Cook, R., Bird, G., Catmur, C., Press, C., & Heyes, C. (2014). Mirror neurons: from origin to function. Behavioral and Brain Sciences, 37(02), 177-192.
This article argues that mirror neurons originate in sensorimotor associative learning and therefore a new approach is needed to investigate their functions. Mirror neurons were discovered about 20 years ago in the monkey brain, and there is now evidence that they are also present in the human brain. The intriguing feature of many mirror neurons is that they fire not only when the animal is performing an action, such as grasping an object using a power grip, but also when the animal passively observes a similar action performed by another agent. It is widely believed that mirror neurons are a genetic adaptation for action understanding; that they were designed by evolution to fulfill a specific socio-cognitive function. In contrast, we argue that mirror neurons are forged by domain-general processes of associative learning in the course of individual development, and, although they may have psychological functions, they do not necessarily have a specific evolutionary purpose or adaptive function. The evidence supporting this view shows that (1) mirror neurons do not consistently encode action “goals”; (2) the contingency- and context-sensitive nature of associative learning explains the full range of mirror neuron properties; (3) human infants receive enough sensorimotor experience to support associative learning of mirror neurons (“wealth of the stimulus”); and (4) mirror neurons can be changed in radical ways by sensorimotor training. The associative account implies that reliable information about the function of mirror neurons can be obtained only by research based on developmental history, system-level theory, and careful experimentation.
“It could be argued that, since [MNs] were “specially created” by evolution, MNs are likely to have a highly distinctive, largely independent, and previously unrecognized psychological function. In contrast, by showing that established psychological theory – associative learning theory – can cast light on the origin of MNs, the associative account underlines the value of embedding research on MN function within system-level psychological and computational theories of how the brain produces behavior” (Cook et al 2014).
ReplyDeleteThe associative-model hypothesis for mirror neurons (MNs) is argued rather strongly in this article, for reasons including their inability to encode for “goals,” their adaptability, that MNs do not engage in prediction, and that sensorimotor stimuli are enough for infants to learn. This hypothesis is used mainly to counter the genetic hypothesis of MN origin, which argues that MNs are not “specially created” by evolution. What strikes me most in the quote above is the sort of ‘reverse-engineering’ (not the good cognitive science kind) that is being done with psychological theory: the associative learning model, which is “well-established” is being deconstructed to explain the origin story of MNs. It seems as though to study the phenomenon of MNs, neuroscientists tried to neatly pack their function into a pre-existing theory. The origin and function MNs, as Cook et al emphasize throughout their paper (citing a plethora of studies) is complicated and could be multi-disciplinary -- that is, assuming we will ever be able what MNs ‘do’ or where they ‘come from’.
I find more compelling the speculation of other approaches to study MNs which are informed by (not constrained by) associative learning -- Cook et al advocate for this too, the sort of “hybrid" approach that takes into account an evolutionary perspective and genetic perspective as well as systems theory/computational neuroscience.
All scientific inquiry is nested in a historical background and the work of thinkers past; it is impossible (and quite ignorant) to avoid the application and guidance of previously established theory. Yet it seems to be a trend in science to over-contextualize studies -- to try and neatly ‘fit’ the ‘answers’ to questions into theories that have already been established. This seems to be the case with MNs and associative-learning theory. The authors advocate for a more broad exploration, which I strongly agree with.
Much of the investigation around MNs cannot ‘fit’ into any previous theories because much of the questions posed are on the basis of *feeling*: understanding, intentionality, and goals (and, as we’ve talked about in great length, no theory exists that will give us any answers...)
We all know that humans (and other organisms) have "MN" capacity: We know when someone else makes the same movement we do. The interesting question is not whether they are born that way or learn it or both. The interesting question is how does their brain do it? Answer: we don't know, but it certainly can't be done by associative learning. Finding neurons whose activity correlates with when we are doing it does not explain how we are doing it.
DeleteI agree with Gili that it seems like “neuroscientists tried to neatly pack their function into a pre-existing theory”.
DeleteIt first seems like Cook et al. aim to undo the error that what was done with the genetic explanation. This error is to fit something we can’t explain into a pre-existing theory, here natural selection, to feel like we solved the problem while not answering the real question of that problem. Here the real question is how we (our brain, or some neurons) have this mirror capacity.
Cook’s account of associative learning does not answer that question either. Rather, it resorts to Hebbian theory to avoid answering the real question.
This passage clearly states Hebb’s rule and spike-timing-dependent plasticity (for kid-sib: what fire together wire together) :
“The kind of learning that produces MNs occurs when there is correlated (i.e., contiguous and contingent) excitation of sensory neurons and motor neurons that code similar actions… Correlated excitation of the sensory and motor neurons increases the strength of the connection between them, so that subsequent excitation of the sensory neuron propagates to the motor neuron”
The problem is that Hebb’s rule refers to a molecular process, and the authors try to apply it to a cognitive one. If we consider the famous Pavlovian dog, it doesn’t understand why it salivates when hearing that bell sound. So yes the connection between two neurons of different modalities (visual and motor, visual and auditory...) can be strengthened via associative learning, but that does not explain motor neuron capacity. It only explains the “wiring”.
“genetic account assumes that they play a fundamental role in action under- standing, and that this is why a specific genetic predisposition to develop MNs was favored by natural selection.”
ReplyDelete“If MNs were a genetic adaptation, it is likely that their properties would be relatively invariant across developmental environments.”
“In its starkest form, the genetic hypothesis would suggest that gene-based natural selection has provided each individual monkey and human–with MNs that code the mapping between a fixed set of observed and executed actions, and that experience plays a minimal role in the development of the observation-execution matching prop- erties of these neurons.”
Does the dynamic power of MNs, which seems essentially unobjectionable, exclude the possibility that they are in some sense generated by a genetic process? Is the notion that mirror neurons are the implementation of a hard coded “mapping between a fixed set of observed and executed actions” not separable from the genetic hypothesis? The edges of this genetic/associative distinction are loosely defined here, though it’s surely intentional given the author's assertion that this is not a question of nature vs. nurture but of the “interaction of nature and nurture”.
It seems unlikely that the authors deem both approaches incompatible, but rather that they’re advocating for a redirection in the path of experimental research; it’s not so much helpful to draw a hard line in the sand here as it is to point in a new direction and ask that we try going that way. If the push that Cook et al makes here is just that further research should respect that mirror neurons aren’t derived from some “fixed set” and absolutely hold up to an infinity of new scenarios, there’s not a ton of room for objection. Because mirror neurons obviously occur across humans everywhere (& across species), it’s absolutely necessary that our understanding of them is dynamic enough to hold up across cultures and beyond.
You are right that "innate or learned?" is a red herring. That's not the relevant question for cogsci (reverse-engineering) when it comes to our mirror capacity (including imitation). The question is how. And "associative learning" certainly does not give the answer.
DeleteI am not sure about how the genetic account view experience in terms of the genetic learning. “In its starkest form, the genetic hypothesis would suggest that gene-based natural selection has provided each individual – monkey and human – with MNs that code the mapping between a fixed set of observed and executed actions, and that experience plays a minimal role in the development of the observation-execution matching properties of these neurons.”
ReplyDeleteHere I do not understand, in the genetic account, how the gene-based selection implies the minimal role of experience in mirror neuron. Even if the mirror neuron might have occurred because of the genes (exist at birth), but does there exist a possibility that experience also reinforces the synapses and then the strength of neural connections between neurons that plays a crucial hole in the function of mirror neuron? By analogy, in terms of height, it is heavily determined by genes, but people can well argue that experience (later environment that you grow up in, e.g. nutrition, childhood lifestyle) play an important role in determining the height.
In this paper, I think the example the authors give is trying to say that even if motor neurons are genetically determined, there is still the need for motor experience to play a role in the maturation of mirror neurons. I don't think that in this section they argue that there can be stronger connections or synapses but that there is a facilitation and preparation that requires experience.
DeleteJust as with Fodor (4b), who asks whether learning when and where something happens in the brain can reveal how and why, the relevant cogsci question about mirror capacity is how and why the brain produces it, not when and where (nor whether it is innate or learned) but how.
Delete“Associative learning is found in a wide range of vertebrate and invertebrate species, indicating that it is an evolutionarily ancient and highly conserved adaptation for tracking predictive relationships between events.”
ReplyDeleteSpeaking on predictive events and out of curiosity, there have been some studies done where experimenters will begin by performing an action and then will perform something completely unexpected. I’m just wondering how the authors would classify this with mirror neurons and if the mirror neurons would fire in anticipation for the second action which does not follow.
For reverse-engineering of mirror capacity, "associative learning" is not an explanation. It is an explanation for our capacity to learn the pairing of nonsense syllables and the like.
Delete“[The associative account] suggests that MNs acquire their capacity to match observed with executed actions through domain-general processes of sensorimotor associative learning, and that the role of MNs in action understanding, or any other social cognitive function, is an open empirical question.”
ReplyDeleteMirror neurons (neurons that are activated both by a motor action and the visual perception of this motor action) are a big trend in neuroscience (and cognitive science). This paper argues against the genetic account of mirror neurons that is proposing a specific evolution of neurons with such properties because it gave a specific advantage for humans to recognize the actions performed by fellow humans. The authors then argue that mirror neurons are the result of general “associative learning” between visual and motor stimuli. They argue (among other reasons) that very few cells seem to have exact “mirror” property and that it is possible to reverse the effect of mirroring by counter sensorimotor training (i.e. create new pairing between arbitrary stimuli that will counter the previous ones). According to the argument, associative learning would be better to explain and predict phenomena related to “mirror neurons”.
Some authors in the open-peer commentary are pushing back against the idea of “associative learning” being the only necessary factor for the production of “mirror neurons”. Instead, they argue for a partly genetic account. They base their claim on the specific importance of hand-object association and the fact that human infants (9 months) seem to recognize human actions more easily and non-typical human action, a fact that is not well explained by a full associative account.
Indeed, such neurons could provide some insight for modelling multiple social phenomena (from imitation learning to language acquisition), even though the question of the “where” or “what” inside the brain does not, in itself, provide an understanding of the “how” the brain does it (to say that “the neurons are able to do capacity X” is not a proper explanation of the capacity X). It is not to say that the brain is completely irrelevant as computationalism said, but more that the brain is not the explanation **in itself**.
---
As a side note, I am not sure to understand Prof. Harnad’s claim that “innate and acquired” do not matter in cognitive science. Since human are not static entities (neither during development nor at adult age), wouldn’t be important to understand which features of the cognition are environmentally dependant (and what are the underlying “innate” features) in order to reverse-engineer them? It could be that cognitive science only must provide a mechanism such as it would develop exactly as a human would do in any given environment. Still, I am under the impression that understanding which properties are the “basic/innate” ones would help in this matter.
It is not that innate vs. learned does not matter in cogsci: it is only that is does not answer how.
Delete"As a rich source of testable predictions about when, where, and how MNs develop, associative learning theory can provide firm guidance for future research on the taxonomic distribution, typical properties, and functional roles of MNs."
ReplyDeleteI totally agree that associative learning is a more testable theory comparing to the genetic hypothesis. And I also agree from some of (or in fact, most of) the evidence provided by the authors, associative learning indeed seems to be a more plausible account for the origination and development of MNs, probably because associative account is much broader than the genetic one -- it in fact does not exclude the evidence that the genetic hypothesis tries to explain, but incorporates, though loosely, most claims that the genetic hypothesis holds, and accommodates more in the mean time.
But as Prof. Harnad pointed out above, knowing "when, where, and how MNs develop" and even "the taxonomic distribution, typical properties, and functional roles of MNs" does not necessarily give us the answer of how they do what they do in our brain. What we are trying to explain in cognitive science is how the brain performs, which is a problem that I don't quite see how all the arguments and conclusions in this paper would help to solve, at least directly. However, I could imagine that with more and more knowledge about the (biological, evolutional, not sociological) development of MNs, it might start to shed some light on how it gets "the capacity to match observed and executed actions" step by step.
No doubt someone will eventually come up with a causal model of mirror capacity and imitation. But is associative learning potentially a causal model for mirror capacity and imitation?
DeleteIn their paper, the authors argue that MNs originate in sensorimotor associative learning, refuting the widely held theory that MNs are genetically adapted for action understanding. While the authors propose a refined theory of how MNs ‘came/come to be’ (through learning rather than natural selection), there is no mention of how exactly MNs work - the causal mechanism that could explain their function. We currently have no clue about the exact mechanisms behind our capacities to imitate other people’s behavior or to recognize others’ emotions and intentions, we just know that we do. Whether MNs arise through genetic adaptation, canalization, exaptation, or associative learning doesn’t really matter, since the MNs’ origin story doesn’t clarify the causal mechanism by which they act. To be fair to the authors though, they do argue that “the associative account implies that reliable information about the function of mirror neurons can be obtained only by research based on developmental history, system-level theory, and careful experimentation”, so maybe the knowledge about MNs’ ‘true’ origin can help scientists focus their efforts more effectively, such that one day we may be able to shed light on some of the causal mechanisms underlying cognition?
ReplyDeleteYou are right to ask "how?" rather than "inborn or learned." Is "associative learning" a potential explanation? We know we have sensory and motor maps of our bodies, probably mostly innately. Is this "associated" somehow to others' maps of their bodies by learning? How? And what about self-recognition in the mirror? These are all challenges for robotics.
DeleteWhile reading Cook et al.’s paper, I had the same reaction that Behme did in her response commentary. It seems like Cook et al. are really refuting a straw man genetic hypothesis for mirror neurons (MNs). Their argument presupposes that the genetic hypothesis for MN involves a strict genetic predisposition of what kind of stimulus the MN will respond to. They go on to describe findings from studies showing that new sensorimotor experience is enough to alter the response of MN, which does not support the genetic hypothesis.
ReplyDeleteHowever, I would have assumed that it was implied within the genetic hypothesis that environmental experience would be required for these MN to “learn” which stimulus they will respond to. Similar to Behme, I had also thought of language as the perfect analogy to demonstrate how cognitive capacities that have evolved throughout human evolution require a certain flexibility in response to an altering environment. That is, while language is something that all humans are genetically predisposed to developing, the stimuli to which humans are exposed in their environment will ultimately decide which language they will speak, however, at birth, all humans are equally well-equipped to acquire any one language.
Moreover, it seems like Cook et al. were splitting hairs by proposing the associative account as a competing hypothesis to the genetic account. These two accounts can almost intuitively be consolidated into one account whereby MNs have a genetically predisposed framework upon which input from the environment will be required to give the adaptation its form through associative learning.
Although, the authors’ do respond to the commentaries of others. They noted that they did not intend to argue that the genetic account means that learning does not play a role in the development of MN. They also state that the fact that MN respond maximally to unnatural stimuli, makes it difficult to reconcile it with the genetic hypothesis of MN. Although, I’m not sure I fully understand the reasoning used, it seems that the authors were stating that the genetic hypothesis implies that responses to unnatural stimuli link sensory neurons with other sensory neurons or motor neurons with other motor neurons, whereas maximal responding of unnatural stimuli implies a connection between sensory neurons and motor neurons. The authors did not go on to elaborate further, which is a shame, because I don’t think this really cleared up any confusion, not for me at least.
Evolution is lazy. (This will come up often in this course.) Evolution does not encode genetically any more than necessary for successful survival and reproduction. It "offloads" as much as it can on predictable environmental influences. Your language example is one. (No need to encode any particular language: just the capacity to learn any of them from experience and interaction.) Learning itself is the best illustration of evolution's laziness -- and language is probably its most powerful example, because once you've learned language, you can learn just about everything else through language. (That's also why T2 is still such a powerful version of the Ttest.)
DeleteThe inadequate explanation of why MNs respond maximally to unnatural stimuli is just another reminder that no one has given a causal explanation of how MNs actually produce mirror capacity and imitative capacity. "Association" is just hand-waving.
You might find it fun to google the "uncanny valley" and then look it up in google scholar.
(I was on Karl MacDorman's thesis committee. He originally started out working on the symbol grounding problem, but he ended up in Japan working on the uncanny valley. Can you trace the route?)
So, it seems like the uncanny valley is the “creepy” feeling that we get when we see modern day humanoid robots closely resemble human behavior, but not quite perfectly. An anthropogenic robot induces the uncanny valley only if it is reasonably close to human appearance and behavior, if its too far off we would not get an eerie feeling, and if it were indistinguishable from human appearance and behavior, we would not get an eerie feeling. The reason for this might be rooted in the symbol grounding problem. The symbol grounding problem involves more than just the association of the meaning of a word with its written symbol, but it can also involve how we get our perceptual representations of physical things we encounter through our senses. So, from what I was able to gather, it seems like MacDorman figured that the reason we get an eerie feeling when seeing a humanoid robot imitate human behavior is because it causes us to have conflicting perceptual representations of humans and objects. This is based on the physical symbol grounding problem and social symbol grounding problem, I think. It’s is a confliction between our social symbol association (our perception of humans and our own identity) and our physical symbol association (our perception of inanimate objects). This causes our human identity to feel threatened which causes the eerie feeling I believe.
DeleteI don't know if I am correct though?
I don't think Karl (or anyone else) has any idea why near perfect robots look creepy. (I kind of doubt it has much to do with symbol-grounding, though it might have something to do with turing-testing and mind-reading.
DeleteHarnad mentions above the ‘uncanny valley’. After reading the paper, I instantly thought of the work of art titled ‘Female Figure’ by Jordan Wolfson. Definitely worth a watch… when I saw this in person, I was initially struck by curiosity and attraction, which almost immediately changed into eeriness and discomfort as I approached the robot. Aesthetical appearance of the machine is a perplexing problem facing engineers and designers. I agree with Johnny that this eeriness could in part stem from the battle between our representation of inanimate objects and animate beings. However there are so many possible reasons for humans to experience this feeling with near perfect robots. Karl even states at the end of his paper that he hasn’t figure this out. The solution to this problem may be to ‘purse a nonhuman design’ (MacDorman et al.). The example given in the paper is a wooden hand. How will this affect the Turing Test? Without creating an aesthetically perfect human robot, you are bound to fall in the trenches of the uncanny valley.
DeleteThe associative account of MNs suggests that it is important to understand the environment of the system in order to explain MN function. This argument seems to appeal to similar logic as the System's Reply against Searle's Chinese-Room.
ReplyDeleteIndeed, it does not matter where the rules are (where the causal mechanisms behind MN activity are acquired). That is distinctly not the point. The point is *that* the system is mirroring, and *how* it does so.
I can't see the link between the associative hunch about MN function and the System Reply to the CRA -- other than both sound vacuous!
DeleteHow would mirror neurons function in terms of empathy? If these researchers want to make a point out of associative learning, I think empathy would be a good example. The researchers write:
ReplyDelete"Whether or not an individual has MNs, which actions are encoded by their MNs, and at what level of abstraction, will all depend on the types of sensorimotor experience received by the individual in the course of their development."
Empathy is an example this because an individual’s ability to empathize depends on experience through development (although I am aware that some individuals are “genetically” predisposed to be more empathetic than others). Empathy is defined as the ability to understand and share the feelings of others. Feelings can be expressed in terms of motoric behaviours (e.g., crying, yelling, smiling, etc.). If someone has a lot of empathy, their associative learning between the recognition of feelings and the actions that result from them. If empathy is a result of associative learning, perhaps this could be a (probably small) contributing factor into answering the hard question of cognitive science: how do we feel when we do things?
I don't know about associative learning, but the way we recognize facial (and auditory) expressions of emotion might give a tiny clue about empathy.
DeleteHi Erica,
DeleteI was also thinking that discussing empathy would be an interesting subject to broach. Empathy is a very human trait that all neurotypical individuals possess to a certain degree because it is essential in maintaining relationships. As you mention, there seems to be a genetic component to empathy as well as a learned capacity. This paper focuses on the describing the development of mirror neurons which they believe is associated with the learned capacity rather than genetics. Although it is interesting to learn how mirror neurons come to exist does not help us understand how the mind works. However, if empathy relies on mirror neurons then it would help us answer the hard question of cognitive science because we are understanding a mechanism that allows us to interpret other people’s feelings.
As a side note, this would mean that our robot version of Gabe would need to be able to empathize to pass T3. Therefore, he would have some mechanism similar to mirror neurons built into him which would be a challenge in in of itself to integrate even if we figured out the neuroanatomical mechanisms.
We already knew people could tell when others were making the same movement as themselves. We already knew (most) people could empathize. So we already know the brain could do both (but not how). Do MNs tell us how? Can we now build T3s that empathize by sticking MNs into them? What does that mean? How do you do that? and how do MNs do whatever they do?
DeleteAll this to say that MNs not only do not tell us how the mind works, they don't even tell us how MNs work!
So how can MNs answer any "hard questions" about empathy that we did not already know?
Gabe would only have to do anything a "neurotypical' person can do (including do what someone who empathizes does). But that's all easy-problem stuff. No inroads on whether Gabe feels (the other-minds problem) nor how and why (the hard problem), whether the feeling is what it feels like to be seeing red or feeling empathy. (Think about it...)
I have been struggling to express my unsatisfaction with the authors and the claims they make about understanding behaviour. The biggest problem with this article is, obviously, the blurred lines between correlation and causation, which I feel the authors have walked on the edge of, but the real problem is, like prof Harnad said above, "that MNs not only do not tell us how the mind works, they don't even tell us how MNs work". The biggest issue I see in this article is that the authors make grand claims about MNs, although they introduced the article acknowledging that only one study had been conducted that recorded bimodal activation through single cell recording. There is almost nothing that we know (or so it seems) about what one MN acts like – making it impossible to pretend we know anything at all about how populations of MNs act like, and how they affect a human on a behavioural level. There is probably something to be said about the pressure that's put on the authors to publish something that has meaning in the real-world, but in this case the claims are not mere exaggerations of the significance of the findings, but really projections of what MNs could be, based on a lot of hope and a very limited actual causal understanding of the MNs.
DeleteI also was intrigued by the idea of using mirror neurons to study empathy, as there have been multiple studies conducted regarding a person’s neuronal reaction to watching someone else in physical pain. These studies have shown that the neurons (mirror neurons?) in the observer’s brain fire in the same place as in the injured person’s brain. But what about social/emotional pain? (I work in a lab that actually studies this!)
DeleteAlthough I do think that MNs are an interesting phenomenon, I think there is still a lot to be uncovered about them and their role in social cognition and learning. They don’t offer any insight that will allow us to reverse engineer cognition based solely on the knowledge of their existence.
I agree with Marianne when she argues that the article blurs the lines between correlation and causation. For this reason, we cannot jump to conclusions about the role MNs play in our learning and behavior. In this case, Fodor might be right in asserting that functional localization and neuroimaging studies cannot deliver concrete answers to the actual questions we’re asking.
MNs are a big research fad in neuroscience these days, much like machine learning in computer science. This isn’t necessarily a bad thing, but I do think there could be much benefit in exploring other means of researching this topic.
"However, the picture of action understanding is obscured by the fundamental problem of defining exactly what is meant by “action understanding” and how it differs from action perception."
ReplyDeleteI would argue that the difference between action perception and action understanding is that there is a level of “feeling” in action understanding that action perception does not have. To perceive an action is to recognize that someone is doing something (maybe the authors think otherwise). But understanding (the goal or intention behind) an action is a feeling. For example, if my friend picking up an apple, I am perceiving/seeing that they are picking up that apple. But to understand why they are picking up that apple is a feeling: I feel that they are hungry. I say “feel” because I can’t know and I wouldn’t know unless I am them.
Maybe behavioral contagion (infectious yawning, laughing and aggression) is another tiny clue.
DeleteAnd mirror self-recognition...
The idea of self-recognition in a mirror is actually really interesting in the context of Erica’s comment. I don’t know that I agree that what the difference between action understanding and action perception is feeling (other than the fact that I feel like I understand), but self-recognition does have something to do with the concept of the self.
DeleteThat concept is actually pretty interesting if we’re talking about action understanding, because even when it comes to language understanding (which we agreed in class was the further we could take action understanding), you need an understanding of the boundaries of the self VS others. This is pretty far from the topic of this article, but at the same time, the activity of mirror neurons does have something to do with understanding that an “other” is performing an action, and that the action is something you (the self) can perform as well. Also, nothing about the way computers are programmed tells us that they have a “sense of self” in this way. I wonder if this is another critical difference that makes computationalism false, and that adds relevance to mirror neurons.
To take this a step further, I'd liken action perception to recognizing the action but not being able to faithfully execute it. For example, if I see someone doing a standing backflip, I certainly cannot feel that nor can I execute it myself, but I can perceive what was being done to accomplish that and derive that feeling, but I cannot feel it myself (at least without a lot of practice!).
DeleteThis paper focuses on two hypotheses for the origin and function of mirror neurons (MNs). The genetic hypothesis claims that genetic evolution plays a central role in the development of MNs, and sensorimotor learning plays a facilitative role. The associative learning hypothesis (the hypothesis that the authors advocate) claims that sensorimotor learning plays the central role in MN development, while genetic evolution plays a non-specific background role. The article analyzes the pros and cons of each, previous studies that have investigated MNs, and hybrid hypotheses. The information and conclusions discussed in this paper analyze one of the easy questions of cognitive science - why we can do what we do? The paper looks at why MNs fire when observing or performing actions and why these neurons developed in this fashion. The question the authors do not attempt to analyze is how do we do what we do? How do these MNs produce the effects that they do? How do MNs “decide” what is important to encode and what is not? What are the causal mechanisms behind MNs? For cognitive science, I believe the how questions are more interesting to investigate than the genetics, evolution, or designated function.
ReplyDeleteI don't even see how MN's answer the why question!
DeleteRe: material discussed last class
ReplyDeleteThough this was said to be discussed more extensively later in the semester, the Symbol Grounding Problem came up as a topic of contention. What assigns meaning to symbols? And to what extent do those meanings have to be assigned for an entity to "understand"? In the case of the Chinese Room Argument, the man did not understand any of the Chinese characters, proving that cognition is not merely the computation or manipulation of symbols. We also said that understanding and consciousness is really just a "feeling". It feels like something to eat a strawberry, experience emotions, form relationship -- to understand. Since we may not be able to determine whether or not Gabe can feel, I would like to explore whether a human can be "un-feeling" neurologically.
In another class, we read about a patient who suffered an interruption in their "what" visual pathway; terminating in the temporal lobe, the pathway is responsible for identification of objects, semantic memory, emotional responses, and more. The patient who experienced a lesion here no longer experienced any feelings associated with objects. They could not recognize anything, as if "the entire world…[was] like a gallery of abstract sculptures." (Ramachandran, 78)
I believe that this case presents striking similarities to a machine that cannot understand the world. Just as the patient could not assign feeling or meaning to anything, the man in the Chinese Room and (weak AI) machines cannot assign meaning to the symbols. "One could argue that the term consciousness does not mean anything unless you recognize the emotional significance and semantic associations of what you are looking at." (Ramachandran, 78) I guess it would feel like something to have your "what" pathway damaged and thus not able to identify or have emotions associated with objects and people, but would Gabe not have the same experience? If consciousness is "feeling", how exactly would we define the parameters of a feeling?
The SGP is not really about what "assigns" meaning to symbols, but about how a symbol system (e.g. T3) knows the referents of its words.
DeleteUnderstanding is not just feeling! It's T2 capacity with words plus T3 (grounded) capacity with words... + the fact that it feels like something to understand (hard problem).
Gabe is a healthy T3. A severely brain-damaged person -- e.g. one with severe object agnosia -- is not. But they still feel some things, and can still do some things, even speak and understand (unless they are phasic too).
If I'm humouring the idea that Gabe is like the man in the Chinese room and does not "feel" (I am skeptical of this idea), I think Gabe and the patient you described above are still having explicitly different experiences; that is to say, Gabe is not having a 'conscious'/feeling/mental experience, and for the patient, it still feels like something to be confronted by a variety of abstract objects that he/she/they doesn't understand.
DeleteAs the professor has said in class, understanding and not understanding are distinct feelings, and it still feels like something to not understand the meaning of symbols or shapes.
The only idea to humor is that Gabe is an MIT T3 robot. Whether he (or anyone but oneself) feels is not something we can know for sure. but we are almost sure about other people, and we can't tell Gabe apart from other people, for a lifetime. The question is whether we feel that Gabe feels, and if not, why not?
DeleteThe conclusion that Gabe does not feel, because he is an MIT robot, is completely arbitrary. We have no way of knowing whether it's true or false. So there is no basis for the comparison with an agnosic/aphasic human,
You misunderstood about understanding. It feels like something to understand and it feels like something to not understand, but it does not feel like anything to not feel anything at all (like a rock).
Responding to an idea brought up in class about symbol grounding:
ReplyDeleteIt was stated in class and previous readings that language has the power to refer to and describe anything, which is a big leap ahead of what can be accomplished through pantomime and imitation. Without thinking about it carefully, I found myself tempted to reject this idea.
Upon further consideration, of course its true (there are infinite possibilities for symbols and combinations of symbols that can be assigned uniquely to any unique or category of experience, material reality, idea, et cetera).
I think my initial impulse to reject this idea was guided by the fact that there are some concepts that I think I could never really understand, no matter how well they were described to me, without directly experiencing them. (For example, I don’t think I will ever truly know what it’s like to lose a loved one no matter how many well-written novels and memoirs I read about grief - even if I am able to understand elements of it). However, I think that this is a symbol-grounding problem, rather than a limitation of the system of language.
A student asked in last week’s lecture how much symbol grounding one needs to be able to use language description to understand concepts/objects not directly experienced through sense perception. I’m really interested in this question, and it’s my hunch that the answer is that it’s somewhat continuous, (re: my frustration at not being able to fully understand grief). Surely there is no ‘magic amount’ of symbol grounding that is enough to unlock the total descriptive power of understanding through language.
I guess my point is point is obvious – it’s my hunch that there is no limit to symbol-grounding such that we are able to take advantage of the maximum potential of language for understanding.
The only limit to the power of language is that you have to get enough of it off the ground through direct sensorimotor experience with enough named objects (categories) so that all the rest can be described through recombining words (already directly grounded, or grounded indirectly) to describe and name all other objects. "objects" are any referent of a content word, whether "concrete," like apple or "abstract," like "justice" or "truth." But of course all categories are abstract to some degree, in that they are not individuals but kinds things (apples, reds. zebras, peekaboo unicorns). We'll get to this later in the course.
DeleteThe grief at the loss of a loved one is hard to imagine if you have not felt it yet, but not impossible to describe, to an approximation. It's not like a color unlike any you've ever seen, or a sense-modality (such as bat echo-location) that you don't have, or like describing colors to a blind person -- though even those can be described to a very faint approximation out of words you already know, categories you already have, and feelings you have already felt.
This article argues in favor of an associative account of the emergence of mirror neurons. Specifically, "the associative hypothesis implies that the characteristic, matching properties of MNs result from a genetically evolved process, associative learning, but this process was not “designed” by genetic evolution to produce MNs". The evidence and supporting arguments provided throughout the article are convincing to me. An example I found really interesting and what drove the author's point in for me was the example they used of the visual word formation area, how it is an "old" region of the brain likely to have evolved before the emergence of written language, yet it plays such an important role in understanding the symbol recognition in regards to language.
ReplyDeleteI understand and agree with the article's point, but I still find it somewhat unnecessary to compare these two hypotheses. Although the authors take into account that MNs had to evolve some way, and that they're just highlighting that genetics just play a background role in the associative theory, some of their points feel very general (which is understandable, as there is still a lot to be understood about MNs). I don't fully understand why the two accounts can't just be combined into one.
On another note, the article mentions that MNs do respond to unnatural stimuli (as an argument for the sensorimotor associative learning account), but I'm also curious as to how MNs respond to robots that fall within the "uncanny valley".
Genetics + learning are always compatible, but what we need is a causal model.
DeleteIn reading others' skywritings, a few thoughts popped in my head. This paper said mirror neurons were basically necessary for action understanding. One could take that a little further and just say they're necessary for understanding, at least some part of it. I have two thoughts on this: if we placed NMs in Searle after he had absorbed all components of the room, would he be understanding what he was doing? Let's say yes. That means all we need are NMs and we can build T3, but we don't know how it works, so where does that leave us in answering either easy or hard question? I also feel like the focus on NMs is homuncular in a sense. These neurons see other people doing an action and understand it, so we're getting close. It feels like we found the room the Homunculus lives in, but have no clue where the key is, while also ignoring the entire hotel around the Homunculus' room that may hold other answers. I've ignored the nuance of NMs and what they actually do, but the point isn't about NMs. The point is that NMs don't really bring us any closer to answering the hard question.
ReplyDeleteALSO, this paper acknowledged there are weasel words in the papers about NMs, and then just continued to throw all of them around. Why? Not very kid-sibly.
Place NMs in Searle? What does that mean?
ReplyDeleteIn the Chinese Room, but after he had absorbed everything. His rebuttal to the systems reply I believe
DeleteOver the past few classes we have discussed reverse engineering as a method used by cognitive scientists in order to try and figure out the how of cognition. In class we had defined reverse engineering as having a system that works and trying to figure out how it works. Reverse engineers are looking for causal explanations as to how functional systems are working. When reading Cook’s paper, I kept questioning how studying Mirror Neurons (MN) could help cognitive scientists explain how cognition works or how the brain allows us to do what all that we do. It seems to me that MNs just confirm things we already knew we could do like determining when others are making the same movements as we are.
ReplyDeleteThe genetic hypothesis argues that humans and other living beings are born with MNs for evolutionary purposes and that sensorimotor experience has little to do with the observation and imitation properties of these neurons. This hypothesis does not help explain the how question that cognitive scientists are trying to solve. It simply states that MN are present at birth and help us with action understanding. It does not give us any insight as to how MN help facilitate action understanding.
The associative hypothesis argues that MN characteristics are developed as a result of sensorimotor associative learning by observing and executing or imitating actions similar to others. Associative learning does not help us explain how we do what we do either. The associative theory simply tells us that correlated MN are firing when we perform or observe certain actions. It does not give us any insight as to how our brains are making us do things.
After reading Cook’s article I feel as though we still do not quite understand the causal mechanisms of imitation. We know that MNs fire when we observe and imitate others… but this still does not explain how we do so. I feel as though the idea of MN seemed promising for cognitive scientists, but are in fact not that helpful when trying to solve the hard problem.
However, there is still no consensus about exactly what is meant by “action understanding,” or how it differs from cognate functions such as “action per- ception,” “action recognition,” and “action selection” (Gallese et al. 2011). Attempts to clarify have emphasized that, in comparison with purely visual processing of action, MN activity relates to the “meaning” of an action and yields a “richer understanding,” “real understanding,” or “understanding from within” (Gallese et al. 2011; Rizzo- latti & Sinigaglia 2010). As we discuss further in section 8, these descriptions do not provide an operational definition of action understanding, that is, a definition that would allow behavior based on (this kind of) action understanding to be distinguished empirically from behavior based on other processes.
ReplyDeleteIn the future, would further research about action understanding ( how mirror neurons relate to the meaning of an action) tell us more about human understanding in general and would allow the creation of T3 or T4 "understanding" systems? Has it been hypothetized that the "feeling" of understanding is performed by mirror neurons?
Cook et al aim to address the question, 'What is the process that gives MN their 'mirrorness'; their fascinating, cardinal capacity to match observed with executed action?'. Particularly, they draw from behavioral and neuroimaging data to refute the genetic account of mirror neurons (MNs) that suggests that it is an evolutionarily-driven genetic adaptation for 'action understanding'. Rather, they vouch for an associative account that argues that MNs do not have an adaptive function and instead stun from domain-general processes of associative learning.
ReplyDeleteWhile Cook et al do highlight key assumptions of the associative model that may facilitate MN's 'mirrorness' - strong experience-dependence, social construction, contingency - they ironically do not give insight into 'how' our physiology (sensorimotor and motor systems) is structured and built to respond to and modulate changes in these characteristics. For example, they argue and provide experimental evidence of why contingency sensitivity explains existence of "logically-related, audiovisual and tool-use MNs", yet do not offer any leads on the causal mechanism that allows for this predictive MN mechanism to arise. In my understanding, Cook et al attempts to overcome this gap by acknowledging a lack of sufficient research and stressing the importance of 'testability' where new predictions must be made from existing experimental data about MN's and tested under diverse experiential methods. However, I question how efficient this approach is to effectively reverse engineer a model that can carry out the development of MN's mature properties in the same way as we do. Particularly, given Cook et al's reliance on neuroimaging evidence throughout this paper (MEP, fMRI, single-unit cell recording), I wonder whether their approach will lead to an experimental emphasis on T4 neural behavioral equivalence. If this is so, this approach may be less useful to helping us understanding cognition (in a cognitive science perspective) due to an issue of overdeterminance.
“Automatic imitation is said to occur when observation of an action involuntarily facilitates performance of a topographically similar action (body parts make the same movements relative, not to external frames of reference, but to one another) and/or interferes with performance of a topographically dissimilar action. Humans show robust automatic imitation when they observe hand, arm, foot, and mouth movements.”
ReplyDeleteI agree that humans do automatically imitate actions, but humans do not imitate every action they observe. What determine what is imitated or not? And if imitation is a strictly rule-based behaviour, what do we mean by “automatic”? Or, in another word, to what extend is imitation “automatic”? Also, it is observed that imitation is not limited to movements of body parts, emotions, verbal habits, or some other higher-level behaviours can be easily imitated without being notice (“automatically”). This seems to support the fulfilment of socio-cognitive function hypothesis.
I think that I agree with the associative learning hypothesis more than the genetic hypothesis, yet we do observe individual difference in mirroring effect. For example, in language acquisition, even with the same immersion environment, there are people who are very good at picking up the phonology of a language (arguably with certain fossilization), while others, although can be fluent in this language, will still be said to have “a strong foreign accent”. Sometimes, it is reported that the difference between the sounds are not perceived, thus could not be pronounced, but it is also reported that even when the differences are perceived, the “correct” sound still cannot be pronounced.
When it comes to infants, infants are good at perceiving the difference of the sounds as a lot of research has shown, yet they are not good at pronouncing them – because the sensorimotor neurons are not mature yet. But later on most of the kids will adapt to the “correct” pronunciation – can I say the mirror neurons are modifying themselves with this example?
Cook et al.
ReplyDeleteThe implications of the associative hypothesis of mirror neurons (MN) which suggests that the matching process of MNs emanate from genetically evolved phenomenon of associative learning. The result is that “it produces MNs when the developing system receives correlated experience of observing and executing similar actions” (Cook et al. 182).
What I believe is of most interest regarding mirror neurons is that they involve a certain ‘learning’ mechanism which mimics ‘correct’ neural processes of observed behavior. While the evidence is inconclusive, I would be interested to know if there is some sort of intrinsic causal mechanism which unconsciously catalogues and yields specific reactions for specific scenarios.
Another implication – which I saw was brought up in another common is the situation of the uncanny, in which movements and actions are produced to mimic that of humans, but miss the mark only somewhat noticeably. Is the adverse reaction (discomfort of the uncanniness) to this phenomenon in some way indicative of misfiring or miscued MNs? I understand this is quite a stretch, but the associational hypothesis entrenches the theory within a distinctly social realm, which inherently relies on external factors for such neurons to mirror and continue the associative process which produces “canonical neurons”.
In this paper, the author voices strong arguments against the genetic hypothesis. Their argument strongly roots for the sensorimotor association learning hypothesis. Through this paper, it argues that imitation, and observational mirroring is done through associations with motor neurons. This could even be extended to communication and thus would support the hypothesis that T3 is necessary in order to pass T2. Therefore maybe insight can come from understanding how mirror neurons truly work. However, with all that said, I am not sure what insight mirror neurons provide. At the end it's simply correlational and doesn't provide causal explanation. We don't know how associative learning generates the capacity for imitation.
ReplyDelete"The associative hypothesis says that genetic evolution has played a non-specific background role, and that the characteristic matching properties of MNs are forged by sensorimotor learning" (Cook et al, 2014).
ReplyDeleteCook et al argue for the Associative Hypothesis (AH) which is in opposition to the Genetic Hypothesis. The AH is characterized by MNs acquiring their matching properties through sensorimotor associate learning, i.e MNs develop through associative learning. Cook et al describe the AH as having 1) strong experience-dependence (correlated sensorimotor experience plays inductive or instructive role in MNs), 2) social construction (MNs are largely built through social interaction), and 3) contingency (event provoking firing of one predicts the event provoking of the other). These characteristics of the associative learning of MNs could potentially be helpful in understanding what 'parameters' are needed in order to come by understanding in a T3. In this paper, it is argued that observation, social interaction, sensorimotor experience is what gives MNs their mirroring ability. In this way, understanding MNs would be helpful because it could show that sensorimotor experience and social interaction is needed in order to produce the same mirroring that MNs do. Perhaps we need to allow our T3 robot (if it is ever created) the access to observation, social interaction, sensorimotor experiences if it is able to (has the capacity to) mirror and imitate in the way that MNs do for us. However, this does not really help in creating a robot (T3) that can do everything we can do. This is because even given the parameters of there needing to be social interaction, sensorimotor experience, etc., we are still unsure about how these parameters will cause the ability to mirror. And surely we don't need MNs to be able to recreate what they do.
Following the class, I find myself to be more of an agnostic regarding whether or not neuroscience and the knowledge of neural connectivity can aid in the reverse engineering of a robot as there have not been concrete examples of the how and why of neural connections directly impacting reverse engineering. Although as mentioned, using the brain and all of its connections would make a perfect T4 - which is what motivates me to believe that in theory neuroscience could be of use - it remains to be seen.
ReplyDelete“Only one study purports to offer direct evidence – from single cell recording – of MNs in the human brain (Mukamel et al. 2010). However, there is a considerable body of indirect evidence – from neuroimaging, transcranial magnetic stimulation (TMS), and behavioral studies –
ReplyDeletesuggesting that human brains contain MNs or comparable “mirror mechanisms”; circumscribed cortical areas involved in both action production and observation (Glenberg 2011).
It is interesting how the finding of mirror neurons and their application in understanding human cognition feel like a step in the right direction, but becomes limiting as it causes a slight tunnel vision. Localizing mirror function to singular cells provides the satisfaction of tangibility, but possibly oversimplifies a more complex system of contextual importance and visual kinematics. In terms of ascribing these cognitive abilities to artificial intelligence, it seems that understanding the smaller systems of predictive coding and contextual clues in action understanding seems like a more fruitful area of study rather than deeply understanding how and why motor neurons can do what they do.
I disagree somewhat. I think MN lays some hints at the idea that nature and nurture are more intertwined than we may have considered. Perhaps it's somewhat of a chicken vs. the egg scenario but it seems like nature would create MN the way they are such that they can be nurtured. On the other hand, perhaps MNs were nurtured from something more raw generations ago and inteligence has evolved to include them as a key component of cognition.
ReplyDeleteCook’s paper talk about the what mirror neurons are, how they are more likely to have been produced by associative learning rather than being produced by the genetic account. And in the class, we discussed this dichotomy (whether it was produced by associative learning or by genetic account) was not answering the main question: how these mirror neurons do what they do. This important question is important for t3, since the question that we are trying to answer is, does this type of research teach us how to make a robot at t3 level which can do what mirror neurons can do. We said this question mainly concerned t3, because t4 is about being able to produce the brain itself. Once you have t4, you already have t3. So to get to t4 you need to do t3 first.
ReplyDeleteAfter class I though that maybe we needed to change our perspective regarding the fact that we need to be able to produce t3 to get to t4. Maybe we don’t need these stages in thinking about how the mind works. I see that these stages try to draw lines between different levels of technicality in producing a brain, but this doesn’t really help us in how we should be solving the problems about understanding how brain works.
Maybe we need a different perspective regarding the way we think about the brain. Yes, Turing test will still be applicable even if we change our perspective and produce some other model of producing an understanding machinery. BUT, clearly our methods of reverse engineering regarding cognition/neuroscience are not sufficient in providing us with answers for that matter. It is not directly related but, maybe we need a perspective like the one that neural networks provided. In example for chess machines, we thought using brute force (with some heuristics) way of solving was the was the best way of solving the chess, until AlphaZero’s neural networks showed that it was not actually that good at all. As I said, this example is not directly related but, I think it describes the type of perspective change that I am trying to think about.
Some other ideas about another topic:
ReplyDeleteDuring the class Prof. Harnad stated that language might have originated from the gestures. I think the idea of this situation would be that people might have used gestures primarily for communication, then these gestures got translated into “symbols” overtime. And these translated gestures, ie. symbols, are a part of language.
I am not knowledgeable in history of languages, maybe the theory about gestures have a strong explanatory power regarding languages but, I thought after class that the claim about gestures does not necessarily provide a clear/easy reasoning about how languages started. One of the reason is, some animals seem to have gestures, or at least we infer that they have gestures by observing their behaviors. And as far as we know, we do not claim that any of the animal species has a language. So even though gestures might be necessary (we don’t have a way of disproving it since humans are the only species with language and they look like they had gestures for a long time), they are not sufficient in leading to the creation of a language.
Second, I think that the gestures alone would not able to solve the symbol-grounding problem by themselves. Sure, there are some levels of abstraction between the gestures and symbols, but I think that the levels of abstraction between drawing and language would be fewer since the drawings and the written parts of language shows higher resemblance than gestures (made by arms, face etc. within a visual field of another person) and any part (like written, spoken) part of language. It is not that simplest explanations are the true explanations, but I think language originating from drawings or drawings/ gestures provides a more parsimonious explanation that languages originating from gestures.
Still though, even if we push the idea that the languages are originated with gestures or/and drawing, we leave a big part of language out: the vocal/speaking part. Sure, we must speculate, the vocal part might just be the drawing-sound or gesture-sound associations that humans learned, but it would be assuming that vocal part came later than gestures/drawings. And it seems highly implausible.
In class and in other's skywritings this week it has been brought up that knowing about mirror neurons still doesn't tell us how we're able to imitate another person's movement, and that there isn't anything new being contributed towards our thinking about cognition/TT, etc/.
ReplyDeleteI think the point of bringing up mirror neurons in our discussion about cognition and everything else, is that mirror neurons should be thought of as a step towards the bigger picture. I think it's a very big deal that we have scientific evidence about something 'we already knew about'. Isn’t it important to be able to conclude something scientifically to allows you to know what is for certain and hypothesize further from that? If you don’t have scientific evidence, then your hypotheses are just hypotheses.
Mirror neurons give us the ability to locate, replicate, and add improvements towards replicating communication and understanding. They're included in our discussion about cognition/TT/understanding, etc. for a reason. I'm biased slightly in my opinion of mirror neurons because I'm learning about them in other psychology classes I'm taking, but regardless scientific evidence should be taken into consideration.
Another point I want to bring up is, in our discussion in class about Hubel and Wiesel's experiment with cats (which I am morally against), we came to the point that their discovery doesn't add anything to our discussion. But, I don't think enough credit was given to this study because they learned more than just a confirmation of our intuition. They learned “how visual cortical neurons encoded image features, the fundamental properties of objects that help us build our perception of the world around us”, and this is clear how it’s useful for our discussion. Understanding that perception of our world (understanding and grounding) and visual abilities are connected gives us a look into how understanding and grounding really occur, how our senses impact our understanding, a possible connection between the mind and the brain, etc.
This comment has been removed by the author.
ReplyDeleteWhile Cook et al (2014) defends the origination of MNs through sensorimotor associative learning, they are not necessarily answering the right questions. Rather, the focus of study on MN’s must shift to determine how they work instead of tackling nature vs. nurture.
ReplyDeleteI take a pessimistic approach on the matter of discovering the ‘how?’ The long list of MN functions noted in Cook et al paper makes the trouble of determining how they work much more challenging. This challenge then extends to the harder problem of how cognition works. Where do we go from here? What is the best approach to figure out ‘how’ MN’s work? The following article by Fodor makes me question even further the objectives of neuroscientists.
Re: Mirror neurons
ReplyDeleteGenetics are part of our biology, but I think that mirror neurons bring up a subtle point about how cognition arises. To put this into perspective, perhaps genetic makeup could be thought of as the original code where the “programmer” originally runs; mirror neurons could therefore demonstrate that recognition must occur beyond the programmed code. I can see why we need the T3 model in this case; in order to do what humans do, sensory-motor functionality must exist. However, the authors arguments support rather correlational aspects and does not provide causal explanation of whether neuroscience can aid in the reverse engineering of a robot.
The paper aims to settle a debate between two origins of MNs- genetic and associative. The genetic argument states that MNs are a by-product of evolution, with domain-specific genetic predisposition leading to MNs acquiring their "mirroriness". The associative account states that the mirroriness was acquired by domain-general associative learning.
ReplyDeleteThe basic confusion I had with these two domains was seeing why they are mutually exclusive, i.e., why can't associative learning be a product of evolution? The idea of domain-general vs domain-specific distinction still stands between both accounts.
As discussed in the previous comments on this article it seems that even mirror neurons are fascinating, they do not help explain the hard problem of cognition.
I find it rather difficult not to get lost in the theories about how mirror neurons could possibly contribute to empathy, learning, or predicting the intentions of others. Given that we are particularly social beings, it would be significant to know if mirror neurons are involved in our interpersonal relationships. That said, can acknowledge that even if we new for certain what it is that mirror neurons do, or if we located them in the brain, we would still not know the causal force that could explain how and why mirror neurons function the way that they do.
ReplyDeleteI am not sure how to break down in to more detail what is meant by "how and why" mirror neurons do what they do. I remember that in class I felt stumped as professor Harnad kept repeating and emphasizing that the properties and location were not equivalent to their function. But having reviewed the easy and hard problem of cognitive science and thought about how and why people behave the way that we do or feel the way that we do, I think the difference between properties and causal functions is becoming clearer to me. It might be that the jump from neuroscience, which focuses on structure, to cognitive science, which acknowledges the possibility of a hybrid mechanism that is both computational and observable as well as non-computational, is not simple. It is not obvious what is the right approach to studying these unobservable mental states, in which mirror neurons may or may not play a significant role.
Re: “I felt stumped as professor Harnad kept repeating and emphasizing that the properties and location were not equivalent to their function.”
ReplyDeleteI think one way to clarify that is to remember that the mind is to the brain what a programme is to a computer, as well as the fact that correlation is not causation. You could see specific areas of the brain get activated, but that does not necessarily mean that one causes the other. Also, there might be several ways to explain the same phenomenon. Hope this helps clarify it a little bit.
“The associative hypothesis of mirror neurons explains for example that “The kind of learning that produces MNs occurs when there is correlated (i.e., contiguous and contingent) excitation of sensory neurons and motor neurons that code similar actions. For example, when an adult imitates an infant’s facial movements, there might be correlated excitation of neurons that are responsive to the observation and execution of lip protrusion. Correlated excitation of the sensory and motor neurons increases the strength of the connection between them, so that subsequent excitation of the sensory neuron propagates to the motor neuron. Thereafter, the motor neuron fires, not only during execution of lip protrusion, but also, via its connection with the sensory neuron, during observation of lip protrusion; what was originally a motor neuron has become a lip protrusion MN. Correlated excitation of sensory and motor neurons encoding the same property of action occurs not only when humans are imitated, but also when we observe our own actions (directly or using an optical mirror)”
ReplyDeleteEven though I find this explanation interesting, I can’t see how having a lip protrusion MN that was activated during the observation of someone executing a lip protrusion would allow him to execute the same lip protrusion. For example, if I observe someone executing a back flip, would it mean that I would be able to execute the same back flip because I have a back-flip MN? Only practice of me executing the action and probably visualization of me executing the already practiced action would allow me to do so. If it is true that “when the system receives correlated experience of observing one action and executing a different action, the same associative process produces logically related MNs”, how come I am not able to reproduce any action that I observe? What do having MNs actually tell me about how I should produce this action? Correlating this observation with AI I believe that we are seeing what is actually useful for machine to learn, which is machine learning: the machine learns from past experiences, and continually practicing, making mistake and readjusting, not by observing something that is external to it.
On the example of mirror neurons, mirror neurons don’t tell us fully how the brain does what it does but it does give us a clue of how the system works and operates like knowing that circles are first encoded as points, then lines then curves, as a hierarchy with V1 cells. It is those processes used by the brain to make sense of the world that are interesting for cognitive scientist trying to reverse engineer the brain. I do think that knowing that a group of neurons perform a certain action is useful in designing similar neural sets performing similar actions.
ReplyDeleteThe MN seem to lack a solid ground (which is normal since it’s a fairly new discovery). There doesn’t seem to be any causal explanation and I guess it’s still a work in progress. Hypothetically speaking, when we will find out exactly where the MN come from and what exactly do they do I believe that it can be a good start to understand how the brain works since we will be able to see the connections the neurons makes with environment. However, we must know the exact formulas that are going on in our minds and neurons (T5) to get a better understanding of cognitive science. The discovery on MN cannot be ignored, but it doesn’t, yet, explain what needs to be explained.
ReplyDeleteMirror neurons fire when an animal performs an action or observes someone making a similar or equal action. Evolutionarily they are thought to help animals understand their world and learn from it. The paper presents the view that Cooks mirror neuron system might be incomplete for humans. A point is made questioning whether or not the research done in animals, especially in rats, is transferable to humans. An primary issue is not knowing how exactly mirror neurons work. Additionally, contagious behaviours from multimodal stimuli challenge the specific acts involved in mirror mechanisms.
ReplyDelete