-mnya,” words whose meaning is almost universally recognized (Paget 1930). Thus through such an association mechanism, the meaning of an action, naturally understood, is transferred to sound. … A fundamental step, however, toward speech acquisition was achieved when individuals … became able to generate the sounds originally accompanied by a specific action without doing the action. This new capacity should have led to (and derived from) the acquisition of an auditory mirror system, developed on top of the original audio-visual one, but which progressively became independent of it. " I'm afraid that I am not fully convinced by this account of how we evolved language, not that I don't think it's a plausible theory, but that it's still missing some key ingredients to be able to explain every aspect about human language competence. It makes perfect sense that our ancestors, with the help of their primitive imitation capacities, learned to associate the sound one makes while eating, to the action of eating per se. This explanation could be extended to a number of other daily activities: hunting (perhaps the sound of a spear hurtling through the air), drinking, childbirth, etc. Once they all became familiar with these words (i.e. the audio representations of actions are grounded), they can combine these words to express more complex meanings. My problem with this theory concerns the abstract concepts. Concrete words are the easy part of language. Simply point at or act out the thing you are trying to verbally express, and the others should understand. However, when our ancestors evolved to have deep thoughts about life, how would they communicate with each other about these ideas? Here in the 21st century, in a cognitive science class, we are still having trouble defining what "feeling" is, how did our ancestors ever manage to get that done? How could they ever wrap their head around the concept of a Peekaboo Unicorn? Getting back to my point: I believe this theory is solid, but it still needs to take into account the "jump" from communicating about just concrete entities to complex, abstract ideas.
It's also mildly interesting how this paper, and the multitude of experiments mentioned in it, kind of proves Fodor's (4b) point about brilliant minds spending time and money finding out answers to questions that haven't been raised yet. This paper goes into a lot of details and technicalities about similarities/differences of the neuroanatomy and localization/specification between primate and human mirror neurons, but (as Fodor would say) what does all this have to do with understanding what the functional roles of these neurons are?
“The neuron illustrated in the figure responded to the observation of grasping and holding (A, full vision). The neuron discharged also when the stimulus-triggering features (a hand approaching the stimulus and subsequently holding it) were hidden from monkey’s vision (B, hidden condition). As is the case for most mirror neurons, the observation of a mimed action did not activate the neuron (C, full vision, and D, hidden condition). Note that from a physical point of view B and D are identical. It was therefore the understanding ofthe meaning ofthe observed actionsthat determined the discharge in the hidden condition.”
Any direct neurological evidence of understanding makes it exceptionally hard for those of us who attempt to maintain computationalist ideals (even those of us who tried to hold up through Searle’s thought experiments). Mirror neurons will trigger in response to observed actions. This neuronal activity then seems to correspond to specific muscle activation respective to the detected action, even if the action wasn’t visually observed. Furthermore the neurons won’t react identically to two identical actions based on the observers knowledge of the situation (genuine vs mimed hidden-actions). This is taken even further with human subjects who’s mirror neuron systems seem to react equally to novel behaviors. At this point, it seems irresistible to designate mirror neurons as the causal mechanism behind cognitive functions – which is exactly what cognitive science has set out to explain, and what computationalism (with regards to understanding) fails to explain. Computationalism’s claim to be implementation neutral is also put to rest, seeing as there are no mirror neurons, nor anything remotely akin to mirror neurons present in Chinese room. Of course what this level of neurology fails to explain is our experience of thinking. Of course this train of thought dives headfirst into the hard problem. But when we close our eyes and picture an action we have just observed, it can be certain that there are mirror neurons involved, but they have little to know chance of explaining the internal imagery we feel to be the manifestation of our understanding.
I guess I'm entirely biased in this case. I find neuroscience and the advancements we've made in the way we study the brain to be fascinating. The idea that there are mirror neurons says a lot to me about the evolution of human understanding. Somehow we managed to evolve the ability to abstract sound meaning from gestural meaning (Rizzolatti & Craighero, p.184). This piece of information begins to explain how we learned to speak, how we process non-verbal information like gestures and body language and this in turn begins to tell us why we do certain actions to convey certain messages.
If we are wondering whether or not this information is useful to "reverse-engineering the brain", I'm not sure what to tell you. I don't believe it is possible to do in the first place but hypothetically, understanding neural substrates and the pattern of neuronal firing in the brain prompted by visual stimuli should most certainly be helpful in creating such a brain. It is clear to see that mirror neurons and important to how we learn. Seeing someone lift a glass to their lips, drink from it, and place it down, activates neurons in our motor systems. This in turn is now a motion we can imitate. A more complex example is perhaps ballet moves. In order to learn how to do certain moves requires our brains to put together how others move and then do the same with our own body parts. So say we are creating this robot that passes the Turing Test. Is this kind of information completely unnecessary? Do AI researchers not need to be aware of how the mind actually works to learn things? Or are we just going to be able to upload information into them where there are instructions on how to produce these movements? I feel this constantly boils down to the question of whether or not the insides matter to what we see on the outside. Does it matter what happens in TMS studies? Does it matter that there exist differences between monkey's and human's mirror neuron systems? I believe that it does matter. We exhibit different behaviors from monkeys because of the differences in our brains. We need to know what human brains are like if we ever want to create something that does what we do. If we want to know why we do what we do, we need to look at the control center - our brains. Otherwise, we are just looking at what we do and how we do it as seen by an observer. We get no closer to knowing the answer to why, which I find to be much more important if we are ever to understand how to create a robot that cannot be distinguished from a human being.
Mirror neurons have given us insight on how it is that we learn to do what we do and gives us details on the process of that learning from one step to the other. It is indeed interesting to know that is happen in the brain; however, I don't think it really brings something new to the picture.
Someone could have gotten the same data on those learning steps and learning modes by thoroughly observing someone in the process of learning or from the other side, in the process of teaching. Someone could even figure it out by thinking about it, about how it is that his person does what he does: he is doing this then he must be doing this and that or this and that and so on and so forth.
On another point, if we assume that the discovery of mirror neurons did tell us something new about how we learn new behaviour it would still not be able to tell us how to reverse engineer T3 and this is because we don't how what happens in the brain gets transmitted to our consciousness, for all we know these activations are just electrical changes and metabolic changes occurring on certain instances, but how does these changes are linked to our consciousness ?
I feel that in order to move forward with regards to the integration of neurophysiological research into cognitive science, the language of interaction needs to be developed. We know already that the questions of cognitive science are of the nature: "why do we do what we do?", but the questions of neuroscience are not so well defined. I would like to propose the candidate question: "how is information processed in the brain". Of course, information processing needs to be defined itself in order for this definition to have value. I propose the following: information processing refers to the process by which states of known semantic value (we know what the state is) give rise - potentially through unknown intermediate states - to new states of known semantic value. To illustrate this, I would suggest that the turning on of a light is a form of information processing. The light initially has a state known to be "off". Once the switch is flipped, that state (known to be "a flipped switch") will then deterministically give rise to a new known state in the light, namely "on". Neuroscience asks questions regarding the steps that take us from known state to known state. For example, neuroscience attempts to find intermediary states between our physical sensory inputs (known states of "photon reflection", "waves in air", or at a higher level, "stuff is there"), and the resulting state which is our internal representation of our surroundings.
How can cognitive science use this information? By defining mechanisms of moving from known state to known state in a deterministic way, neuroscience can teach us general principles about how to engineer systems which can deterministically generate the complex state (which is the "do" in the "why we do what we do") that defines cognition. To use Searle's term, I feel that the questions of neurophysiology today are part of a system of "weak" neuroscience - an attempt to understand the brain so as to help create our own models, rather than itself reconstruct our cognition.
This is where I feel that there is value in mirror neurons. They define an intermediate state which could provide a method for transforming sensorimotor information into understanding through the use of states which are triggered both by perceiving and by preforming action. Certainly overlapping patterns of fire in mirror neurons between two related situations is not understanding: I could create a similar system using light switches. What it is is a building block towards understanding - a potential way to define a deterministic relationship between our sensory transduction, and an understanding of what to do and when.
2- even if it had done so I don't believe it would have been worth hurting other living beings especially ones that are able to feel. I think what we are doing as a race can be termed "speceism" it is an immoral abuse of our "power". We have absolutely no right whatsoever to take animals from their natural habitats, young ones from their mothers, put them in cages and cause them unnecessary harm and all that for the benefit of "science". By doing what we are doing we are disturbing the natural equilibrium of this planet which we call our home but it is not just our home, we are at an age where we have to learn how co-habitate with everyone else on this planet, we all have the right to live here as equals.
With regards to the importance of the mirror-neuron system to communication, and gestural communication in particular, Rizzolatti et al. state: "Mirror neurons represent the neural basis of a mechanism that creates a direct link between the sender of a message and its receiver. Thanks to this mechanism, actions done by other individuals become messages that are understood by an observer without any cognitive mediation."
While the neurophysiological evidence of mirror neurons, and the details of their function and operation, seem undeniable, this line of reasoning and its implications seems to me to be a gross jump in the interpretation thereof. It is of course appealing to ascribe the mirror-neuron system to some direct communicatory mechanism, which would be the predecessor of language—that property which is often thought to demarcate humans from other organisms—but that it creates a uniquely "direct link between the sender of a message and receiver" seems to be romantic overgeneralisation when in fact, it is the observer that is ascribing intentionality to a particular gesture (and we have seen that intentionality, derived from contextual understanding of a particular action, is important to the response of mirror neurons). I would not argue against some form of communication arising through this manner—through a multitude of potential misunderstandings due to individual interpretation, but arise nonetheless—but the implied bilaterality of this initial simple interaction that would be required to transmit information between to agents, and not merely between movement of a muscle and neuron, is nonexistent. Mirror neurons may be indicative of how and when we ascribe intentionality to an action, but are insufficient to account for the direct bilateral links required by any system that would call communication its primary purpose.
Do mirror neurons so far go any further than being one correlate and component of our capacity to recognize someone else's movement as the same a movement I can make? Analogue movement recognition.
This is probably a part of our capacity to imitate movements, but language?
Is my being able to say and mean "the cat is on the mat" and my ability to understand what someone else means when they say "the cat is on the mat" recognizing someone else's movement as the same as my own?
It might be if he were singing in Chinese, and I could imitate him.
I agree with what Audrey says! The final sentence of that passage, "Thanks to this mechanism, actions done by other individuals become messages that are understood by an observer without any cognitive mediation,” leaves me asking questions. Without any cognitive mediation, to my understanding, means without any further processing - this doesn’t make sense to me.
I read Fodor’s paper before reading this one, and so I was struck but the abundance of indirect evidence supporting the role of mirror neurons in cognition processes. To answer Prof. Harnad’s question posted on an earlier thread - no, I don’t think finding neurons that fire when we imitate doesn’t explain how the brain causes our capacity to imitate (at least this paper doesn’t).
Mirror neurons as the basis of learning could have huge implications on cognitive science and robotics. While obviously it is not possible to reverse engineer the brain, mirror neurons could be the key to developing a robot that could pass T3 and after some more understanding, T4.
Using the idea of imitation learning, it is currently possible to build a robot that could mimic some speech or some motions. Incorporating a degree of entropy into the response system such that the responses are no longer predictable could be the beginnings of the answer of how to convince the judges during the test.
Furthermore, using this knowledge, teaching machines to learn could be the next step. This is already done with certain algorithms where you enter an input and then it generates a response that only it could figure out and extrapolate on but this could be applied to more "human" operations like language acquisition and skill building.
In the video at the top you said J. Fodor had too hastily dismissed brain imaging.
He doesn't think brain imaging has told us anything new about how the brain works because the question of "where" has nothing to do with "how". How could learning the where of human cognition tell us anything about how human cognition works?
If our problem with this article (and moreover the practice of locating neurons for cognitive capacities) lies in its purely correlative form, I wonder if we will ever be satisfied. Causal mechanism can't be found, they are thought up in connection with correlative findings. They are put simply, ideas. What is lacking might not be an empirical proof of the causal mechanism for cognition in humans, but instead a satisfying theory which explains how mirror neurons cause thinking. Causality is applied to significant correlations by crosschecking it with other correlations and by coming up with logical reasons why two (or more) things co-vary in the way they do.
“The incredibly confusing organization of Broca’s area in humans, where phonology, semantics, hand actions, ingestive actions and syntax are all intermixed in a rather restricted neural space, is probably a consequence to [the aforementioned] evolutive trend.” (186) After reviewing Rizzolatti and Craighero’s in review of mirror-neuron systems, it seems as though one might be drawn to arguments advocating for its extreme locationist perspective, implying that distinct neurons, in distinct locations in the brain, might underlie all semantic-syntactic transformations and might therefore underlie the development of communication systems. The authors steadily reference distinct anatomical/cito-architechtonic areas that become activated during instances in which the action understanding/imitation learning criteria of mirror-neurons are fulfilled. This is where, however, a skeptic might criticise the authors for their inability to highlight a causal mechanism for cognition that might satisfy T3 capacities. Although I cannot support the locationist myself, I have yet to see why empirical findings of the “mirror-neuron system” (with emphasis on system) can’t be a stepping stone in understanding cognition. The authors’ primary evidence comes from correlational techniques that assess the interactions of brain activity both spatially and temporally. Although correlation is not causation, it provides empirical data that might be used to develop hypotheses concerning 3rd party players in a large system of interacting brain regions. I can’t stress enough that the regions might be nowhere as important as the dynamical means by which the system interacts (according to the temporal or spatial forms of the messages the neurons send). The reverse engineering of these dynamic properties (and not a reconstruction of the locations themselves, which may or may not be an epiphenomenal bi-product of evolution), through simulation, or computer modeling, may be enough to satisfy T3 capacities. As imaging, or electrophysiological techniques become more spatially and temporally resolute, how can we say that possible cell-assembly systems, which we must reverse engineer, will never uncover, with more certainty, the potential means by which sensory-motor transformations help provide meaning to symbols, and language processing? With regards to experimentation on animals: I don’t have any good evidence for why or why not animals should be tested on in cognitive science research. However, the moral standard for hurting animals would ultimately rely on whether they are sentient enough to feel, and whether we have a strong enough reason to believe this is so. It seems to me that the best we could do to find this out with certainty, would be to perform some version of the Turing Test in order to reverse engineer animal cognition. I can only rely on a deep intuition that this is possible. Furthermore, it is clear that animal testing, more often than not, is not a utilitarian means for saving lives, but rather a matter of curiosity for academic researchers. In instances when determining the mechanism necessary for cognition, I would hope that functional imaging (on humans), using non-invasive techniques that do not involve the hurting of animals, is a strong enough means for contributing to the construction of T3 passing models.
"This fact suggests that a necessary step for speech evolution was the transfer of gestural meaning, intrinsic to gesture itself, to abstract sound meaning. From this follows a clear neurophysiological prediction: Hand/arm and speech gestures must be strictly linked and must, at least in part, share a common neural substrate. [...] A number of studies prove that this is true." p.16
At the beginning of the article, the author noted that the section on language is very speculative. When reading the discussion on mirror neurons and language, however, I got the impression of assertiveness in the truthfulness behind the findings. I feel like the passage I quoted above captures the sense of assertiveness in what was initially admitted to be speculation. Often in reading research articles it becomes apparent that the author may have some degree of bias in how the data is presented. Especially in pop science magazines, the findings are often sensationalized and the colored brain images seem to be meant more for aesthetic appeal than for analysis.
The paper shows the evidence suggesting the existence of mirror neurons in other primates and human. Mirror neurons are the neurons in our brain that enable us to understand the actions of others and to learn by imitation. For example, learning to drive by observing the demonstrator.
I can see how it is exciting for researchers to discover mirror neurons. It is like Columbus first discovered North America. Except the map is on our brain. But: 1. It is not the discovery of new cognition ability; the discovery of mirror comes after we realize that we have the ability to imitate. We knew that we can imitate, research had been done on imitation before discovering mirror neuron. 2. It did not helped cognitive science, cognitive science tries to figure out how it works, where it located does not matter.
Admittedly, discovering our brain map is important for areas such as clinical neurology. Another benefit would be it satisfied our curiosity, I had taken a course in Brain lesion and the information is intriguing. But neural science is always a controversial subject, especially before neural-imaging. Unlike other subject where you can observe easily, there's no other way to study brain than damaging part of it. Because how inhumane this is, research of neural science are mostly done on animals. It is cruel. I can only say that with the help of new technology such as neural imaging. We can study the brain without damaging it.
I'll try my best not to repeat anything my classmates have said, but it's hard to do when 50 of them have posted before I have:
ReplyDeletemy main takeaway from the mirror neurons article by R+C is the way in which mirror neurons may explain action understanding. There are several questions that necessarily must be asked as followup: first, what IS understanding, actually? Are we to believe it's just the firing of neurons? I know that some others have already discussed why this is an insufficient answer that doesn't provide a causal mechanism for how we can understand anything. I feel like this is an argument we get into a lot - either we end up with a homonculus, or we end up with neurons firing; either way, someone is unsatisfied with the explanation of how we understand/know/think. This makes me wonder how will we know when we've actually found the right answer? If we're not satisfied with neurons firing, must we turn to dualism? Is cognitive science the space between neuroscience and philosophy which can once and for all answer the question of 'what is thinking?' or will it just attempt to bridge the gap between what two disciplines have already established, in their own fields, that thinking is?
Another thought I had was that, if we have these exciting mirror neurons that can supposedly indicate action understanding, then might not we also have even more interesting neurons elsewhere in the brain, that can indicate understanding of other, more complicated things - e.g., ideas, emotions, etc.? Based on what I have read about Theory of Mind, which is becoming increasingly popular, it does seem possible that we will within the next few decades have discovered other types of neurons that code for understanding of other types of knowledge, like knowlege of others' thoughts/emotions. So when we put all these together, will that amount to thinking? And that brings me back to the first few questions I asked in this increasingly curious post - will we ever be satisfied with the neurons-are-firing explanation for thinking?
In an earlier comment, Prof Harnad asked:
ReplyDelete"Did finding that there were mirror neurons tell us anything that would help as generate (reverse-engineer) the T3 capacities that are correlated with firing by mirror neurons?"
The strongest claims that can be made on the basis of findings such as those in this article are claims like "the mirror neuron system is involved in action understanding". It is possible that mirror neurons play some role in generating T3 performance, as action understanding seems to be a critical part of T3 ability. But it would be an overstatement to say that it helps explain the causal mechanism that generates T3 performance. Correlative cognitive neuroscience findings point to possible components of a causal mechanism, but are unable to do this with confidence, since they are incapable of explaining how one possible component fits into the rest of the found possible components.
Given the relative youth of cognitive neuroscience, I'm not sure what a better method for finding new components for a causal mechanism would consist in. At this point, however shallow, the resulting insights into neural causal mechanisms might at least serve to confirm or correct our existing cognitive theories.
The fact that the size of objects in observed human-object interactions influences the aperture of the mouth during pronunciation indicates that we should place hand motion and mouth motor movements in the same category with respect to their neural mechanisms. However, this isn’t very informative about whether the cognitive mechanisms that produce these phenomena are related. Piecewise correlative findings give us as much insight into the mechanism underlying T3 capacities as laying out a billion unordered puzzle pieces uniform in shape and asking us to solve.
I think it's interesting that the article attempts a definition of understanding, or rather presupposes what it is. According to the text, one form of understanding would manifest as the cortical activation of zones of the premotor cortex corresponding to the limbs it is observing. In this process, the visual cortex recognizes the input as corresponding to its own body parts, and thus the cortical activation due to observation of an action is similar to that of the execution of that action, with some limits due to species. This means understanding would be ''making yours'', identifying in some way, with what you perceive in the world. So we say of these subjects that they understand because their actions(or activations) correspond with material reality, such as when the neuron fires when the monkey knows there is an object behind the screen because it saw it before even though it doesn't see it at the moment the test is effectuated. This understanding, effectuated ''without any cognitive mediation'', is then what allowed us to develop language by transferring the meaning of actions onto the sounds that represent them. However, when R&C say that meaning ''transferred, in evolution, from hand gestures to oro-laryngeal gestures'' I think it would be more appropriate to say that when humans developed an auditory mirror system, new words were created increasingly independent from gesture.
ReplyDeleteIf the goal of cognitive scientists is to reverse engineer a cognizing system that passes as a T3 model, I cannot help but think that neuroscience is endlessly informative. The challenge is ambitious to say the least, since we are not only cognizing but also looking into our own minds and brains to design cognition outside of the body.
ReplyDeleteIn terms of activation, we saw in this article that simple observation of goal-directed actions performed by closely related primates triggers the same neurons that are used when we do those actions ourselves. The development of our cognitive capacities was presumably jump-started when our species started sharing ideas through an agreed-upon system of symbols. In order for two individuals to know a symbol represents the same thing in both of their heads, there has to be some sort of shared experience. This is where the mirror neuron becomes interesting. It seems to be a mechanism that takes meaningless symbols (for example, sound wave fluctuations produced by speech), relates that set of symbols to a time in history when they existed in the receptive individual, thus providing the sound wave with meaning, so that meaning can be interpreted by the individual. If this is the case, then it could give cognitive scientists a shot at engineering a physical symbol/semantics-interpreting mechanism that could get us one step closer to understanding cognition.
“Second, the temporal characteristics of cortical excitability, during action observation, suggest that human mirror-neuron systems code also for the movements forming an action and not only for action as monkey mirror-neuron systems do. These properties of the human mirror-neuron system should play an important role in determining the humans’ capacity to imitate others’ action.”
ReplyDeleteIf there are aspects of motor neurons that are uniquely human, than using neuroscience to find out what mechanism allows human motor neurons to anticipate actions has an implication for cognitive science as well. To be able to imitate others’ actions, as the article argues, not only has to do with physical mimicry but with learning as well. Learning, imitating, and anticipating actions are topics that fall under cognition as well as neuroscience. If the human motor neuron system is indeed coding for certain actions, then that is a powerful way to consider how our brain might organize information—even information that is not reflected in the present, physical world we see at a given moment (I’m talking about intuition, essentially). If our brain is coding for actions and sorting them into categories, then this certainly falls under the domain of cognitive science. While neuroscience can help discover the mechanism, it appears that cognitive science could extrapolate to describe a more general model of how the brain might function based on the human motor neuron system.
“In the case of humans, there is another faculty that depends on the observation of others’ actions: imitation learning. Unlike most species, we are able to learn by imitation, and this faculty is at the basis of human culture”
ReplyDeleteRizolatti and Craighero’s article described mirror neurons: neurons that fire both while performing an action and while observing others perform an action. Most interestingly, they talk about the role of mirror neurons in imitation learning and their potential role in the development of language.
Reading this article and thinking about what we have read and learned in class so far, I thought about the potential for modeling neural networks of the brain, particularly those of mirror neurons, in order to create a robot that would pass the Turing Test. It also made me think more about “imitation” and how we said that the Turing Test was not about imitation at all, but that it was about finding a causal mechanism behind cognition. If we consider “imitation” as a way that we grow and learn, particularly to interact with one another in society, imitation may actually play an integral role in this causal mechanism. Now, my question is: would this be achievable at the level of T3, or would this require T4?
In the paper The Mirror-Neuron System, Rizzolatti and Craighero use different neuro-imaging methods to compare the mirror-neuron systems in humans and monkeys in neuroscientific and neuopsychological aspects. They examine which parts of the brain are activated during viewing objects and moving our bodies. They tell us what humans can do while monkeys cannot. They put forward ideas like language/speech might be derived from hand movements.
ReplyDeleteHowever, the experiments and examinations they run still cannot tell us what exactly happens when imitation happens. In last week's reading, I recall when Searle comments on brain's involvement in imitation. He says:"If we had to know how the brain worked to do AI, we wouldn't bother with AI. " I think this sentence can best express my feelings towards the paper by Rizzolatti and Craighero. It is true that from the experiments we can discover some relations among movements and sights, differences in brain activation between monkeys and human and then understand what brain areas are correlated with what functions; however, these findings are of little use if we want to explain how and why imitation actually works (the materials we talk in class - instead of what part of the brain is associated with imitation); otherwise, we do not need to bother AI or the imitation game. These problems are easy to solve.
Nevertheless, the experiments that Rizzolatti and Craighero conduct are influential in a general way. They give accounts for some behavioral differences between monkeys and humans. Although imitation is not only for humans, there are some things that monkeys are unable to do. They connect speech with hand movements, and I believe encourage more researches in this field.
"When we eat, we move our mouth, tongue, and lips in a specific manner. … If, while making this action, we blow air through the oro-laryngeal cavities, we produce a sound like “mnyam-mnyam,” or “mnya-mnya,” words whose meaning is almost universally recognized (Paget 1930). Thus through such an association mechanism, the meaning of an action, naturally understood, is transferred to sound. … A fundamental step, however, toward speech acquisition was achieved when individuals … became able to generate the sounds originally accompanied by a specific action without doing the action. This new capacity should have led to (and derived from) the acquisition of an auditory mirror system, developed on top of the original audio-visual one, but which progressively became independent of it. "
ReplyDeleteI'm afraid that I am not fully convinced by this account of how we evolved language, not that I don't think it's a plausible theory, but that it's still missing some key ingredients to be able to explain every aspect about human language competence. It makes perfect sense that our ancestors, with the help of their primitive imitation capacities, learned to associate the sound one makes while eating, to the action of eating per se. This explanation could be extended to a number of other daily activities: hunting (perhaps the sound of a spear hurtling through the air), drinking, childbirth, etc. Once they all became familiar with these words (i.e. the audio representations of actions are grounded), they can combine these words to express more complex meanings. My problem with this theory concerns the abstract concepts. Concrete words are the easy part of language. Simply point at or act out the thing you are trying to verbally express, and the others should understand. However, when our ancestors evolved to have deep thoughts about life, how would they communicate with each other about these ideas? Here in the 21st century, in a cognitive science class, we are still having trouble defining what "feeling" is, how did our ancestors ever manage to get that done? How could they ever wrap their head around the concept of a Peekaboo Unicorn? Getting back to my point: I believe this theory is solid, but it still needs to take into account the "jump" from communicating about just concrete entities to complex, abstract ideas.
It's also mildly interesting how this paper, and the multitude of experiments mentioned in it, kind of proves Fodor's (4b) point about brilliant minds spending time and money finding out answers to questions that haven't been raised yet. This paper goes into a lot of details and technicalities about similarities/differences of the neuroanatomy and localization/specification between primate and human mirror neurons, but (as Fodor would say) what does all this have to do with understanding what the functional roles of these neurons are?
“The neuron illustrated in the figure responded to the observation of grasping and holding (A, full vision). The neuron discharged also when the stimulus-triggering features (a hand approaching the stimulus and subsequently holding it) were hidden from monkey’s vision (B, hidden condition). As is the case for most mirror neurons, the observation of a mimed action did not activate the neuron (C, full vision, and D, hidden condition). Note that from a physical point of view B and D are identical. It was therefore the understanding ofthe meaning ofthe observed actionsthat determined the discharge in the hidden condition.”
ReplyDeleteAny direct neurological evidence of understanding makes it exceptionally hard for those of us who attempt to maintain computationalist ideals (even those of us who tried to hold up through Searle’s thought experiments). Mirror neurons will trigger in response to observed actions. This neuronal activity then seems to correspond to specific muscle activation respective to the detected action, even if the action wasn’t visually observed. Furthermore the neurons won’t react identically to two identical actions based on the observers knowledge of the situation (genuine vs mimed hidden-actions). This is taken even further with human subjects who’s mirror neuron systems seem to react equally to novel behaviors. At this point, it seems irresistible to designate mirror neurons as the causal mechanism behind cognitive functions – which is exactly what cognitive science has set out to explain, and what computationalism (with regards to understanding) fails to explain. Computationalism’s claim to be implementation neutral is also put to rest, seeing as there are no mirror neurons, nor anything remotely akin to mirror neurons present in Chinese room.
Of course what this level of neurology fails to explain is our experience of thinking. Of course this train of thought dives headfirst into the hard problem. But when we close our eyes and picture an action we have just observed, it can be certain that there are mirror neurons involved, but they have little to know chance of explaining the internal imagery we feel to be the manifestation of our understanding.
I guess I'm entirely biased in this case. I find neuroscience and the advancements we've made in the way we study the brain to be fascinating. The idea that there are mirror neurons says a lot to me about the evolution of human understanding. Somehow we managed to evolve the ability to abstract sound meaning from gestural meaning (Rizzolatti & Craighero, p.184). This piece of information begins to explain how we learned to speak, how we process non-verbal information like gestures and body language and this in turn begins to tell us why we do certain actions to convey certain messages.
ReplyDeleteIf we are wondering whether or not this information is useful to "reverse-engineering the brain", I'm not sure what to tell you. I don't believe it is possible to do in the first place but hypothetically, understanding neural substrates and the pattern of neuronal firing in the brain prompted by visual stimuli should most certainly be helpful in creating such a brain. It is clear to see that mirror neurons and important to how we learn. Seeing someone lift a glass to their lips, drink from it, and place it down, activates neurons in our motor systems. This in turn is now a motion we can imitate. A more complex example is perhaps ballet moves. In order to learn how to do certain moves requires our brains to put together how others move and then do the same with our own body parts.
So say we are creating this robot that passes the Turing Test. Is this kind of information completely unnecessary? Do AI researchers not need to be aware of how the mind actually works to learn things? Or are we just going to be able to upload information into them where there are instructions on how to produce these movements?
I feel this constantly boils down to the question of whether or not the insides matter to what we see on the outside. Does it matter what happens in TMS studies? Does it matter that there exist differences between monkey's and human's mirror neuron systems? I believe that it does matter. We exhibit different behaviors from monkeys because of the differences in our brains. We need to know what human brains are like if we ever want to create something that does what we do. If we want to know why we do what we do, we need to look at the control center - our brains. Otherwise, we are just looking at what we do and how we do it as seen by an observer. We get no closer to knowing the answer to why, which I find to be much more important if we are ever to understand how to create a robot that cannot be distinguished from a human being.
Mirror neurons have given us insight on how it is that we learn to do what we do and gives us details on the process of that learning from one step to the other. It is indeed interesting to know that is happen in the brain; however, I don't think it really brings something new to the picture.
ReplyDeleteSomeone could have gotten the same data on those learning steps and learning modes by thoroughly observing someone in the process of learning or from the other side, in the process of teaching. Someone could even figure it out by thinking about it, about how it is that his person does what he does: he is doing this then he must be doing this and that or this and that and so on and so forth.
On another point, if we assume that the discovery of mirror neurons did tell us something new about how we learn new behaviour it would still not be able to tell us how to reverse engineer T3 and this is because we don't how what happens in the brain gets transmitted to our consciousness, for all we know these activations are just electrical changes and metabolic changes occurring on certain instances, but how does these changes are linked to our consciousness ?
I feel that in order to move forward with regards to the integration of neurophysiological research into cognitive science, the language of interaction needs to be developed. We know already that the questions of cognitive science are of the nature: "why do we do what we do?", but the questions of neuroscience are not so well defined. I would like to propose the candidate question: "how is information processed in the brain". Of course, information processing needs to be defined itself in order for this definition to have value. I propose the following: information processing refers to the process by which states of known semantic value (we know what the state is) give rise - potentially through unknown intermediate states - to new states of known semantic value. To illustrate this, I would suggest that the turning on of a light is a form of information processing. The light initially has a state known to be "off". Once the switch is flipped, that state (known to be "a flipped switch") will then deterministically give rise to a new known state in the light, namely "on". Neuroscience asks questions regarding the steps that take us from known state to known state. For example, neuroscience attempts to find intermediary states between our physical sensory inputs (known states of "photon reflection", "waves in air", or at a higher level, "stuff is there"), and the resulting state which is our internal representation of our surroundings.
ReplyDeleteHow can cognitive science use this information? By defining mechanisms of moving from known state to known state in a deterministic way, neuroscience can teach us general principles about how to engineer systems which can deterministically generate the complex state (which is the "do" in the "why we do what we do") that defines cognition. To use Searle's term, I feel that the questions of neurophysiology today are part of a system of "weak" neuroscience - an attempt to understand the brain so as to help create our own models, rather than itself reconstruct our cognition.
This is where I feel that there is value in mirror neurons. They define an intermediate state which could provide a method for transforming sensorimotor information into understanding through the use of states which are triggered both by perceiving and by preforming action. Certainly overlapping patterns of fire in mirror neurons between two related situations is not understanding: I could create a similar system using light switches. What it is is a building block towards understanding - a potential way to define a deterministic relationship between our sensory transduction, and an understanding of what to do and when.
2- even if it had done so I don't believe it would have been worth hurting other living beings especially ones that are able to feel. I think what we are doing as a race can be termed "speceism" it is an immoral abuse of our "power". We have absolutely no right whatsoever to take animals from their natural habitats, young ones from their mothers, put them in cages and cause them unnecessary harm and all that for the benefit of "science". By doing what we are doing we are disturbing the natural equilibrium of this planet which we call our home but it is not just our home, we are at an age where we have to learn how co-habitate with everyone else on this planet, we all have the right to live here as equals.
ReplyDeleteWith regards to the importance of the mirror-neuron system to communication, and gestural communication in particular, Rizzolatti et al. state: "Mirror neurons represent the neural basis of a mechanism that creates a direct link between the sender of a message and its receiver. Thanks to this mechanism, actions done by other individuals become messages that are understood by an observer without any cognitive mediation."
ReplyDeleteWhile the neurophysiological evidence of mirror neurons, and the details of their function and operation, seem undeniable, this line of reasoning and its implications seems to me to be a gross jump in the interpretation thereof. It is of course appealing to ascribe the mirror-neuron system to some direct communicatory mechanism, which would be the predecessor of language—that property which is often thought to demarcate humans from other organisms—but that it creates a uniquely "direct link between the sender of a message and receiver" seems to be romantic overgeneralisation when in fact, it is the observer that is ascribing intentionality to a particular gesture (and we have seen that intentionality, derived from contextual understanding of a particular action, is important to the response of mirror neurons). I would not argue against some form of communication arising through this manner—through a multitude of potential misunderstandings due to individual interpretation, but arise nonetheless—but the implied bilaterality of this initial simple interaction that would be required to transmit information between to agents, and not merely between movement of a muscle and neuron, is nonexistent. Mirror neurons may be indicative of how and when we ascribe intentionality to an action, but are insufficient to account for the direct bilateral links required by any system that would call communication its primary purpose.
Movement vs Meaning
DeleteDo mirror neurons so far go any further than being one correlate and component of our capacity to recognize someone else's movement as the same a movement I can make? Analogue movement recognition.
This is probably a part of our capacity to imitate movements, but language?
Is my being able to say and mean "the cat is on the mat" and my ability to understand what someone else means when they say "the cat is on the mat" recognizing someone else's movement as the same as my own?
It might be if he were singing in Chinese, and I could imitate him.
But language? and meaning? and understanding?
I agree with what Audrey says! The final sentence of that passage, "Thanks to this mechanism, actions done by other individuals become messages that are understood by an observer without any cognitive mediation,” leaves me asking questions. Without any cognitive mediation, to my understanding, means without any further processing - this doesn’t make sense to me.
ReplyDeleteI read Fodor’s paper before reading this one, and so I was struck but the abundance of indirect evidence supporting the role of mirror neurons in cognition processes. To answer Prof. Harnad’s question posted on an earlier thread - no, I don’t think finding neurons that fire when we imitate doesn’t explain how the brain causes our capacity to imitate (at least this paper doesn’t).
umm, did you mean you don't think it doesn't or you don't think it does?
DeleteMirror neurons as the basis of learning could have huge implications on cognitive science and robotics. While obviously it is not possible to reverse engineer the brain, mirror neurons could be the key to developing a robot that could pass T3 and after some more understanding, T4.
ReplyDeleteUsing the idea of imitation learning, it is currently possible to build a robot that could mimic some speech or some motions. Incorporating a degree of entropy into the response system such that the responses are no longer predictable could be the beginnings of the answer of how to convince the judges during the test.
Furthermore, using this knowledge, teaching machines to learn could be the next step. This is already done with certain algorithms where you enter an input and then it generates a response that only it could figure out and extrapolate on but this could be applied to more "human" operations like language acquisition and skill building.
Professor,
ReplyDeleteIn the video at the top you said J. Fodor had too hastily dismissed brain imaging.
He doesn't think brain imaging has told us anything new about how the brain works because the question of "where" has nothing to do with "how". How could learning the where of human cognition tell us anything about how human cognition works?
If our problem with this article (and moreover the practice of locating neurons for cognitive capacities) lies in its purely correlative form, I wonder if we will ever be satisfied. Causal mechanism can't be found, they are thought up in connection with correlative findings. They are put simply, ideas. What is lacking might not be an empirical proof of the causal mechanism for cognition in humans, but instead a satisfying theory which explains how mirror neurons cause thinking. Causality is applied to significant correlations by crosschecking it with other correlations and by coming up with logical reasons why two (or more) things co-vary in the way they do.
ReplyDeleteThis comment has been removed by the author.
ReplyDelete“The incredibly confusing organization of Broca’s area in humans, where phonology, semantics, hand actions, ingestive actions and syntax are all intermixed in a rather restricted neural space, is probably a consequence to [the aforementioned] evolutive trend.” (186)
ReplyDeleteAfter reviewing Rizzolatti and Craighero’s in review of mirror-neuron systems, it seems as though one might be drawn to arguments advocating for its extreme locationist perspective, implying that distinct neurons, in distinct locations in the brain, might underlie all semantic-syntactic transformations and might therefore underlie the development of communication systems. The authors steadily reference distinct anatomical/cito-architechtonic areas that become activated during instances in which the action understanding/imitation learning criteria of mirror-neurons are fulfilled. This is where, however, a skeptic might criticise the authors for their inability to highlight a causal mechanism for cognition that might satisfy T3 capacities. Although I cannot support the locationist myself, I have yet to see why empirical findings of the “mirror-neuron system” (with emphasis on system) can’t be a stepping stone in understanding cognition. The authors’ primary evidence comes from correlational techniques that assess the interactions of brain activity both spatially and temporally. Although correlation is not causation, it provides empirical data that might be used to develop hypotheses concerning 3rd party players in a large system of interacting brain regions. I can’t stress enough that the regions might be nowhere as important as the dynamical means by which the system interacts (according to the temporal or spatial forms of the messages the neurons send). The reverse engineering of these dynamic properties (and not a reconstruction of the locations themselves, which may or may not be an epiphenomenal bi-product of evolution), through simulation, or computer modeling, may be enough to satisfy T3 capacities. As imaging, or electrophysiological techniques become more spatially and temporally resolute, how can we say that possible cell-assembly systems, which we must reverse engineer, will never uncover, with more certainty, the potential means by which sensory-motor transformations help provide meaning to symbols, and language processing?
With regards to experimentation on animals:
I don’t have any good evidence for why or why not animals should be tested on in cognitive science research. However, the moral standard for hurting animals would ultimately rely on whether they are sentient enough to feel, and whether we have a strong enough reason to believe this is so. It seems to me that the best we could do to find this out with certainty, would be to perform some version of the Turing Test in order to reverse engineer animal cognition. I can only rely on a deep intuition that this is possible. Furthermore, it is clear that animal testing, more often than not, is not a utilitarian means for saving lives, but rather a matter of curiosity for academic researchers. In instances when determining the mechanism necessary for cognition, I would hope that functional imaging (on humans), using non-invasive techniques that do not involve the hurting of animals, is a strong enough means for contributing to the construction of T3 passing models.
Mirror Neurons
ReplyDelete"This fact suggests that a necessary step for speech evolution was the transfer of gestural meaning, intrinsic to gesture itself, to abstract sound meaning. From this follows a clear neurophysiological prediction: Hand/arm and speech gestures must be strictly linked and must, at least in part, share a common neural substrate. [...] A number of studies prove that this is true." p.16
At the beginning of the article, the author noted that the section on language is very speculative. When reading the discussion on mirror neurons and language, however, I got the impression of assertiveness in the truthfulness behind the findings. I feel like the passage I quoted above captures the sense of assertiveness in what was initially admitted to be speculation. Often in reading research articles it becomes apparent that the author may have some degree of bias in how the data is presented. Especially in pop science magazines, the findings are often sensationalized and the colored brain images seem to be meant more for aesthetic appeal than for analysis.
The paper shows the evidence suggesting the existence of mirror neurons in other primates and human. Mirror neurons are the neurons in our brain that enable us to understand the actions of others and to learn by imitation. For example, learning to drive by observing the demonstrator.
ReplyDeleteI can see how it is exciting for researchers to discover mirror neurons. It is like Columbus first discovered North America. Except the map is on our brain. But: 1. It is not the discovery of new cognition ability; the discovery of mirror comes after we realize that we have the ability to imitate. We knew that we can imitate, research had been done on imitation before discovering mirror neuron. 2. It did not helped cognitive science, cognitive science tries to figure out how it works, where it located does not matter.
Admittedly, discovering our brain map is important for areas such as clinical neurology. Another benefit would be it satisfied our curiosity, I had taken a course in Brain lesion and the information is intriguing. But neural science is always a controversial subject, especially before neural-imaging. Unlike other subject where you can observe easily, there's no other way to study brain than damaging part of it. Because how inhumane this is, research of neural science are mostly done on animals. It is cruel. I can only say that with the help of new technology such as neural imaging. We can study the brain without damaging it.