Phenomenology and the Cognitive Sciences, DOI: 10.1007/s11097-011-9209-z This is a preprint version. Please cite the published version available at: http://www.springerlink.com/openurl.asp?genre=article&id=doi:10.1007/s11097-011-9209-z
On the Role of Social Interaction in Social Cognition: A Mechanistic Alternative to Enactivism Mitchell Herschbach Department of Philosophy University of California, San Diego 9500 Gilman Drive #0119 La Jolla, CA 92093-0119, USA Email:
[email protected] Abstract: Researchers in the enactivist tradition have recently argued that social interaction can constitute social cognition, rather than simply serve as the context for social cognition. They contend that a focus on social interaction corrects the overemphasis on mechanisms inside the individual in the explanation of social cognition. I critically assess enactivismʼs claims about the explanatory role of social interaction in social cognition. After sketching the enactivist approach to cognition in general and social cognition in particular, I identify problems with an enactivist taxonomy of roles for social interaction in the explanation of social cognition (contextual, enabling, and constitutive). In particular, I show that this enactivist taxonomy does not clearly distinguish between enabling conditions and constitutive elements, which would make them in danger of committing the coupling-constitution fallacy found in some attempts to extend cognition. I explore resources enactivism has to more clearly demarcate constitutive parts of a cognitive system, but identify problems in applying them to some of the main cases of social cognition enactivists characterize as being constituted by social interaction. I offer the mechanistic approach to explanation as an alternative that captures much of what enactivists want to say about the relations between social and individual levels, but views social interactions from the perspective of embedded cognition rather than as being constitutive of social cognition.
1.
Introduction
A common feature of embodied, embedded, enactive, extended approaches to cognition is that they challenge cognitivismʼs focus on the internal operations of the brain by proposing that features of the nonneural body and environment are essential to explanations of cognition, and that attending to these nonneural processes will lead us to revise our understanding of the brainʼs role in these brain-body-world interactions. The last 10-15 years have seen much discussion of the idea that parts of the physical environment can be essential to individual cognition, but only recently has an emphasis been put on the role of the social environment. Some have addressed whether social groups can count as cognitive systems in their own right (e.g., Huebner 2008; Hutchins 1995), but only recently has the focus been on the influence of social interactions on individual social cognition.
1
A rather radical voice in this discussion comes from enactivists who argue that social interaction can constitute social cognition (e.g., De Jaegher 2009a,b; De Jaegher & Di Paolo 2007; De Jaegher et al. 2010; De Jaegher & Froese 2009; Di Paolo 2005, 2009a,b; Di Paolo et al. 2011; Di Paolo et al. 2008; Froese & Di Paolo 2009, 2010; Fuchs & De Jaegher 2009; McGann & De Jaegher 2009; Torrance & Froese 2011). This position is characterized as opposed to attempts to explain social cognition solely in terms of internal neural mechanisms, for example, a “mindreading” or “theory of mind” mechanism that enables the understanding of othersʼ mental states. Cognitivism is one target of this enactivist criticism, but it is not the only one. Even adherents of embodied cognition are targeted for “rely[ing] on neurological mechanisms such as so-called mirror neurons when explaining social understanding” and failing to recognize social interaction processes themselves as playing a constitutive role in social cognition (De Jaegher & Di Paolo 2007, p. 495). Against these forms of internalism, enactivists protest that “social cognition is not reducible to the workings of individual cognitive mechanisms” and that “interactive processes are more than a context for social cognition: they can complement and even replace individual mechanisms” (De Jeagher et al. 2010, p. 441). Enactivists appear to be making more than the methodological claim that the unit of analysis for cognitive science should include the social context for individual behavior and cognition. Most everyone today would agree with this claim. For example, cognitivists treat the environment as being the site of the inputs to and outputs from cognitive processes, and thus worthy of study—even if it has not been given adequate attention in most research in this tradition, and even if cognitivists typically think no cognition proper happens outside the head. Enactivists are in addition making a claim about the explanation of social cognition and seemingly a metaphysical claim about what constitutes social cognition: that social interaction processes, rather than (or in addition to) internal neural mechanisms, can explain some social behavior of individuals because interaction processes can constitute social cognition. For example, Froese and Di Paolo (2010) clearly express the methodological claim when they write that most cognitive science suffers from “methodological individualism,” the “fundamental assumption…that the individual agent is the correct unit of analysis for understanding mind, cognition and behaviour” (p. 43). They move into explanatory and/or metaphysical claims, however, when they explicitly characterize this methodological
2
individualism as “an exclusive focus on individual agency whereby all social phenomena including social cognition is reducible without remainder to individual mechanisms” (p. 43). The explanatory and metaphysical claims are clear when enactivists talk about the constitution of social cognition including social interactions, in the way extended mind theorists say the physical environment can partly constitute cognitive processes. De Jaegher and Frose (2009), for example, write that “social agents, as well as being constitutive of the social interactions they engage in, are also constituted by them” (p. 446). Even clearer, Froese and Di Paolo (2009) explicitly use the language of “cognitive extension” from extended mind approaches when characterizing the constitutive role of social interaction in social cognition (p. 442). Enactivists thus reject forms of reductionism that metaphysically characterize and therefore explain social cognition only in terms of the brain, and instead propose that social interactions can constitute social cognition. In this paper I offer a critical analysis of the enactivist position that social interaction can explain social cognition because it is a constitutive element of it. I argue that De Jaegher et al.ʼs (2010) attempt to distinguish contextual, enabling, and constitutive explanatory roles does not adequately characterize what is required to be a constitutive element of social cognition. Further, I argue that even other enactivist proposals about the constitutive elements of cognitive systems do not clearly cover some paradigm cases of social-interaction-based explanations of social cognition appealed to by enactivists—namely, human and computational modeling studies of perceptual crossing. I offer the mechanistic approach as a better way to formulate much of what the enactivists want to say about the relation between multiple levels of organization: the collective level of social interactions, the personal level of individual agents, and subpersonal levels iteratively decomposing agents into their parts. The mechanistic perspective I endorse does not, however, support the enactivist idea that social interactions constitute social cognition. The social interactions with other agents rather serve as part of the environment in which agents are situated, and to which the behavioral repertoire made possible by their internal mechanisms can be better or worse adapted. I begin in section 2 by outlining the enactivist approach. Then in section 3 I describe enactivist accounts of social interaction and its role in social cognition. I then describe and identify problems with an
3
enactivist taxonomy of possible explanatory roles for social interaction. In so doing I describe empirical evidence often appealed to by enactivists in support of the idea that social interaction can constitute social cognition: a human experiment (Auvray et al. 2009) and computational models of perceptual crossing (Di Paolo et al. 2008; Froese & Di Paolo, 2010). In section 4 I attempt to interpret in more detail how enactivists demarcate constitutive parts of cognitive systems. Having found problems with the enactivist justification for treating social interactions as constituents of social cognition, in section 5 I describe the similarities and differences between enactivism and the mechanistic approach, and show how to explain perceptual crossing within the mechanistic framework.
2.
The Enactivist Approach to Cognition
”Enactivism” is used to describe several related approaches emphasizing the embodied, dynamic, environmentally situated nature of cognition. Here I focus on the enactivist tradition exemplified by Varela et al. (1991), Di Paolo (2005) and Thompson (2007). According to this enactivism, mind and life are intimately connected. Enactivists believe that living organisms are “the paradigm cognitive beings,” and that “what makes living organisms cognitive beings is that they embody or realize a certain kind of autonomy—they are internally self-constructive in such a way as to regulate actively their interactions with their environments” (Thompson & Stapleton 2008, p. 24). Autonomous systems are “internally self-constructive” in that their constitutive parts dynamically interact and self-organize so as to form a unitary, bounded system that has an identity distinct from its environment. Enactivists use the phrase operational closure to characterize the way an autonomous systemʼs constituent parts recursively depend on one another while generating and sustaining the systemʼs identity. Relatedly, an autonomous system is far from thermodynamic equilibrium, and accordingly must continuously take in matter and energy from its environment in order to continuously generate and sustain its identity. As Thompson and Stapleton (2008) summarize, “an autonomous system is a thermodynamically open system with operational closure that actively generates and sustains its identity under precarious conditions,” where “precarious conditions” refers to how the parts of the system would “tend to run down or extinguish” outside the context of the system (p. 24). The living cell serves as a paradigm of an autonomous system: its chemical constituents
4
self-organize into a metabolic network that produces its own semipermeable membrane through which matter and energy can be exchanged in order to repair and rebuild the component parts of the cell. This form of autonomy is known as autopoiesis (Thompson 2007, ch. 5). But other kinds of systems can exhibit non-autopoietic autonomy, including immune systems, nervous systems, and social groups. Enactivists often use the language of dynamical systems to describe the internal activity of autonomous systems and their environmental interactions. A system is said to be coupled to another “when the conduct of each is a function of the conduct of the other,” or in more technical terms, when “the state variables of one system are parameters of the other system, and vise versa” (Thompson 2007, p. 45). An autonomous system regulates its coupling with its environment when it influences the coupling so as to maintain its self-generated identity (De Jaegher et al. 2010, p. 441). An autonomous system is called adaptive when capable of this regulated environmental coupling (Di Paolo 2005, 2009; Thompson 2007, p. 148). Adaptive autonomous systems are thus inherently active rather than static, in that their inner workings have their own endogenous dynamics and these systems actively regulate their interactions with the environment, rather than passively reacting to stimuli. It is the adaptive response to the environment that points to another core concept of the enactive approach: sense-making. That an adaptive autonomous systemʼs activity and interaction with its environment has an aim or goal—the self-production and self-maintenance of the autonomous system— means that the system has a perspective on the world, that worldly entities take on a significance or meaning for the system (e.g., as food, as dangerous). Sense-making is this interaction with the environment by which it takes on meaning and value for the adaptive autonomous system. Enactivists often talk in terms of sense-making rather than cognition, or define cognition in terms of sense-making (Thompson 2007, p. 159), partly to emphasize the connection between autonomy and cognition but also to emphasize that sense-making is value-laden and that the distinction between cognition and affect is artificial. While it will not be a focus here, enactivism also treats experience as intimately connected to life and sense-making, and emphasizes that the scientific study of cognition should take experience seriously in its descriptions and explanations.
5
In sum, enactivists see adaptive autonomy as necessary for cognition/sense-making. Cognition is understood as not merely a passive reaction to sensory input in order to mediate action, but rather as the generation of meaning for an endogenously active, autonomous system as it regulates its environmental interactions.
3.
Enactivism on Social Cognition
3.1.
Enactivist Accounts of Social Interaction and Social Cognition
How does the enactivist picture apply to social cognition? Enactivists usually begin by characterizing what they mean by “social interaction,” and then how social interaction can influence the cognition/sense-making of individuals. As mentioned above, enactivism uses dynamical systems theory to characterize the interactions between an autonomous system (e.g., a human) and its environment. The same language is used to characterize the social interactions between agents (De Jaegher & Di Paolo 2007; De Jaegher et al. 2010). Two systems are coupled when the behavior of one depends on the behavior of the other. Coordination is defined as “the non-accidental correlation between the behaviours of two or more systems that are in sustained coupling, or have been coupled in the past, or have been coupled to another, common, system” (De Jaegher & Di Paolo 2007 p. 490). For example, during a conversation two peopleʼs speech and bodily movements may synchronize. Synchronization is not, however, the only way to be coordinated. The coordination patterns of a social interaction may involve shifts between synchronized, desynchronized, and in-between states. Further, coupled systems can exhibit varying degrees of coordination (De Jaegher & Di Paolo 2007, p. 491). Finally, note that coordination is something merely physical, non-living systems are capable of achieving, and thus need not involve cognition. As explained above, an agent (an adaptive autonomous system) acts to influence its coupling with the environment, so as to ensure the agentʼs continued existence. An agent is said to regulate its coupling by engaging in motivated changes to the constraints or parameters that influence the coupling. Coregulated coupling occurs when the two coupled systems are both agents and they each regulate their coupling (De Jaegher et al. 2010, p. 441).
6
These dynamical concepts of coupling, coordination, and regulation are used to offer the following definition of social interaction: Social interaction is the regulated coupling between at least two autonomous agents, where the regulation is aimed at aspects of the coupling itself so that it constitutes an emergent autonomous organization in the domain of relational dynamics, without destroying in the process the autonomy of the agents involved (though the latterʼs scope can be augmented or reduced). (De Jaegher & Di Paolo 2007, p. 493) A social interaction is thus co-regulated coupling between two autonomous agents, such that the social interaction itself becomes autonomous—the interacting partners act such that the interaction takes on a “life of its own,” however fleeting it may be, that can influence the behavior and cognition of the interactors without destroying their individual autonomy. If one agent comes to dominate the regulation—for example, through coercion—the interaction would no longer count as social interaction, as it would be analogous to one agent interacting with a non-social object (De Jaegher et al. 2010, p. 443). The enactivist definition of social interaction thus applies to only a subset of the causal interactions between agents. It seems a motivation for this particular definition of social interaction is that the autonomy of the social interaction is important to establishing the coordination between social agents as “a proper level of analysis in itself,” such that the autonomous social interaction is “an identifiable pattern with its own internal structure” and can influence the cognition and behavior of the individual participants (De Jaegher & Di Paolo 2007, pp. 491-492). Enactivists often use as an example of an autonomous social interaction the way two people might interact as they approach from opposite directions down a narrow hallway: In attempting to walk past each other, you both step towards the same side. This may happen a few times before you are finally able to bypass each other. Here, the coordination of movements (a temporally synchronised mirroring of sideways steps) ensures (for a brief while) that the interaction process is sustained despite the fact that you both want to stop interacting in this way. This example illustrates how the interpersonal coordination of movements endows the interaction process with a form of autonomy. We can even go so far as to say that the interaction process itself influences the interactors. (De Jaegher 2009a, p. 539; cf. De Jaegher & Di Paolo 2007, p. 493) The social interaction is autonomous in the sense that its constituents (the two agents) self-organize to maintain the interaction process over time, even in cases where it conflicts with the intentions of the individual agents.
7
Enactivists use the phrase participatory sense-making to characterize the way individual cognition/sense-making can be influenced by social interactions. De Jaegher and Di Paolo (2007) define it as “the coordination of intentional activity in interaction, whereby individual sense-making processes are affected and new domains of social sense-making can be generated that were not available to each individual on her own” (p. 497). This latter point—that the socio-cognitive abilities of individuals can be expanded when in the right context of interaction—is essential to the enactivist claim about the importance of social interaction to social cognition, as I will discuss below.
3.2.
Taxonomy of Explanatory Roles for Social Interaction in Social Cognition
Although it is not explicitly described as an enactivist position, De Jaegher et al. (2010) offer the following taxonomy of possible roles for social interaction in the explanation of social cognition. For a given phenomenon X, the situation in which X occurs is “the collection of past and present events, processes and relations that are observed with a phenomenon X” (p. 443). For the explanatorily relevant elements in this collection, the authors identify three possible explanatory roles they could play, presented in order of increasing specificity: • • •
F is a contextual factor if variations in F produce variations in X, C is an enabling condition if the absence of C prevents X from occurring and P is a constitutive element if P is part of the processes that produce X. (p. 443)
De Jaegher et al. (2010) illustrate these different explanatory roles using the phenomenon of water boiling in an electric kettle. A change in air pressure would affect this phenomenon, but is not necessary for its existence, so is a contextual factor but not an enabling condition. A supply of electricity is necessary for but not a part of the phenomenon, so is an enabling condition (and contextual factor) but not a constitutive element. The phenomenonʼs constitutive elements (which also count as contextual and enabling conditions) are “an appropriate heat exchange between a metal plate and the water provoking a phase transition from liquid to steam” (p. 443). Before addressing its application to social cognition, there is an important ambiguity about the nature of constitution in this example and the general account. In various passages De Jaegher et al. (2010) treat a constitutive element as a part of the whole phenomenon, such that “The set of all the
8
constitutive elements is the phenomenon itself” (p. 443). But in their explicit definition quoted above, a constitutive element is described as “a part of the processes that produce the phenomenon” (p. 443, italics added). Does this mean that their definition of “constitutive elements” covers not just not those internal parts that compose or constitute the whole phenomenon, but also elements spatially external to the phenomenon which help produce it? How to interpret the authors on this point seems to depend on what is identified as the explanandum. In their illustration, is the phenomenon to be explained just the boiling of the water, or this plus the kettleʼs metal plate that heats the water? If the former, then the authors are saying that the metal plate is a constitutive element of the waterʼs boiling even though it is not a material component of the water. If the phenomenon is the water plus metal plate, then the “constitutive elements” identified by the authors are all parts which compose it. These two interpretations offer very different pictures of what is meant by “constitutive element.” Since the authors more often say that constitutive elements are parts of the phenomenon, rather than elements that produce the phenomenon, it suggests they intend the second reading of the boiling water example. But as we will see below, this sort of ambiguity about the phenomenon of interest and its constitutive elements pervades enactivist writings. De Jaegher et al. (2010) first apply their explanatory framework to social interaction using Murray and Trevarthenʼs (1985) experiments on “protoconversations”: face-to-face interactions between parents and infants where they contingently respond to each otherʼs facial expressions, vocalizations, and gestures expressive of emotion. Their double TV experiment had mothers interact with their 2-month-old infants via a television link. In the live interaction condition, where the mothers and infants received live video feeds of each other, infants behaved quite normally, actively interacting with their mothers. But in the replay condition, when infants were shown a recording of their motherʼs previous actions, they disengaged from the interaction and acted distracted and distressed. One possible explanation of infant behavior in the double TV experiment is that infants possess a contingency detection module, a neural mechanism that detects contingencies between their behavior and environmental events (e.g., Gergeley & Watson 1999). De Jaegher et al. (2010) identify this as treating social interaction as a mere contextual factor: “As input to this module, interaction is only a contextual factor: variations in the interaction pattern change the output of the module” (p. 443).
9
When illustrating enabling and constitutive roles for social interaction, De Jaegher et al. (2010) appeal to related research on the recognition of perceptual crossing: the ability of an agent to detect the presence of another agent when the two agents are in perceptual interaction “(as in the case of mutual touch or catching one anotherʼs eye)” (Auvray et al. 2009, p. 33). Auvray et al. (2009) studied human recognition of perceptual crossing in a very minimal virtual environment. Computational modeling work based on Auvray et al.ʼs setup was conducted by Di Paolo et al. (2008) and Froese and Di Paolo (2010). This research is repeatedly cited by enactivists as providing empirical evidence of social interactionʼs playing a constitutive role in social cognition, so I will describe it in some detail. 3.2.1.
Behavioral and Computational Modeling Studies of Perceptual Crossing
These perceptual crossing studies use the following setup (Fig 1). Two agents face each other in a 1dimensional environment that wraps around itself after 600 units of space. Each agent controls the position of a sensor (their “avatar”) in this 1-D space, which is activated when it overlaps with another environmental object. There are three objects in this environment: a static, fixed object; the other agentʼs sensor/avatar; and a “shadow” object that copies the movements of the other agentʼs sensor at a fixed distance from it. All three objects are the same size, thus producing the same sensory stimulation. The agentʼs task is to detect when it is in perceptual contact with the other agent, rather than the fixed object or shadow.
Fig 1 Setup for the perceptual crossing experiments of Auvray et al. (2009) and computational models of Di Paolo et al. (2008) and Froese and Di Paolo (2010). In Auvray et al.ʼs (2009) experiment, the adult participants were paired up, seated in different rooms, then blindfolded. Each controlled their sensor with a computer mouse and received tactile feedback
10
on an index pad on their other hand. Participants were instructed that the virtual environment contained three types of tactile objects—the partnerʼs avatar, fixed objects and mobile objects—but not informed of the connection between the shadow and the other agentʼs avatar. Their task was to click the mouse whenever they judged their sensor activation to be due to encountering the other agentʼs avatar, and to click as many times as they judged perceptual crossing to occur. The experimenters found that participants successfully detected perceptual crossing even in this minimal sensory situation. 65.9% of participantsʼ clicks were in response to another agent, compared to 23% of clicks due to the shadow and 11% of clicks due to the fixed object. Whatʼs notable is that while participants could perceptually discriminate between moving and fixed objects, they did not seem to be able to discriminate between the other agentʼs avatar and the moving shadow. The probability of clicking given stimulation by a moving object significantly differed from the probability of clicking given an encounter with a fixed object; but there was no significant difference between the probability of clicking given an encounter with the shadow and the other agent. The authors explained the ability to discriminate between fixed and moving objects in terms of “sensorimotor contingencies,” i.e., the relations between a participantʼs actions and sensory stimulation. Participants were observed to engage in a general strategy of changing their direction of movement after a sensory encounter. In the case of fixed objects, this strategy led to a regular symmetrical oscillation around the fixed source of stimulation. But in the case of moving objects, this strategy led to an asymmetrical oscillation around a source of stimulation that was constantly displaced. Since the perceptual information was so impoverished, it is not surprising that relations between the agentʼs behavior and sensory stimulation, rather than simply features of the sensory input, were needed to perceptually discriminate between even moving and fixed objects. What enactivists emphasize most is the explanation of why participants were able to succeed on the task even though their clicking behavior suggests they could not perceptually discriminate between the other agent and their shadow. Their success is explained in terms of the collective dynamics of the social interaction between the two participants, which led to a higher frequency of agent-agent interaction than agent-shadow interaction—52.2% of sensory encounters were caused by the other agent versus 15.2% for
11
shadows—and thus more clicks for agents than shadows. How did the dynamics of the social interaction produce this result? Remember that participants used a strategy of reversing the direction of their movement when encountering a source of stimulation, such that they tend to oscillate around an object. When another agent is encountered, that agent will similarly receive sensory stimulation, and engage in the same oscillatory behavior. This leads to a stable state of two-way mutual scanning—i.e., agent-agent interaction is an “attractor” in the landscape of possible behaviors by the two agents. When a participant encounters a shadow, however, the other participant will not receive any sensory stimulation and thus will continue moving in the same direction, dragging their shadow along with them, away from the first participantʼs sensor. This means agent-shadow interaction is inherently unstable and will terminate much sooner than agent-agent interaction. As Froese and Di Paolo (2010) summarize, “in contrast to the expectations of a methodological individualist framework, the solution to the task does not rely on an individual ability to detect social contingency, but emerges from the mutual perceptual activity of the subjects that are oriented towards each other” (p. 47). Given the dynamics of the social interaction, agentagent interactions are more likely to occur than agent-shadow interactions; so even though an individual canʼt perceptually discriminate between agents and shadows, they will still click on agents much more often than shadows. Di Paolo et al. (2008) created computational models of artificial agents engaging in the same task. The artificial agents were controlled by continuous-time recurrent neural networks consisting of several input nodes activated by the sensor, output nodes for left and right movement, respectively, and several hidden nodes. A genetic algorithm was used to evolve agents well adapted to the task of detecting the other agent (operationalized as maximizing time close to the other agent). In addition to replicating the search strategy, successful performance, and collective dynamics of Auvray et al.ʼs (2009) participants, Di Paolo et al. (2008) also modeled a closer approximation of Murray and Trevarthenʼs (1985) double TV experiment by evolving agents that would interact with a live interactive partner but disengage from a noncontingent partner that is replaying motions from a previous live interaction. Just like Murray and Trevarthenʼs mothers and infants, the artificial agents created stable interactions in the live condition, but the interactions broke down in the replay condition. Like with Auvray et al.ʼs (2009) participants, the stable
12
interaction in the live condition is explained by the collective dynamics generated by their oscillatory search strategy. In contrast, the interaction broke down in the replay condition, since the individual agent cannot alone sustain the oscillatory behavior needed for stable coordination between the two agentsʼ movements. The input to their neural network controllers are clearly different in the two conditions. But like with the human subjects, this internal mechanism is not capable of discriminating between a contingently responding live agent and a non-contingent replay agent. So, according to enactivists, the internal mechanism is not responsible for the difference in behavior across the two conditions. Rather, it is a difference in the collective dynamics of the social interaction across the two conditions that explains the difference in the individualʼs behavior.
3.2.2.
Evaluating the Enactivist Explanatory Framework as Applied to Social Interaction
How do enactivists characterize these explanations of perceptual crossing in terms of their taxonomy of explanatory roles? Somewhat strangely, De Jaegher et al. (2010) identify Di Paolo et al.ʼs (2008) experiment involving live vs. recorded conditions as a case where social interaction is only an enabling condition for social cognition, but describe Auvray et al.ʼs (2009) experiment as a case where social cognition is constituted by social interaction. With regard to the former case, where the phenomenon is whether an individual sustains or disengages from interaction, De Jaegher et al. (2010) write that “Without interaction the necessary processes would not function and the phenomenon would not occur, making interaction an enabling condition” (p. 444). This interpretation is consistent with requiring constitutive elements to be internal parts which compose the phenomenon. But with regard to Auvray et al.ʼs experiment, where the phenomenon is the participantʼs clicking more often in cases of perceptual crossing than agent-shadow or agent-object interaction, they write: The variation in the number of clicks is attributable only to the differences in the stability of the coupling and not to individual strategies. This experiment shows that the interaction process is not only enabling but plays a constitutive role. The phenomenon is a manifestation of the properties of the interaction pattern. (De Jaegher et al. 2010, p. 445) This interpretation is inconsistent with requiring constitutive elements to be internal parts. The authors clearly indicate that the social interaction is necessary for both cases of social cognition. But since the same kind of explanation in terms of the collective dynamics of the social
13
interaction is given in both cases, it is unclear why the language of constitution is only applied to one but not the other. They do not explicitly discuss how social interaction could enable social cognition without also constituting it, or more generally whether being an internal part is relevant to whether something is a constitutive element. This invites the question of what enactivists really mean when they say social interaction constitutes social cognition. There are also problems with De Jaegher et al.ʼs (2010) illustration of social interactionʼs playing a contextual factor: social interactionʼs serving as input to a contingency detection mechanism that is to explain infantsʼ behavior in the double TV experiment. It is surely right that variation in the social interaction produces differences in the activity of such a mechanism. But it seems natural to say social interaction would also count as enabling the infantʼs behavior. If the contingency detection module is a passively reactive mechanism, as traditionally assumed by cognitivists, it will fail to respond without some input. The appropriate input (contingent social interaction from the mother) is thus necessary for the mechanism to indicate the presence of social contingency and produce the infantʼs interactive behavior. Whether social interaction (as defined by enactivism) is necessary to generate the appropriate input for such an internal mechanism is an open question. But if this were true, it would mean that De Jaegher et al. (2010) are wrong to characterize internalist, cognitivist explanations of social cognition as treating social interactions as only contextual factors. Here the distinction between contextual factors and enabling conditions seems fairly clear; the problem is with the authorsʼ application of these concepts to particular explanations of social cognition. A further problem comes when De Jaegher et al. (2010) add that, for this contingency-detectionmechanism hypothesis, “the cognitive mechanism is not constituted by social interaction” (p. 443). In what sense could a social interaction be a constitutive element of a neural mechanism (p. 443)? If a constitutive element is understood as a “part of the phenomenon” itself (p. 443), this statement would involve a substantial confusion between levels of organization, between the social level constituted by interacting agents and the level of the neural mechanisms inside an individual agent. Perhaps the authors meant to say that what is not constituted by social interaction is the socio-cognitive phenomenon of contingency detection, rather than the internal cognitive mechanism, the contingency-detection module. But if this
14
passage is as the authorsʼ intended, then it confuses between levels of organization or uses a noncompositional sense of constitution inconsistent with what they say elsewhere about constitution. In sum, De Jaegher et al.ʼs (2010) taxonomy of explanatory roles is not as clear as it first appears. If these explanatory roles are not well delineated and how they apply to social interaction is unclear, it suggests there might be ambiguity in what enactivists mean when they claim social interaction can constitute social cognition. I will next analyze in more detail what enactivists have to say about social interactionʼs constituting social cognition, and the constitution of cognitive systems in general.
4.
Enactivism and the Constitution of Cognitive Systems
I content there is a significant ambiguity in what enactivists mean by “constitution.” In some places enactivists seem to use “constitutive element” in a fairly standard sense, as referring to the parts that compose a whole phenomenon. Their claim that social interaction can constitute social cognition would thus mean that processes of social cognition extend out beyond the boundaries of the individual agentʼs brain and non-neural body. This reading is supported by some enactivistsʼ descriptions of social interactions in terms of “cognitive extension” in the sense intended by extended mind/cognition approaches (Froese & Di Paolo 2009, p. 442). But other times enactivists do not seem to require constitutive elements to be literal parts of the cognitive system. Some of enactivistsʼ talk about the constitutive explanatory role of social interaction in social cognition seem to collapse the distinction between the thesis of extended cognition, according to which cognitive processes are partly constituted by parts of the extra-organismal environment, and the thesis of embedded cognition, according to which “cognitive processes depend very heavily, in hitherto unexpected ways, on organismically external props and devices and on the structure of the external environment in which cognition takes place” but do not include these environmental resources as constituent parts (Rupert 2004, p. 393). For example, De Jaegher and Froese (2009) describe the claim that social interaction can constitute cognition as saying that “the constitution of individual agency can be influenced by the interactional contexts in which social agents routinely find themselves” (p. 449, italics added), and that “individual agency, at least individual human agency, is to a large extent determined by
15
social factors” (p. 456, italics added). These passages seem to treat social interaction as an external factor influencing an individual agentʼs social cognition rather than being a constituent part of these cognitive processes. In so doing, enactivists appear to treat any factor that is necessary for a cognitive process to be a constituent of that cognitive process, thus collapsing the distinction between enabling conditions and constitutive elements in the explanation of cognition. Accordingly, enactivism would be susceptible to what Adams and Aizawa (2001, 2008) call the coupling-constitution fallacy: the mistaken belief that anything that causally interacts or is coupled with a cognitive system/process, and thus anything relevant to the explanation of cognitive phenomena, counts as a constituent of the cognitive system/process. With Adams and Aizawa, I believe it would be a mistake to call every explanatorily relevant factor a constituent of a cognitive system/process. Enactivists do not go quite this far, since they do distinguish contextual and enabling conditions. But their apparent treatment of spatially external enabling factors as constituents of a cognitive system still falls afoul of the coupling-constitution fallacy. And while there is admittedly no consensus in the literature about the proposed criteria for the constitutive parts of a cognitive system, the question is whether enactivism can itself avoid the coupling-constitution fallacy. Enactivists do seem to have the resources to make a distinction between a cognitive system and its environment using their notion of autonomy, and thus to avoid the coupling-constitution fallacy. Those elements that are part of a self-organizing, operationally closed autonomous system which self-produces and self-maintains its identity over time, are constitutive parts of this system and distinct from environmental structures with which the system might be coupled at different points of time. A cognitive agent might be constituted simply by parts of the organismʼs biological body; but enactivists also allow for environmental objects to be incorporated within the living, cognitive agent. Extra-organismal elements can count as literal parts of the cognitive agent if they are incorporated into the operationally closed, selforganizing network by which the agent attains its identity and autonomy (Di Paolo 2009b; Thompson & Stapleton 2009). Adding to this “external” characterization of bodily extension, Thompson and Stapleton (2009) focus on the phenomenological experience of incorporating environmental elements into our bodies. They claim that, like the rest of our biological bodies, incorporated elements are “transparent,” they are that
16
by which we experience the world outside us, rather than being experienced as worldly objects. Accordingly, their “transparency constraint” specifies that “For anything external to the bodyʼs boundary to count as a part of the cognitive system it must function transparently in the bodyʼs sense-making interactions with the environment” (p. 29). In sum, enactivists offer at least two possible criteria for being a constituent of a cognitive agent: (1) being an element in the operationally closed network which constitutes the systemʼs as autonomous, and (2) being experienced by the agent transparently rather than as an environmental object. These criteria could be used to distinguish environmental factors that are or are not constitutive of the cognitive system, and thus avoid committing the coupling-constitution fallacy. But do these two criteria apply in cases of social cognition involving autonomous social interactions? That is, should the social cognition of an individual agent be described as constituted by social interactions? I will evaluate this with regard to the human and computational studies of perceptual crossing so often appealed to by enactivists as “proof of concept” (Torrance & Froese, 2011, p. 29) that social interaction can constitute social cognition. Since enactivists make no particular claims about the phenomenology of perceptual crossing, the autonomy-based criterion for constitution seems most appropriate here. Given their own account of autonomy, it seems accurate to describe cases of perceptual crossing as autonomous social interactions. The question, however, is how the autonomy at the social level relates to the individual-level autonomy of each interacting agent. Remember that, as defined by enactivists, social interactions do not destroy the autonomy of the interactors, even if the autonomy of the individual agents can enhanced or reduced by such interactions. Does perceptual crossing then involve an augmentation of the autonomy of each individual agent to include the other? As they themselves admit, the computational models of perceptual crossing do not involve full-blown agents, even if the modeled agents have some autonomy. But putting that aside, enactivists do not seem to describe perceptual crossing cases this way. They describe the way the autonomous social interaction (constituted by the two interactors) can influence the social cognition/sense-making of each individual agent—e.g., social interaction enables successful recognition of perceptual crossing by both agents when it is otherwise beyond the capabilities of each agent on their own. Is this enough to say that the cognitive system of each agent has incorporated the other agent? To do so
17
would seem to collapse the distinction between the level of each individual agent and the level of the social collective. For if we expanded the operationally closed network constituting each agent to include the other agent, we would end up with the two agents being constituted by the identical social-level system consisting of each of them. But this goes against enactivismʼs definition of social interaction as preserving the distinct autonomies of each agent. If social interaction maintains the autonomy of each individual agent, then the constitutive elements of each autonomous agent should not be fully coextensive. This proposal also implies social cognition is really a feature of a collective agent rather than either individual agent. But even enactivistsʼ characterizations of “joint sense-making” that is only possible within the context of social interaction (De Jaegher & Di Paolo 2007) seem to refer to the social cognition of individual agents rather than group minds. Could the constitutive elements of the two agents overlap without being fully coextensive? This would seem to require carving up the parts of each agent involved in social interaction, or saying each individual agent consists of multiple autonomous identities, one of which is incorporated into the social interaction (e.g., the cognitive system responsible for perceptual crossing) and others of which do not (e.g., the autopoietic system defining the agent as an organism), so that as a whole the autonomy of the agent is preserved despite partial incorporation with the other agent. At least Di Paolo (2009b) makes such distinctions between overlapping identities; but as he recognizes, this conception of multiple overlapping identities raises the issue of how to relate these identities. If the cognitive system engaged in perceptual crossing is not to be identified with the agentʼs organismal boundary, i.e., the autopoietic system, how exactly are these two systems related? Will it is make sense to say there is a single cognitive system across the many cognitive processes in which an agent engages? Di Paolo (2009b) seems to offer a way out of this whole problem of identifying the constituents of a cognitive system by clarifying that enactivism treats cognition as a relational property between an agent and its environment: “As relational in this strict sense, cognition has no location. It simply makes no sense to point to chunks of matter and space and speak of containment within a cognitive system” (p. 19). This account distinguishes between the “operational domain” of elements in the operationally closed network by which the agent is constituted as an adaptive autonomous system, and the “relational domain” of this
18
autonomous systemʼs interactions with its environment. I do not, however, see this relational view of cognition as avoiding the problem of demarcating the constituents of a cognitive system. Even if we define cognition as a relation, what prevents us from identifying the constituents of this relational domain? Indeed, it seems what enactivists define as a social interaction is just such an autonomous relational domain consisting of two interacting agents. Di Paolo (2009b) seems to restrict talk of “constitution” to the components of an autonomous system, and avoid it when discussing the autonomous systemʼs relations to its environment. But if relational domains are themselves autonomous systems, as in the case of social interaction, what warrants this restriction of constitution talk to one side of a relation, i.e., to an agent engaged in interaction? In sum, De Jaegher et al.ʼs (2010) taxonomy of explanatory factors does not adequately distinguish constitutive elements of a cognitive system from external elements that merely causally interact with a cognitive system. I have therefore tried to interpret what enactivists mean by a constitutive element of a cognitive system in a way that does not fall afoul of the coupling-constitution fallacy—a clear goal for enactivists themselves (Di Paolo 2009b; Thompson & Stapleton 2009). To do so I have identified autonomy-based and transparency-based criteria for constitution offered by enactivists. Only the first easily applies to the studies of perceptual crossing widely cited by enactivists as cases of social interaction constituting social cognition. I have argued, however, that appealing to autonomy is problematic in these cases, since there are multiple autonomous systems at issue: the operational autonomy of the individual agent versus the autonomous relational domain of the social interaction between agents. Even if one endorses the enactivistsʼ relational definition of cognition, one must pick which of these autonomous systems to focus on as the phenomenon of interest. The problem is that enactivists seem to shift between these two levels of organization when talking about social cognition, and use “constitutive element” sometimes to refer to internal parts and sometimes to refer to anything relevant to the explanation of the phenomenon of interest. This ambiguity is a barrier to understanding what enactivists mean by the claim that social cognition can be constituted by social interaction. One way of resolving this ambiguity while avoiding the coupling-constitution fallacy is to adopt the perspective of the “new mechanistic philosophy of science” developed over the last few decades by
19
philosophers of biology (e.g., Bechtel & Richardson 1993/2010; Bechtel & Abrahamsen 2005; Machamer et al. 2000) and cognitive science (e.g., Bechtel 2008; Craver 2007). In the next section I will show how mechanism can capture much of what enactivists want to say about the relations between the social and individual levels, without falling afoul of the coupling-constitution fallacy.
5.
Mechanism and the Explanation of Perceptual Crossing
The mechanistic approach has emerged over the last few decades as philosophers of science attempt to understand the investigative and explanatory practices of biologists and cognitive scientists. Theyʼve discovered that explanation in these sciences often takes the form of describing the mechanism responsible for a given phenomenon. A core feature of mechanistic explanation is thus decomposing a system into its parts and determining what operations those parts perform. But decomposition is not all there is to the mechanistic approach. Although the mechanistic approach is compatible with modular accounts localizing cognitive functions in single components, mechanists find that biological systems tend to involve a host of interacting parts (Bechtel 2009a). Mechanists thus emphasize “looking around” to how the various parts and operations of a mechanism are organized (Bechtel 2009b). In terms of the operations cognitive components are thought to perform, the mechanistic approach is compatible with (though not necessarily committed to) representationalism. Mechanism is thus compatible with cognitivist explanations of social cognition, such as explaining mental state attribution (e.g., performing a standard false-belief task; Baron-Cohen et al. 1985) in terms of internal neural mechanisms whose operations can be described in terms of theory theory and/or simulation theory. Mechanism does not, however, just focus on internal parts; it emphasizes also “looking around” to the environment in which the mechanism is situated (Bechtel 2009a,b). Like enactivists, Bechtel and Abrahamsenʼs (2010, 2011) dynamic mechanistic explanation emphasizes that tools of dynamical systems theory can be used to characterize the temporal organization of parts and operations in a mechanism and the mechanismʼs activity in its environment. This version of mechanism also shares with enactivism the conception of biological systems as autonomous, endogenously active systems.
20
There is clearly much common ground between mechanism and enactivism, particularly Bechtelʼs account of dynamic mechanistic explanation. What can the mechanistic approach offer in characterizing the perceptual crossing cases so emphasized by enactivists? I believe mechanism better emphasizes the explanatory roles of elements at multiple levels of organization—at social, individual/personal, and subpersonal levels—while distinguishing the constituents of the individual agent from environmental factors influencing the agent. In their attempt to distance themselves from methodological individualistic accounts of social cognition, enactivists sometimes underemphasize the role of internal mechanisms when identifying the “constitutive” explanatory role of social interactions. For example, when discussing Auvray et al.ʼs (2009) study of perceptual crossing, De Jaegher et al. (2010) write that “Only the interaction process fully explains the result” (p. 445). More generally, they say that “interactive processes can…even replace individual mechanisms” and that: Maximally, if we take seriously the idea that interaction can enable and constitute social cognition, we can conceive of interaction dynamics as, in some cases, delivering the necessary cognitive performance. There is no need to duplicate their effects by an individual [internal] mechanism. (p. 445) While elsewhere they admit that “Of course, individual factors enable this social performance” (p. 445), they tend not to offer integrated, multilevel accounts of social cognition, at each level distinguishing between constitutive and causal but non-constitutive factors. The mechanistic approach offers such a multilevel perspective, while also maintaining the system-environment distinction needed to avoid the couplingconstitution fallacy. Mechanists would describe perceptual crossing cases at the social level as involving an autonomous social network composed of the two interacting agents. The social network has its own dynamics which can be described in terms of stable states (perceptual crossing) and unstable states (agent-shadow and agent-fixed-object interactions). These properties of the social network can be explained by (a) decomposing the system into its parts—the agents and potentially other environmental objects—and determining how each part behaves, (b) examining how those parts are organized spatially and temporally to constitute the entire social network, and (c) determining how that network interacts with anything external to it (although in the perceptual crossing experiments this is not relevant).
21
To understand the behavior of a part of this social network, i.e., an individual agent, requires: (a) identifying within the agent the mechanism responsible for the agentʼs behavior—in the case of the computational models of perceptual crossing, the recurrent neural network controller; in the case of humans, the brain—and decomposing the mechanism into its constituent parts; as well as (b) looking up to how the mechanismʼs parts are organized spatially and temporally to constitute the whole phenomenon; and (c) looking around to the mechanismʼs environmental interactions. The phenomenon of interest here is the agentʼs ability to interact with the other agent rather than with other objects. The dynamics of the agentʼs interaction with the environment can be described, such that we know the relationships between environmental conditions and the behavior of the agent. For example, we observe that the agent engages in a general search strategy of moving in one direction until encountering an object, then reversing direction. To explain this search strategy employed by the agent will require decomposing the internal mechanism that mediates between sensory stimulation and motor behavior, as well as determining how those parts are organized to form the whole mechanism. Since the neural network controllers in Di Paolo et al. (2008) and Froese and Di Paolo (2010) were recurrently connected, there is not much specialization in the operations performed by the individual nodes of the network; so decomposing the controller and studying each node in isolation will not be especially helpful to explaining the agentʼs behavior. The controller as a whole should be understood as an endogenously active dynamical system: environmental influences modulate the endogenous activity of the neural network, rather than network only reacting to sensory inputs. Froese and Di Paolo (2010), for example, analyzed the internal dynamics of their threenode recurrent neural network controller as consisting of two attractor landscapes, which the network switches between depending on the presence or absence of sensory input, which is itself a function of both the agentʼs own behavior and the behavior of the other agent. Enactivists describe this environmental input as partly constituting the social behavior of the individual agent. But mechanists make a distinction between the internal mechanism that constitutes the phenomenon of interest (e.g., the search behavior of the agent) and environmental influences on the mechanism and its parts. Mechanists do this because they understand a mechanism as an autonomous system in the same sense as enactivists: as a system that produces and maintains its identity over time
22
through the self-organized activity of its internal parts. It is thus the autonomy of the system that marks the distinction between the parts constituting the mechanism and external elements causally interacting with the mechanism. In summary, when explaining cases of perceptual crossing, mechanists would say that agentʼs search strategy, which is constitutively explained by the nature of its neural network controller, only succeeds in finding and maintaining interaction with the other agent when the agent is situated in the appropriate social environment. And this appropriate social environment consists in an autonomous social interaction that is partly constituted by the individualʼs own actions, and partly constituted by the other agentʼs actions. The mechanistic framework thus emphasizes the import of elements at various levels of organization: the social level of interacting agents, the individual level of the embodied agent, the subpersonal level of the neural mechanism(s) inside the agent, and however far down one wants to go in decomposing this mechanism into its component parts. The mechanistic explanation of the individuals engaged in perceptual crossing is most consistent with the thesis of embedded cognition rather than the thesis of extended cognition. The cognitive system is here an autonomous system constituted by its internal parts that interacts with an external environment, rather than being a distributed or extended system crosscutting internal and external factors.
6.
Conclusion
The mechanistic approach has much in common with enactivism. Both: •
while recognizing the relevance of lower-level parts to explaining the activity of the whole, see the complex, non-aggregative organization of those parts as a crucial part of such explanations. They thus reject forms of reductionism that expect explanations to be purely in terms of the lower-level entities (e.g., Bickle 2003);
•
make the methodological claim that to understand cognition, cognitive scientists must study the environments in which agents are situated;
•
allow for the possibility that these social interactions can acquire an autonomous existence, which can influence the cognition of individual members of the network;
23
•
see the environmental complexity provided by social interactions as being able to account for complex social cognition without the need for similarly complex internal mechanisms;
•
define the boundaries between a system and its environment in terms of autonomy.
A main point of departure between them is that mechanism does not treat factors external to the agent as constitutive of the agentʼs social cognition. Instead, mechanism treats social cognition as the activity of an agent that is situated in a particular environment. The social behavior constituted by an agentʼs internal mechanisms will be successful in some environments but not others. These environments can exist independently of the actions of the agent, or be constructions of the agent. Enactivist accounts of social cognition provide fascinating cases where individual agents themselves partially constitute social interactions which influence their own social cognition. But from the mechanistic perspective, these social interactions do not themselves constitute the social cognition of individual agents. Since enactivism and mechanism seem to share an autonomy-based account of how to distinguish a cognitive system from its environment, I propose that enactivists could resolve the ambiguity in their talk of “constitution” by dropping the more radical claim that social interaction can constitute individual-level social cognition, and instead endorsing mechanismʼs embedded cognition perspective. If enactivists are to reject this mechanistic perspective, perhaps because they reject its compositional picture of levels, they owe us an account of how to understand their radical claim about the constitution of social cognition while avoiding the couplingconstitution fallacy.
References Adams, A., & Aizawa, K. (2001). The bounds of cognition. Philosophical Psychology, 14, 43–64. Adams, A., & Aizawa, K. (2008). Defending the bounds of cognition. Oxford: Basil Blackwell. Auvray, M., Lenay, C., & Stewart, J. (2009). Perceptual interactions in a minimalist virtual environment. New Ideas in Psychology, 27(1), 32-47. Baron-Cohen, S., Leslie, A. M. & Frith, U. (1985). Does the autistic child have a “theory of mind”? Cognition, 21, 37–46.
24
Bechtel, W. (2008). Mental mechanisms: Philosophical perspectives on cognitive neuroscience. New York: Routledge. Bechtel, W. (2009a). Explanation: Mechanism, modularity, and situated cognition. In P. Robbins & M. Aydede (Eds.), The Cambridge handbook of situated cognition (pp. 155-170). Cambridge: Cambridge University Press. Bechtel, W. (2009b). Looking down, around, and up: Mechanistic explanation in psychology. Philosophical Psychology, 22(5), 543-564. Bechtel, W., & Abrahamsen, A. (2005). Explanation: A mechanist alternative. Studies in History and Philosophy of Science Part C: Studies in History and Philosophy of Biological and Biomedical Sciences, 36(2), 421-441. Bechtel, W., & Abrahamsen, A. (2010). Dynamic mechanistic explanation: computational modeling of circadian rhythms as an exemplar for cognitive science. Studies In History and Philosophy of Science Part A, 41(3), 321-333. Bechtel, W., & Abrahamsen, A. (2011). Complex biological mechanisms: Cyclic, oscillatory, and autonomous. In C. A. Hooker (Ed.), Handbook of the philosophy of science: Vol. 10. Philosophy of complex systems. New York: Elsevier. Bechtel, W., & Richardson, R. C. (2010). Discovering complexity: Decomposition and localization as strategies in scientific research. Cambridge, MA: MIT Press. (Original 1993 edition published by Princeton University Press) Bickle, J. (2003). Philosophy and neuroscience: A ruthlessly reductive account. Dordrecht: Kluwer. Craver, C. F. (2007). Explaining the brain: Mechanisms and the mosaic unity of neuroscience. Oxford: Clarendon Press. De Jaegher, H. (2009a). Social understanding through direct perception? Yes, by interacting. Consciousness and cognition, 18(2), 535-42. De Jaegher, H. (2009b). What made me want the cheese? A reply to Shaun Gallagher and Dan Hutto☆. Consciousness and Cognition, 18(2), 549-550.
25
De Jaegher, H., & Di Paolo, E. (2007). Participatory sense-making. Phenomenology and the Cognitive Sciences, 6(4), 485-507. De Jaegher, H., Di Paolo, E., & Gallagher, S. (2010). Can social interaction constitute social cognition? Trends in Cognitive Sciences, 14(10), 441-447. De Jaegher, H., & Froese, T. (2009). On the role of social interaction in individual agency. Adaptive Behavior, 17(5), 444-460. Di Paolo, E. A. (2005). Autopoiesis, adaptivity, teleology, agency. Phenomenology and the Cognitive Sciences, 4(4), 429-452. Di Paolo, E. (2009a). Editorial: The social and enactive mind. Phenomenology and the Cognitive Sciences, 8(4), 409-415. Di Paolo, E. (2009b). Extended Life. Topoi, 28(1), 9-21. Di Paolo, E. A., Rohde, M., & De Jaegher, H. (2011). Horizons for the enactive mind: Values, social interaction, and play. In J. Stewart, O. Gapenne, & D. Paolo (Eds.), Enaction: Towards a new paradigm for cognitive science (pp. 33-88). Cambridge: MIT Press. Di Paolo, E. A., Rohde, M., & Iizuka, H. (2008). Sensitivity to social contingency or stability of interaction? Modelling the dynamics of perceptual crossing. New Ideas in Psychology, 26(2), 278-294. Froese, T., & Di Paolo, E. A. (2009). Sociality and the life–mind continuity thesis. Phenomenology and the Cognitive Sciences, 8(4), 439-463. Froese, T., & Di Paolo, E. (2010). Modeling social interaction as perceptual crossing: an investigation into the dynamics of the interaction process. Connection Science, 22(1), 43-68. Fuchs, T., & De Jaegher, H. (2009). Enactive intersubjectivity: Participatory sense-making and mutual incorporation. Phenomenology and the Cognitive Sciences, 8(4), 465-486. Gergely, G., & Watson, J. S. (1999). Early socio-emotional development: Contingency perception and the social-biofeedback model. In P. Rochat (Ed.), Early social cognition: Understanding others in the first months of life (pp. 101–136). Mahwah, NJ: Erlbaum. Huebner, B. (2008). Do you see what we see? An investigation of an argument against collective representation. Philosophical Psychology, 21(1), 91-112.
26
Hutchins, E. (1995). Cognition in the wild. Cambridge: MIT Press. Machamer, P., Darden, L., & Craver, C. F. (2000). Thinking about mechanisms. Philosophy of Science, 67, 1-25. McGann, M., & De Jaegher, H. (2009). Self–other contingencies: Enacting social perception. Phenomenology and the Cognitive Sciences, 8(4), 417-437. Murray, L., & Trevarthen, C. (1985). Emotional regulation of interactions between two-month-olds and their mothers. In T. Field & N. Fox (Eds.), Social perception in infants (pp. 177-197). Norwood, NJ: Ablex. Rupert. (2004). Challenges to the hypothesis of extended cognition. Journal of Philosophy, 101(8), 389328. Steiner, P., & Stewart, J. (2009). From autonomy to heteronomy (and back): The enaction of social life. Phenomenology and the Cognitive Sciences, 8(4), 527-550. Thompson, E. (2007). Mind in life: Biology, phenomenology, and the sciences of mind. Cambridge: Harvard University Press. Thompson, E., & Stapleton, M. (2008). Making sense of sense-making: Reflections on enactive and extended mind theories. Topoi, 28(1), 23-30. Torrance, S., & Froese, T. (2011). An inter-enactive approach to agency: Participatory sense-making, dynamics, and sociality. Humana.Mente, 15, 21-53. Varela, F. J., Thompson, E., & Rosch, E. (1991). The embodied mind: Cognitive science and human experience. Cambridge, MA: MIT Press.
27
