The scientific merit and importance of mirror neurons and mirroring properties are well established.  What I wish to examine here are some recent statements by the Parma group about the significance of mirror neurons, statements more of a philosophical or programmatic nature than a narrowly scientific one.  In a recent paper, “A unifying view of the basis of social cognition” (2004), Gallese, Keysers and Rizzolatti make proposals about the nature of social understanding and about the prospects for unifying all of social cognition under the banner of mirroring phenomena.  These intriguing proposals are the ones to be considered here. 

Understanding Social Understanding

One aspect of social cognition is social understanding.  In their 2004 paper, Gallese, Keysers and Rizzolatti (GKR) focus heavily on this aspect. 

We will posit that, in our brain, there are neural mechanisms (mirror mechanisms) that allow us to directly understand the meaning of the actions and emotions of others by internally replicating (‘simulating’) them without any explicit reflective mediation.  Conceptual reasoning is not necessary for this understanding.  As human beings, of course, we are able to reason about others and to use this capacity to understand other people’s minds at the conceptual, declarative level.  Here we will argue, however, that the fundamental mechanism that allows us a direct experiential grasp of the mind of others is not conceptual reasoning but direct simulation of the observed events through the mirror mechanism.  The novelty of our approach consists in providing for the first time a neurophysiological account of the experiential dimension of both action and emotion understanding. (p. 396, emphases added)

Given the centrality of the understanding theme, one would like to be clear on the exact meaning of ‘understanding’.  What does it mean to understand an action or an emotion, or for that matter, to understand events in general?   In their 2004 paper and in other recent writings, the Parma group suggests several accounts of understanding actions and emotions.  These accounts are not equivalent to one another, however, and one would like to know which one they intend.  Moreover, several of these accounts seem incompatible with their above-quoted opposition to the “conceptual” approach to understanding.

In an earlier discussion, Rizzolatti, Fogassi and Gallese (2000) provide the context of social understanding and its application to mirror neurons. 

Primates are social animals living in continuous mutual relationship with conspecifics….  Therefore, it is crucial for each member of a given social group to be able to recognize the presence of another individual performing an action, to discriminate the observed action from others, and to “understand” the meaning of the observed action to react appropriately to it….  [W]e propose that the mirror system is a basic system for the recognition of action….  According to this view, there are “vocabularies” of motor actions at the core of the cortical motor system.  Neurons forming these vocabularies store both knowledge about an action and the description … of how this knowledge should be used.  The ensemble of neurons related to a given action forms the global motor schema of that action.  When an appropriate stimulus is presented, the relevant schema is activated.  (pp. 549-550)

Putting all this together, the Parma group implies that to understand an action is to recognize it, and to recognize it is to subsume it under a schema, which is a way of typing or grouping actions in terms of some important descriptions or categories.  This suggests to me a general approach to event understanding that fits the Parma group’s proposals.

(1)     To understand an event E is to subsume it under a familiar kind or type. 

This proposal bears some analogy to a widely-discussed model of explanation in philosophy of science, the so-called “subsumption” model of explanation (Hempel, 1965).[i] 

Account (1) leads one to expect an account of social understanding through concepts.  Identifying a kind or type presupposes a concept or category; that’s roughly what a schema is.  Moreover, a concept-friendly approach is endorsed in an earlier paper by Vittorio Gallese (2003), in which he links mirror neurons to the grasp of concepts.  Borrowing a theme from Dretske (1988), Gallese writes:  “it is the natural function of natural information to produce intentional representations, concepts included (2003, 1232; original emphasis deleted, new emphasis added).  Gallese defends the idea that monkeys have concepts, in part via their mirror-neuron systems.  “[T]he same conceptual content … results from a multiplicity of states subsuming it, namely, differently triggered patterns of activation within a population of ‘audio-visual mirror neurons’ (2003, 1238; emphasis added).  Dan Sperber (2004) also urges an interpretation of mirror-neuron activity in terms of conceptual content. 

However, GKR’s position in their 2004 paper is incompatible with these concept-friendly ideas.  The first passage quoted earlier says that mirror mechanisms allow us to “directly” understand other people, where direct understanding is said to occur without reflective or conceptual mediation.  This seems at odds with the account of understanding offered by (1).   I find the current approach to mirror systems that attempts to keep concepts at a distance less congenial than their older approach.  Has their view changed, and, if so, what are the reasons for the change?

In the 2004 paper, GKR offer another quite specific account of understanding, at least action understanding.  “The observer understands the action because he knows its outcomes when he does it” (p. 396).  A similar passage is found in Rizzolatti and Craighero (2004):  “Each time an individual sees an action done by another individual, neurons that represent that action are activated in the motor cortex.  This automatically induced motor representation of the observed action corresponds to what is spontaneously generated during active action and whose outcome is known to the acting individual” (p. 172).  These passages suggest the following account:

(2)     To understand an action or an event is to recognize its standard, customary outcome(s).

What is the relationship between (2) and (1)?  (2) might be viewed as just a special version of (1), in which the familiar type or kind is the usual outcome of the action or event.  Here we still have the problem of the incompatibility of this approach with GKR’s opposition to any role for concepts in “direct” understanding.

A different problem with (2) is excessive narrowness.  GKR propose a unifying basis for social understanding, intended at least for actions and emotions.  But how does (2) work in the case of emotions?  Is there a standard outcome associated with each emotion?  When one recognizes disgust or fear in another, what are the alleged outcomes that are recognized?  If GKR’s approach is to be generalized to all forms of social understanding, what is the guarantee that all events identified in others (especially mental events) are understood through their distinctive outcomes?

Still another conception of social understanding seems to be contained in the initially quoted passage plus ensuing material.  I already quoted this sentence:  “[T]he fundamental mechanism that allows us a direct experiential grasp of the mind of others is … direct simulation of the observed events through the mirror mechanism.”  The next paragraph continues: 

What makes social interactions so different from our perception of the inanimate world is that we … carry out similar actions and we experience similar emotions.  There is something shared between our first- and third-person experience of these phenomena: the observer and the observed are both individuals endowed with a similar brain-body system.  A crucial element of social cognition is the brain’s capacity to directly link the first- and third-person experiences of these phenomena (i.e. link ‘I do and I feel’ with ‘he does and he feels’).  We will define this mechanism ‘simulation’.  (Gallese et al. 2004, p. 397)

A preliminary reading of this passage suggests that it is experiential sharing that confers social understanding, or the experiential “grasp” of the minds of others.  This might inspire the following encapsulation:

(3)     Person A “directly” understands the mind of person B if A experiences a mental event that matches one experienced by B. 

If this account were correct, mirroring activity would obviously suffice for social understanding. 

Is (3) correct, however?  Two people each undergo a bout of nausea while sailing on an ocean vessel, one in the Atlantic and one in the Pacific.  The two are not in communication; they don’t even know of one another’s existence.  Does either person understand the other’s mind?  No.  So (3) doesn’t provide a sufficient condition for understanding other minds (perhaps not a necessary condition either).  If mirroring makes a contribution to social understanding, it cannot be in so simple a fashion as (3) suggests.

The trouble with the nausea case, arguably, is that there is no causal connection between the two passengers.  Their matching bouts of nausea are causally independent of one another.  So let us strengthen (3), yielding something perhaps closer to what GKR had in mind (but didn’t make explicit), namely (4).

(4)     Person A “directly” understands the mind of person B if A experiences a mental event that matches one experienced by B, and A’s mental event is caused by B’s mental event, via a brain mechanism that can reproduce such matches on similar occasions.

This clearly excludes the nausea case, as desired.  The reference to a brain mechanism capable of reproducing such matches is inserted to avoid cases of purely accidental causal relations.  For instance, one passenger’s bout of nausea might accidentally trigger his computer to send a nausea-inducing message to A.  That still wouldn’t qualify as direct social understanding.

Does (4) suffice for social understanding?  Not yet, I suspect.  What (4) arguably captures is a notion of mental contagion.  Is contagion sufficient for understanding, or mindreading?  I think not.  The receiver of a contagious mental state might not know the identity of the sender, and might not think of the sender as a subject of a matching state.  Crying among infants is contagious, and we may assume that the crying behavior transmits upsetness from one infant to another.  Do babies who experience upsetness as a consequence of hearing another’s cry represent the latter as upset?  As undergoing a matching state?  That is debatable.  If not, then I don’t think it’s a case of either mindreading or interpersonal understanding.  For these to occur, the receiver must represent the sender as a subject of a similar state, must impute such a state to the sender.  No such requirement is included in (4).

Perhaps GKR take note of this requirement in their discussion.  Recall this sentence: “A crucial element of social cognition is the brain’s capacity to directly link the first- and third-person experiences of these phenomena (i.e. link ‘I do and I feel’ with ‘he does and he feels’)” (2004, p. 396).  Perhaps their talk of “linkage” covers the element of imputation I am insisting upon.  If so, well and good.  But now the question arises of whether mirror-neuron activation constitutes, all by itself, such imputation.  Suppose monkey M₁ undergoes mirror-neuron activation in area F5 upon observing monkey M₂ engage in object grasping.  And suppose both monkeys’ patterns of activation constitute mental states of planning to perform object grasping.  Does the pattern of activation in M₁ constitute an act of representing M₂ as planning to perform object grasping?  Is it an imputation of such a plan to M₂?  At this juncture, such an interpretation would be very problematic.  It is certainly conceivable that monkeys make such imputations to other monkeys (or humans) when their mirroring systems are activated in observation mode.  But imputation may well be executed by a different part of the relevant circuit or network.  Is there any reason to think that mirroring per se constitutes such a “cognitive” element? 

Apply this to the case of human emotions, to a case the Parma people themselves have investigated.  A functional imaging study by Wicker et al. (2003) found that essentially the same brain regions were activated when a subject inhaled a foul odor and underwent an experience of disgust and also when the same subject observed somebody else inhale a foul odor and form a disgust-expressive face.  Here we have interpersonal mirroring activity.  Is it sufficient for social understanding to occur in the observation version of the test?  Does it entail that the observer mindreads the observed?  No.  Mindreading is consistent with this evidence but isn’t guaranteed by it.  Their study didn’t have subjects perform mindreading tasks, e.g., verbally classify the expressed emotion.  If spontaneous mindreading did occur, moreover, it might have been subserved by activity in other brain regions, not in the mirroring region(s). 

It is worth noting that the first paper to propose a connection between mirror neurons and the simulation theory of mindreading, viz., Gallese and Goldman (1998), did identify the elements of imputation and classification that I am now emphasizing.  Admittedly, these matters appeared in very condensed form and different terminology.  Our paper talked about an “attributor” who either predicts or retrodicts a target’s mental state.  The term “attribution” in effect introduces the ideas of representing the target, classifying the state, and imputing it to the target.  A specific mental state cannot be predicted or retrodicted for a target without a concept or classification of it.  The representation and imputation elements are also implied in this passage:  “[T]he subject of the MN activity knows (visually) that the observed target is concurrently performing this very action.  So we assume that he ‘tags’ the plan in question as belonging to that target” (1998, 497).  To say that the subject of MN activity “tags” the plan occurring in his own mind “as belonging” to the target implies that he imputes it to the target.  So Vittorio and I there presented a fairly thick account of social understanding, or mindreading.  Now Vittorio and colleagues seem to subscribe to a thinner account, which seems to exclude elements of conception or classification and ignore imputation.  This makes for a very “streamlined” account, which contrasts sharply with other well-known approaches in cognitive science.  By my lights, however, it goes too far in that direction.

Does Mirroring "Unify" Social Cognition?

Recall the title of the GKR paper: “A unifying view of the basis of social cognition”.  What is the scope and nature of their unity ambition?  Social cognition presumably includes interpersonal mindreading, all interpersonal mindreading.  Is that the intended scope of their view?  Their abstract reads:

In this article we provide a unifying neural hypothesis on how individuals understand the actions and emotions of others.  Our main claim is that the fundamental mechanism at the basis of the experiential understanding of others’ actions is the activation of the mirror neuron system.  A similar mechanism, but involving the activation of viscero-motor centers, underlies the experiential understanding of the emotions of others.  (2004, p. 396). 

Is the abstract meant to restrict the unity thesis to the understanding of actions and emotions (rather than the entire range of mental states)?  GKR explicitly set aside intention ascription in their paper.  Is this because of space limitations, or because they don’t believe that the action mirror-neuron system and its counterpart emotion-mirroring system are the basis of all interpersonal mindreading?  I am not sure, but will interpret their intended scope as more universal.

In addition to questions about the scope of their unity thesis, there is a question of its nature.  Do GKR claim that a single mechanism underlies all (“experiential”) understanding of action and emotions?  Or do they claim that a single kind of mechanism underlies such understanding?  The latter interpretation is compatible with many different mechanisms of the same kind.  This seems to be their intent, for the last sentence of the abstract speaks of  “a similar mechanism” underlying emotions.  Thus, GKR’s claim is that a certain kind of mechanism underlies all social understanding, and I interpret this to mean all actions and all mental states.  Which kind of mechanism do they have in mind?

At least three possibilities come to mind.  One possible kind, the narrowest one, is the class of mirror-neuron mechanisms first investigated, viz., motor mirroring mechanisms.  A second possible kind is mirroring mechanisms in general, with no restriction to motoricness.  This is perhaps suggested by GKR’s use of the phrase “a similar mechanism” when discussing emotion systems.  On the other hand, since they talk of emotions as involving viscero-motor centers, motoricness may be what qualifies these centers for inclusion.  A third possible kind is one of my invention.  It is simulational mechanisms, where ‘simulation’ is understood as a broader, more inclusive, process than mirroring.  GKR seem to restrict the term ‘simulation’ to mirroring.  At least this is suggested in the following passage:  “A crucial element of social cognition is the brain’s capacity to directly link the first- and third-person experiences of these phenomena …  We will define this mechanism ‘simulation’.” (p. 396)  This passage implies an equivalence between simulation and mirroring[ii]; but I prefer to think of mirroring as just one form of simulation.  (More on this shortly.)

If the theory being advocated were the theory that the unifying basis for social cognition concerns the activation of motor mirroring systems, then GKR’s unity thesis could aptly be labeled a “motor theory of social cognition”.  Precisely this label for the Parma group’s position is used by Jacob and Jeannerod (2004), so they presumably think that the relevant kind is motor mirroring mechanisms.  However, in his paper prepared for the present workshop, Gallese is very clear that this isn’t what he intends.  “According to the use I make of this notion, embodied simulation is not conceived of as being exclusively confined to the domain of motor control, but rather as a more general and basic endowment of our brain” (2004; emphasis in the original).  So, unless Gallese here differs from his colleagues in their joint 2004 article, the “motor theory of social cognition” isn’t an apt label for their approach. 

Let me now press for simulation mechanisms (broadly understood) as the best option in this territory as a potential unifying basis of social cognition.  Think of mental simulation mechanisms as systems prone to duplicate targeted mental events or processes, or as mechanisms prone to produce attempted duplications of such events.  Mirroring mechanisms fall into the first category of mental simulation mechanisms.  The second category of mental simulation mechanisms would be driven by systems for mental pretense or imagination.  These are the processes that traditional simulation theorists have typically invoked (see Gordon, 1986; Heal, 1986; Goldman, 1989; Harris, 1991).  When a mindreader tries to predict or retrodict someone else’s mental state by simulation, she uses pretense or imagination to put herself in the target’s “shoes” and generate the target state.  In other words, she tries to “take the perspective” of the target.  This kind of simulation mechanism may not produce matching states very reliably.  That’s one way they differ from mirroring systems.  Two other differences are, first, that these simulation mechanisms are partly voluntary rather than automatic, and they work partly at the conscious level rather than exclusively at the nonconscious level (as is true of mirroring systems).  Pretense- or imagination-driven simulation systems are an important subclass of simulation systems, and should not be confused with mirroring systems (see Goldman, in preparation).  A full theory of social cognition, therefore, cannot afford to neglect them. 

For example, mindreading other people’s epistemic states (knowledge and belief states) is an important type of social interaction.  Deliberate deception is based on this kind of mindreading and is unquestionably an important species of social interaction.  But it is unlikely that epistemic mindreading falls within the scope of mirroring systems.  On the other hand, epistemic mindreading probably does fall within the scope of pretense-driven or imagination-driven simulation systems.  Thus, an approach that appeals to simulation generally rather than mirroring is more promising as a unifying basis of social cognition.  If one attempts to create unification by exclusive attention to mirroring systems, the case for unification will be in jeopardy. 

Return now to Jacob and Jeannerod’s critique of the motor theory of social cognition.  They argue that the motor theory of social cognition cannot account for a mindreader’s understanding of what they call “social intentions”.  I suspect they are right; there isn’t any motor mirror system that can execute such a piece of mindreading.  That would be a problem for a theory of mindreading focused entirely on motor mirroring, but the broad form of simulation theory is not so restricted.  Anticipating this maneuver, Jacob and Jeannerod disparage non-motor simulation as undesirable “tinkering” with the concept of simulation.  “We disapprove this strategy because it relaxes the fundamental link between simulation and the requirements of the motor system, which we take very seriously”.  This step is inadequately defended.  The simulation theory of mindreading predates the discovery of mirror neurons,[iii]and there is no good reason to reconfigure simulation theory so as to confine it to motor mirroring processes.[iv]  To my knowledge no simulation theorist of mindreading has ever explicitly claimed that all mindreading is executed by motor mirror systems.  It is not “tinkering” with the simulation theory (of mindreading) to extend simulation beyond the motor realm.  Gallese, Keysers and Rizzolatti also seem to advocate too restrictive a position in claiming that all social cognition is underpinned by mirroring systems.  As discoverers of various forms of mirroring phenomena, both motor and non-motor, one can hardly begrudge them hopes for a very wide field of application.  But a slightly less ambitious program, I suspect, has a greater chance of being right.

References

Dretske, F. I. (1988).  Explaining Behavior, Reasons in a World of Causes.  Cambridge, MA: MIT Press. 

Gallese, V. (2003).  A neuroscientific grasp of concepts: from control to representation.  Philosophical Transactions of the Royal Society of London, B.  358: 1231-1240.

Gallese, V. (2004).  Intentional attunement.  The mirror neuron system and its role in interpersonal relations.  http://www.interdisciplines.org/mirror

Gallese, V. and Goldman, A. (1998).  Mirror neurons and the simulation theory of mindreading.  Trends in Cognitive Sciences 2: 493-501.

Gallese, V., Keysers, C. and Rizzolatti, G. (2004).  A unifying view of the basis of social cognition.  Trends in Cognitive Sciences 8, 9: 396-403.

Goldman, A. (1989).  Interpretation psychologized.  Mind and Language 4: 161-185.

Goldman, A. (in preparation).  The Simulating Mind.  New York: Oxford University Press.

Gordon, R. (1986).  Folk psychology as simulation.  Mind and Language 1: 158-170.

Harris, P. L. (1991).  The work of the imagination.  In A. Whiten, ed., Natural Theories of Mind.  Oxford: Blackwell.

Heal, J. (1986).  Replication and functionalism.  In J. Butterfield, ed., Language, Mind and Logic.  Cambridge: Cambridge University Press.

Hempel, C. G. (1965).  Aspects of Scientific Explanation and Other Essays in the Philosophy of Science.  New York: Free Press.

Jacob, P. and Jeannerod, M. (2004).  The motor theory of social cognition. A critique.  http://www.interdisciplines.org/mirror

Rizzolatti, G. and Craighero, L. (2004).  The mirror-neuron system.  Annual Review of Neuroscience 2: 661-192.

Rizzolatti, G., Fogassi, L. and Gallese, V. (2000).  Cortical mechanisms subserving object grasping and action recognition: a new view on the cortical motor functions.  In M. S. Gazzaniga, ed., The New Cognitive Neurosciences, 2^nd ed. (pp. 539-552).  Cambridge, MA: MIT Press.

Sperber, D. (2004).  “Mirror neurons” or “concept neurons”?  http:/www.interdisciplines.org/mirror

Wicker, B., Keysers, C., Plailly, J., Royet, J-P., Gallese, V., and Rizzolatti, G. (2003). Both of us disgusted in my insula: The common neural basis of seeing and feeling disgust.  Neuron 40: 655-664.

Acknowledgement.  Thanks to Holly Smith for valuable comments on an earlier draft of this paper.