In their target article, Cook et al. argue that mirror neurons (MNs) did not evolve to support action understanding. Instead, they argue that MNs emerge through a domain-general associative mechanism: Neurons that are contingently activated by observing and performing a particular action ultimately become selectively activated by either observing or performing that action. We agree with the authors that it is premature to conclude that MNs have the specific adaptive function of subserving action understanding (and that the term “action understanding” itself is typically ill-defined). Nevertheless, while we think Cook et al. are right that we know little about the function and origin of MNs, we think they may be right for the wrong reasons.
The first problem concerns the fact that Cook et al. use evidence that MN responses can change with experience to support their domain-general association account. Unfortunately, the finding that MNs are tuned by experience does not in principle mean that MNs are not innate and do not have an adaptive function. Instead, MNs may be responsive to goal- or object-directed actions more abstractly, perhaps starting out prepared to represent a fixed set of canonical actions that are phylogenetically ancient and then through experience becoming more finely tuned such that non-canonical actions can be represented as goal-directed (e.g., using a grasping tool to reach an object). For these reasons, we don't see the fact that MNs respond to a broader set of stimuli to which the “evolutionary ancestors of contemporary monkeys could not possibly have been exposed” (sect. 4.1, para. 3) to be difficult to reconcile with an innateness account. The idea that mirror neurons may start with a fixed set of prepared representations that gradually expand through experience seems to fit well with how many domain-specific cognitive mechanisms work.
Indeed, the science of the mind has several examples of abstract innate biases that are tuned by specific experiences. To take a classic example, rats have an innate bias to connect nausea with what they have eaten, but they must also use associative learning to identify which specific foods are to be avoided after different experiences (Garcia & Koelling Reference Garcia and Koelling1966). As this case illustrates, showing that MNs change with experience does not necessarily mean that they result in domain-general learning mechanisms that can take in any sort of inputs. Instead, MN representations could still emerge from domain-specific processes, ones that allow only certain kinds of experiences to act as inputs but become tuned to new inputs across experience.
A second set of reasons to doubt the target article's claim that MN learning is domain-general concern (1) the rarity of these neurons (according to one study, only about 17% of neurons sampled from a common monkey F5 region; Gallese et al. Reference Gallese, Fadiga, Fogassi and Rizzolatti1996), and (2) their seeming specificity to only particular types of visual-motor contingencies (e.g., grasping). If MNs develop through domain-general processes whenever there is contingent stimulation, Cook et al. should expect any and all neurons to develop “mirror” properties, and thus should expect to observe MNs for a variety of contingencies all over the brain. A domain-general account of MN learning should predict that species would develop mirror-like neurons that respond to Pavlovian contingencies, such as a neuron that fires both when a bell rings and when the subject salivates. If neurons are truly domain-general associative learners as the authors contend, then it is hard to explain why only some kinds of associations are represented by MNs to the exclusion of others and why only a small subset of neurons turn into MNs over the course of development. Even if MNs do acquire mirror properties through experience with contingent activation, this only pushes the question of specialization back to why these neurons change in response to contingencies while the majority of neurons in the brain do not.
Finally, the target article attempts to marshal evidence that human learners receive enough of the necessary types of inputs to create MNs through associative learning, stating that “much of the sensorimotor experience required for MN development comes from being imitated, synchronous action, and exposure to action words” (sect. 3.2., para.3). Unfortunately, this account of MN formation would likely not predict the existence of MNs in nonhuman primates, who rarely imitate (see Lyons et al. Reference Lyons, Santos and Keil2006) and do not have language at all (eliminating two of the proposed routes to MN formation). Thus, the experiential account proposed by Cook et al. to account for their proposed origin of MNs seems incomplete given the data.
For all these reasons, we believe that the target article's approach of searching for MNs' developmental origin before their function may not be the best way to examine or falsify an adaptive argument for these neurons. In order to falsify the hypothesis that MNs are involved in action understanding, we recommend that MN researchers instead directly test the role that MNs play in action understanding empirically – through experiments investigating the level at which actions are represented by these neurons. Since most of what we know about MNs comes from single-cell recording in monkeys, we propose to integrate neurophysiological recording techniques with behavioral methods that allow us better insight into the types of actions that monkeys understand (e.g., Umiltà et al. Reference Umiltà, Kohler, Gallese, Fogassi, Fadiga, Keysers and Rizzolatti2001). Primate cognition researchers have recently developed a number of new behavioral methods in which monkeys appear to represent others' object-directed actions (Rochat et al. Reference Rochat, Serra, Fadiga and Gallese2008), perceptions (Flombaum & Santos Reference Flombaum and Santos2005; Santos et al. Reference Santos, Nissen and Ferrugia2006), and knowledge (Marticorena et al. Reference Marticorena, Ruiz, Mukerji, Goddu and Santos2011). We argue that recording monkey MN activity while subjects perform these action understanding tasks could provide unique insight into the specific ways that action is represented in these neurons. Cook et al.'s article points us in the right direction by questioning the current accounts of the role MNs play in “action understanding,” but we believe that careful empirical work will be needed to test these accounts going forward. A domain-general association account is an inadequate replacement.
You have
Access
In their target article, Cook et al. argue that mirror neurons (MNs) did not evolve to support action understanding. Instead, they argue that MNs emerge through a domain-general associative mechanism: Neurons that are contingently activated by observing and performing a particular action ultimately become selectively activated by either observing or performing that action. We agree with the authors that it is premature to conclude that MNs have the specific adaptive function of subserving action understanding (and that the term “action understanding” itself is typically ill-defined). Nevertheless, while we think Cook et al. are right that we know little about the function and origin of MNs, we think they may be right for the wrong reasons.
The first problem concerns the fact that Cook et al. use evidence that MN responses can change with experience to support their domain-general association account. Unfortunately, the finding that MNs are tuned by experience does not in principle mean that MNs are not innate and do not have an adaptive function. Instead, MNs may be responsive to goal- or object-directed actions more abstractly, perhaps starting out prepared to represent a fixed set of canonical actions that are phylogenetically ancient and then through experience becoming more finely tuned such that non-canonical actions can be represented as goal-directed (e.g., using a grasping tool to reach an object). For these reasons, we don't see the fact that MNs respond to a broader set of stimuli to which the “evolutionary ancestors of contemporary monkeys could not possibly have been exposed” (sect. 4.1, para. 3) to be difficult to reconcile with an innateness account. The idea that mirror neurons may start with a fixed set of prepared representations that gradually expand through experience seems to fit well with how many domain-specific cognitive mechanisms work.
Indeed, the science of the mind has several examples of abstract innate biases that are tuned by specific experiences. To take a classic example, rats have an innate bias to connect nausea with what they have eaten, but they must also use associative learning to identify which specific foods are to be avoided after different experiences (Garcia & Koelling Reference Garcia and Koelling1966). As this case illustrates, showing that MNs change with experience does not necessarily mean that they result in domain-general learning mechanisms that can take in any sort of inputs. Instead, MN representations could still emerge from domain-specific processes, ones that allow only certain kinds of experiences to act as inputs but become tuned to new inputs across experience.
A second set of reasons to doubt the target article's claim that MN learning is domain-general concern (1) the rarity of these neurons (according to one study, only about 17% of neurons sampled from a common monkey F5 region; Gallese et al. Reference Gallese, Fadiga, Fogassi and Rizzolatti1996), and (2) their seeming specificity to only particular types of visual-motor contingencies (e.g., grasping). If MNs develop through domain-general processes whenever there is contingent stimulation, Cook et al. should expect any and all neurons to develop “mirror” properties, and thus should expect to observe MNs for a variety of contingencies all over the brain. A domain-general account of MN learning should predict that species would develop mirror-like neurons that respond to Pavlovian contingencies, such as a neuron that fires both when a bell rings and when the subject salivates. If neurons are truly domain-general associative learners as the authors contend, then it is hard to explain why only some kinds of associations are represented by MNs to the exclusion of others and why only a small subset of neurons turn into MNs over the course of development. Even if MNs do acquire mirror properties through experience with contingent activation, this only pushes the question of specialization back to why these neurons change in response to contingencies while the majority of neurons in the brain do not.
Finally, the target article attempts to marshal evidence that human learners receive enough of the necessary types of inputs to create MNs through associative learning, stating that “much of the sensorimotor experience required for MN development comes from being imitated, synchronous action, and exposure to action words” (sect. 3.2., para.3). Unfortunately, this account of MN formation would likely not predict the existence of MNs in nonhuman primates, who rarely imitate (see Lyons et al. Reference Lyons, Santos and Keil2006) and do not have language at all (eliminating two of the proposed routes to MN formation). Thus, the experiential account proposed by Cook et al. to account for their proposed origin of MNs seems incomplete given the data.
For all these reasons, we believe that the target article's approach of searching for MNs' developmental origin before their function may not be the best way to examine or falsify an adaptive argument for these neurons. In order to falsify the hypothesis that MNs are involved in action understanding, we recommend that MN researchers instead directly test the role that MNs play in action understanding empirically – through experiments investigating the level at which actions are represented by these neurons. Since most of what we know about MNs comes from single-cell recording in monkeys, we propose to integrate neurophysiological recording techniques with behavioral methods that allow us better insight into the types of actions that monkeys understand (e.g., Umiltà et al. Reference Umiltà, Kohler, Gallese, Fogassi, Fadiga, Keysers and Rizzolatti2001). Primate cognition researchers have recently developed a number of new behavioral methods in which monkeys appear to represent others' object-directed actions (Rochat et al. Reference Rochat, Serra, Fadiga and Gallese2008), perceptions (Flombaum & Santos Reference Flombaum and Santos2005; Santos et al. Reference Santos, Nissen and Ferrugia2006), and knowledge (Marticorena et al. Reference Marticorena, Ruiz, Mukerji, Goddu and Santos2011). We argue that recording monkey MN activity while subjects perform these action understanding tasks could provide unique insight into the specific ways that action is represented in these neurons. Cook et al.'s article points us in the right direction by questioning the current accounts of the role MNs play in “action understanding,” but we believe that careful empirical work will be needed to test these accounts going forward. A domain-general association account is an inadequate replacement.