Voluntary Behavior in Cognitive and Motor Tasks

Many previous treatments of voluntary behavior have viewed intentions as causes of behavior. This has resulted in several dilemmas, including a dilemma concerning the origin of intentions. The present article circumvents traditional dilemmas by treating intentions as constraints that restrict degrees of freedom for behavior. Constraints self-organize as temporary dynamic structures that span the mind-body divide. This treatment of intentions and voluntary behavior yields a theory of intentionality that is consistent with existing findings and supported by current research. 1. Dilemmas with Traditional Views of Intention Extensive catalogs of empirical data stand behind contemporary theories of mind, body, and behavior. However, one important fact, common to all experiments with human participants, is rarely considered: Before meaningful data can be collected, a participant’s intentions to perform as instructed must be solicited (Lindworsky 1923). In other words, the backbone of what scientists know about mind and behavior – experimental data – depend fundamentally on the will, purpose, and goals of participants. But despite its prevalence, an agreed-upon coherent understanding of intentionality is not available. This is not for lack of data, given that every study of human behavior solicits human intentions. Instead, theoretical progress has been hamstrung in the dilemmas of a dualist mind-body separation. The current essay addresses these dilemmas under the umbrella of complexity science, emphasizing intentions as constraints on emergent behavior. We start by describing the dilemmas inherited from the perspective of mind-body separation. We then provide a view of intentionality that transcends the mind-body divide. Along the way, we explain how contents of intentions accomplish selective attention, and how voluntary control reveals its presence in behavior. Consistent with the masthead Mind and Matter, mental events have abundant and reliable material consequences for the body and behavior 20 Kloos and Van Orden (Markman et al. 2005), and they permeate all human activities (cf. Gibbs 1994, Lyons 1995, Mandler 1997, Mitchell et al. 2009). The mere expectation of an adverse effect can lead to strong bodily reactions (known as nocebo effects; e.g., Barsky et al. 2002). And mental imagery can improve sports performance, especially when images are enacted repeatedly (Wakefield and Smith 2009, Weinberg 2008). Intentions are one kind of those mental events – with similarly remarkable effects on the body. For example, an intention to act results in muscle tensions across the body, including “heightened tonicity of the reactive mechanisms ... widespread contraction of skeletal muscles ... marked changes in breathing, heart rate, and vascular processes ... and an increased readiness of arousal for associations within a given sphere” (Bills 1934, p. 408). Even more striking effects of intentions come from classical-conditioning research on autonomous bodily reactions of human participants (for a summary see Woodworth and Schlosberg 1954). In a Pavlovian-type paradigm, a sound was paired with eating food, while the amount of secreted saliva was measured. Or a light was paired with an electric shock, while psycho-galvanic reflexes were measured. Notwithstanding the fact that people can be classically conditioned, neither conditioning nor extinction proceeded in a regular fashion. For example, the conditioned sound sometimes led to a decrease, rather than an increase, in the level of saliva secretion, depending on the intentional stance (Razan 1935). Or in the case of shock conditioning, the conditioned reflex in response to the light disappeared immediately after the participant was told there would be no more shocks (Cook and Harris 1937, Mowrer 1938). Changes in the intentional set, in effect, turned the participant into a differently conditioned animal (Razan 1939). Finally, the intentional stance in one task can infect performance of another task that requires a different intention altogether (for a summary see Bills 1943). For instance, performing a task that demands high accuracy will improve accuracy in a subsequent task. Likewise, performing a task that emphasizes speed will speed up performance in a second task. Even intended rhythm is infectious across tasks. A skilled pianist, for example, after playing a composition with a fast tempo, will then play a slower-tempo composition closer to the pace of the first piece, or vice versa (Cathcart and Dawson 1928). It is even possible to fatigue the participant’s goal to pay attention in voluntary behavior and transfer that fatigue to a second task. Immediate effects of intentions on bodily reactions create dilemmas for theories of mind and body because theories lack the necessary bridging concepts to connect mind and body (Mandler 1984). How does intentionality affect a body reflex that lives outside of voluntary control (cf. Fearing 1970)? And how does the intentional stance elicited for one task reappear in the performance of a different task? The traditional solution to these Voluntary Behavior in Cognitive and Motor Tasks 21 dilemmas is to assume a special causal status for intentions: a capacity of intentions to bring into existence behavior across the mind-body divide, and from one task to another. There are several problems with this view, however, whether on methodological, theoretical, or empirical grounds. On a methodological level, the muscle tensions that result from intentions will vary in quality from person to person and from task to task. While muscle tensions can reliably predict a host of factors (e.g., task difficulty, fatigue, quality of performance, even broad content of thought; Woodworth and Schlosberg 1954), idiosyncratic muscle tensions across participants and tasks prevent aggregating data to discover common causes. In other words, methodological attempts to isolate the causal power of intentions are bound to fail, given the idiosyncratic variability in embodied content from person to person, or task to task (cf. Molenaar 2008). On a theoretical level, conceiving of intentions as causal entities inevitably raises the question of the cause of intention. If the origins of nocebo effects, effects of mental imagery, paradoxes of conditioning, and the embodiment of the intentional stance are explained by an intention, then what caused the intention in the first place? Intentions must either acquire a magical status, as prime movers, or we enter an endless logical regress of seeking the cause of the cause of the intention to behave (Juarrero 1999). Finally, on an empirical level, if the origin of voluntary behavior is explained by intentions, it is not clear why intentions sometimes have so little effect on behavior. It is well known, for example, that a dieter forbidding himself to eat his favorite non-diet foods will likely fail with that diet solution (Baumeister and Heatherton 1996, Rachlin 2000). In general, mindful, forbidding self-control is notoriously difficult to put into action, leaving us vulnerable to temptation (Nordgren et al. 2009). If intentions are prime movers, why do they fail to move us when it matters? One solution to the dilemmas is to deny intentions any status in bringing about behavior. In fact, experiments are frequently thought to tap into involuntary, automatic, or unconscious processes exclusively (Science Watch 1999). This solution not only ignores the role of participants’ intentions in data collection (Vollmer 2002), but it also creates new dilemmas. Most prominent is the mind’s ability to attend selectively to relevant factors. Take for example the well-known finding that participants – instructed to focus on the ball handling in a basketball game – fail to notice a man in a gorilla suit pounding his chest on the basketball court (Simons and Chabris 1999). How could such striking selective attention be explained without reference to the focus of the participant or the intentions to perform as instructed? In other words, how does the mind stay open to the outside factors that are necessary to promote intended goals, while at the same time ignoring irrelevant factors that might derail them? 22 Kloos and Van Orden In sum, the mind-body divide has led to dilemmas about how to understand the role of intentions in behavior. Conspicuous in the protracted failure to connect mind to body, intentions can either take the role of functional causes of behavior; or they become irrelevant facets without causal impact. Neither of these options fit the existing data – data that show strong influences of intentions. In the remainder of the essay, we describe a way to bridge the mind-body divide that escapes these dilemmas. To build this bridge, we borrow concepts from complexity science, developed over decades of research with living and non-living nonlinear systems. 2. Intentions as Constraints Complexity science offers a framework for an integrated understanding of intentionality, one that avoids isolable functional causes of cognitive activities (for contemporary issues of the functional view, see Bechtel 2009, Lyons 1995). We align ourselves with the idea that intentions are not effective causes in the sense of billiard-ball causality, but function instead as constraints in emergent coordination (e.g., Juarrero 1999, Riley and Turvey 2001). To unpack this claim, we first discuss constraints and emergent coordination more generally and then turn to how control parameters bring about selective attention. 2.1 Constraints and Emergent Coordination Constraints are relations among a system’s components that reduce the degrees of freedom for change. An intuitive example of constraints comes from the arrangement of muscles and bones across the skeleton. Similar to tensegrity structures in architecture and robotics (e.g., Tomassian 1997, Tur and Juan 2009), the skeleton supplies struts, while the muscles (ligaments and fascia) form tension lines, which together eliminate slack across the musculoskeletal structure (Levin 2002). The limits on degrees of freedom of this tautly poised arrangement limit the range of motion of body parts, such that they can move in some directions, but not others. Other examples of constraints on body motion, less constant than musculoskeletal tensegrity, are temporary coordinative structures (Turvey 1990). They comprise webs of constraints across the body, which constrain how the parts of the body will change together, in coordination. Playing tennis, for instance, is constrained by coordinative structures to run for the ball, forehand shots, backhand shots, serves, and return a serve. And swimming is constrained by coordinative structures to enact the strokes of swimming. The web of constraints of a coordinative structure delimits the possibilities for coordinated movement of the body in the actions at Voluntary Behavior in Cognitive and Motor Tasks 23 hand. As a result, the coordinative structure will temporarily constrain the body to move in some ways but not others. Like coordinative structures, intentions can be conceived as temporary sources of constraint that concern the specific needs and goals of an actor. For example, the instructions about how to act as a participant in an experiment are temporary sources of constraint that the participant takes on. Thus intentions contribute self-control by limiting the options for behavior to suit the immediate requirements of the task. They are temporary dynamical structures that emerge to constrain mind and body and sustain purposeful behavior. In this sense, intentions are ordinary ingredients of nature, as commonplace as causes. It even becomes plausible that emergent structures of physical systems express a kind of proto-intentionality or proto-mentality (Shaw 2001). Temporary dynamical structures have several important features that apply to intentions. In particular, a temporary structure constrains local interactions at the same time as the local interactions sustain it. A model physical system to intuit this constrain-sustain feature is a layer of fluid, heated from below (e.g., Kelso 1995). At a critical difference between heat coming in at the bottom and heat going out at the surface, the fluid molecules self-organize into orderly Bénard convection cells to transport heat through the fluid (Nicolis 1989). A hexagon pattern forms across the surface of the fluid such that each cell of the hexagon circulates fluid molecules in a direction opposite from its neighbors. This pattern of convection cells constrains the motions of its component molecules (i.e., each molecule moves in the direction of the convection roll that it happens to be part of). At the same time, the interdependence of motion across the entire fluid is sustained by local interactions among the motions of neighboring molecules. This relation between the hexagon of Bénard convection cells and the molecules that sustain them is called a strange loop. It refers to the constrain-sustain relation in which local mutually reinforcing motions sustain an emergent global structure in their collective activity, which “loops” back to constrain the local motions within the global structure. Such a relation is present in all strongly emergent phenomena. In fact, interdependent strange-loop behavior is at the heart of complexity science. Strange loops short-out the logical regress of searching for ultimate causes because local changes are constrained by the global coordination they sustain (Juarrero 1999). Thus temporary dynamical structures respect the local physics of cause and effect while, at the same time, they acquire dynamical properties in their global organization (that constrain the local dynamics). An intention is analogous to the emergent coordination of a Bénard cell, part of a temporary coordinative structure of the entire system. As a Bénard cell self-organizes within a hexagonal pattern, spanning the system 24 Kloos and Van Orden of heat and fluid, so do intentions self-organize within dynamics spanning the system of mind, body, and context. Like Bénard cells constrain the movements of molecules, intentions constrain the changes in their embodied elements. And like Bénard cells are sustained in the interdependent interactions of molecules, intentions are sustained by interdependent interactions among their embodied elements. In both systems, elements change on faster timescales than the coupling that sustains the pattern emerging among them. And in both systems, the emergent pattern constrains the local behavior of the elements. 2.2 Control Parameters and Critical States Relevant constraints are summarized in control parameters. In a simplified illustration of the heated fluid, a control parameter is the ratio between heat entering the bottom layer of the fluid, and heat dissipated at the surface: simplified fluid dynamics : incoming heat outgoing heat (1) This ratio of incoming heat and outgoing heat predicts the observed global behavior of the molecules. When the ratio is less than 1, disorderly movements of molecules are sufficient to dissipate heat. And if the ratio becomes greater than 1 (all else equal), orderly Bénard cells emerge. Applied to human behavior, a control parameter has been illustrated in infants’ voluntary stepping behavior (Thelen and Smith 1994). Sources of constraint include the weight of an infant’s leg and the strength of the infant’s leg muscles. The control parameter combines these constraints, again in a ratio: simplified stepping : weight of leg strength of leg (2) Weight of the leg is in the numerator, and strength of the leg is in the denominator. Stepping behavior is possible when the strength of the leg exceeds its weight, and stepping behavior disappears when the weight of the leg exceeds its strength, which correctly predicts typical and atypical patterns of development. In more general terms, the numerator of the control parameter – e.g., leg weight – summarizes constraints that embed the infant in her environment. Such embedding constraints delimit affordances, the dispositions of the surrounding environment that are directly relevant for action (Gibson 1979). Conversely, the denominator – e.g., leg strength – concerns embodied constraints of the actor, constraints supplied by the body itself. Embodied constraints delimit effectivities, the capacities and capabilities of the actor to exploit the available affordances (Shaw et al. 1982). Thus, Voluntary Behavior in Cognitive and Motor Tasks 25 the control parameter of voluntary stepping captures the relations between the infant and its environment: simplified behavior : affordances effectivities (3) Working out the details of the control parameters allows us to discuss the concept of critical states, a concept with special significance for our understanding of intentions. When numerator and denominator of the ratio are equal, the control parameter reaches a critical value, and the system enters a critical state. In the example of the heated fluid, the critical value is reached when the incoming heat equals the outgoing heat. The opposing actions that are equally available in this state are random dispersion vs. clockwise or counter-clockwise movement within convection cells. In infant stepping, the critical value is reached when the pull of gravity exactly equals leg strength. In this state, two opposing behaviors (e.g., stepping and not stepping) are in precise balance, and therefore equally likely. The simultaneous presence of opposing actions creates a symmetry that can be broken by the smallest perturbation. Even tiny changes – with seemingly miniscule causal power – can tip the balance of the poised alternatives and enact behavior. In the heated-fluid analogy, just before molecules self-organize as convection rolls, any relevant contingency, even a single molecule’s movement, can determine whether a particular Bénard cell will roll clockwise or counter-clockwise within the global pattern. And in the example of body motion, a very local change in musculoskeletal position can be amplified through the tensegrity structure and, as a result, change the movement of the entire body (Carello et al. 2008, Turvey 2007, Turvey and Fonseca 2008). Even the spontaneous contraction of a single muscle at the critical point of movement may bring about a movement that was not possible before. Selective attention, one of the dilemmas of intentionality, is resolved in critical states. This is because critical states can only be perturbed by events that favor an available action alternative. For example the next meal of a hungry dieter may be enacted by finding a candy bar, but not by finding a toy car. The toy car is not sufficiently relevant to the specified critical state, like the color of her mother’s blouse may be irrelevant to a baby’s stepping behavior. In this way, critical states allow the actor to stay open to even the smallest changes in events relevant to the critical state, without being captured by irrelevant contingencies. Taking this idea a step further, it is the intentional content of the critical state that determines which contingencies may sway the system one way or another (cf. Mandler 1984, 1997). A dieter’s focus on healthy versus unhealthy edible things creates a critical state in which foods – whether healthy or unhealthy – become part of the relevant contingencies. 26 Kloos and Van Orden It therefore leaves the dieter susceptible to eating candy bars despite the intention not to. A more effective strategy might be to concentrate on more abstract end-goals of dieting, such as personal wellbeing, to better disconnect candy bar contingencies from critical states of behavior (Fujita and Han 2009). Before a contingency can enact behavior, the system must already be in a relevant critical state. Available constraints must first specify propensities to act. Only then do contingencies have the power to cause behavior. Absent a relevant critical state, neither voluntary nor involuntary behavior will occur. And once a relevant contingency occurs, the specific critical state ceases to exist, dissipating the causal powers of the contingency. Eating a candy bar, for example, may rob the next candy bar of the power to enact behavior. Instead, new constraints for behavior emerge. 3. Evidence of Complexity in Human Performance We have suggested that intentions are best conceived as temporary dynamic structures. Intentions are emergent constraints that span the mind-body divide and shape the critical states that anticipate purposeful behavior. But how would we know that voluntary performance is the product of emergent coordination? The answer lies in the fact that emergent coordination requires positive feedback among a system’s components (e.g., Camazine et al. 2001). Positive feedback of this sort predicts a specific pattern of intrinsic variation in measurements of a system’s behavior, known as a scaling relation (Van Orden et al. 2003). In this section, we elaborate on these ideas. 3.1 Scaling Relations In an emergent coordinative structure, changes in any relevant part of the body are correlated with changes in every other relevant part. This appears as long-range correlations in a repeatedly measured human performance. A scaling relation reflects such long-range correlations. It pertains to the relation between the size of changes in repeated measurements and how often changes of that size occur. The scaling relation at issue has been called pink noise, but has also been referred to as flicker noise, 1/f noise, 1/f scaling, multiplicative noise, edge of chaos, fractal time, longrange correlations, red noise, or self-affinity. The many names reflect the many phenomena and disciplines in which scaling relations have been observed. We will use the term “pink noise” throughout (but see Ihlen and Vereijken 2010). Pink noise can be portrayed in a spectral plot that results from decomposing a data series into sine waves of different amplitudes. Figure Voluntary Behavior in Cognitive and Motor Tasks 27 ! l l log frequency reaction times lo g po w er de co m po se d re ac ti o n ti m es de co m po se d re ac ti o n ti m es trial number ti m es de co m po se d re ac ti o n ti m es Figure 1: upper right – reaction times of one subject versus trial number; left – reaction times decomposed into sinusoidal components of different wavelength; lower right – spectral plot of reaction times with an average slope of −0.94 and four marked points referring to the sinusoidal components indicated. 1 shows such a data series (top right) and how it can be decomposed into sine waves of particular amplitudes. Slow changes in the data series are captured by low-frequency high-amplitude sine waves (top left of Fig. 1), and fast changes are captured by high-frequency low-amplitude waves (bottom left of Fig. 1). A power spectrum is then constructed, with relative amplitude on the vertical axis, and frequency f of change on the horizontal axis (on log-log scales). The amplitude represents the relative size of change S(f), also referred to as power. The slope of the regression line in the spectral plot defines the scaling relation between amplitude and frequency. In Fig. 1, the size of changes S(f) is inversely proportional to their frequency f : S(f) = 1/f = f−α, with scaling exponent α ≈ 1, the scaling exponent of pink noise. Pink noise is so commonly observed in cognitive and motor task performance that it has been claimed to capture a universal feature of human performance (for reviews see Gilden 2001, 2009, Kello and Van Orden 2009). Furthermore, the universality of pink noise beyond human performance supports the idea that common dynamical organizations appear in systems of different material construction, in living as well as nonliving matter. However there is some difficulty related to pink noise. This is 28 Kloos and Van Orden because pink noise is both a regular and an irregular phenomenon: Regularity is seen in the stable scaling relation of a power spectrum, while irregularity is seen in the unstructured aperiodic waveform in a data graph. In truth, empirical pink noise is neither regular nor irregular. Instead, it is a strongly nonlinear pattern that exists between the two extremes (Nicolis and Rouvas-Nicolis 2007, Sporns 2007, Tsonis 2008). A physical system provided the analogy for our interpretation of pink noise: avalanches of sand or rice piles. In actual experiments, sand granules were dropped, one at a time, to build a pile in which eventually the next dropped granule triggered an avalanche (for a review see Jensen 1998). Time between avalanches of different sizes was measured repeatedly. Initial results showed that sand pile avalanches never became sufficiently large to reproduce the very large avalanches predicted in a scaling relation. They instead yielded overly random avalanche behavior driven by the inertia of sand granules. Results changed after grains of sand were replaced with rice kernels (Frette et al. 1996). The rice kernels varied in their aspect ratio of kernel length to kernel width. Kernels of low aspect ratio (less surface area) behaved like sand; too little friction resulted in over-random avalanche behavior. Kernels of higher aspect ratio (more surface area) allowed more friction among kernels, and thus more regular avalanche behavior. More friction made it possible for small piles of rice to form throughout the larger pile, at or near their toppling threshold. With so much rice poised to topple, rice piles could produce the rare large avalanches to fill out a scaling relation between size S(f) and frequency f of avalanches. Taking the riceand sand-pile results together, granules with too little friction were too strongly governed by inertia. Although inertia itself is a highly regular phenomenon – i.e., the tendency of a particle to maintain its current state trajectory – inertia is a source of over-random behavior in the case of avalanches. This is because too little friction, relative to inertia, minimizes the tendency to build local structure in a sand pile, necessary for scaling behavior. In order to capture granules in local piles, poised near their toppling threshold, more friction is needed, relative to inertia. On the other end of the spectrum, too much friction (or too little inertia) would produce piles that are too coherent and too over-regular in their behavior. This is the case in a mud-pile in which rare largeavalanche mudslides dominate behavior. Only when friction and inertia are in relative balance do avalanches reveal scaling relations. The tradeoff between friction and inertia can again be represented as a control-parameter ratio: variation in avalanche behavior : inertia friction = over-random over-regular (4) Inertia contributes to over-random behavior, and friction between granVoluntary Behavior in Cognitive and Motor Tasks 29 ules contributes to over-regular behavior. When these two factors are in balance, a pink-noise scaling relation can be observed. As such, pink noise is neither over-random nor over-regular but balanced between the two. Applied to cognitive and motor performance then, pink-noise scaling relations might be indicative of a mind-body-context system that is poised at a critical state in which over-random tendencies are balanced with over-regular tendencies. 3.2 Attraction to Critical States As we discussed earlier, a balance between numerator and denominator in the control parameter is constitutive of critical states – a state in which opposing options are available simultaneously. And so far, we have shown how a pink-noise scaling relation is also a form of balance, namely between over-random and over-regular tendencies. A new dilemma arises, however. Absolute symmetry in critical states is behaviorally unstable, given that the smallest relevant contingency will collapse the symmetry. Yet pink noise is common in nature. How can empirical critical states be unstable while at the same time be associated with behavior as commonly observed as pink noise? The solution is a system that is attracted toward critical states (Bak 1997, Bak et al. 1987). What is the evidence that intentional acts are poised at criticality? Suggestive evidence comes from speech experiments in which participants repeated the same word again and again (Kello et al. 2008). Each recorded instance of the word was then parsed identically into dozens of frequency bins, and the amplitude of each frequency-bin was tracked across a participant’s tokens of the spoken word. This resulted in dozens of separate data-series per participant, each with a spectral exponent. Aggregating the estimated scaling exponents in a histogram revealed a normal Gaussian distribution with a central tendency near the pink-noise scaling exponent of α ≈ 1. In other words, taking into account variations in scaling exponents, their central tendency in a repetitive speech task appears to be pink noise. More direct evidence for attraction towards pink noise was observed as adults gained practice with a Fitt’s tracing task (Wijnants et al. 2009). Participants used a stylus to repeatedly trace between two dots on an electronic tablet. The measurement was the time required to trace from one dot to the other, yielding a trial series of trace times. Across practice blocks, the central tendency of spectral plots approached α ≈ 1 of pink noise. Interestingly, in the earliest practice block, the trace-time exponents were distributed below α ≈ 1, reliably toward α ≈ 0 (see Figure 2). A scaling exponent of α ≈ 0 would reflect white noise, an over-random coordination in which changes of every size S(f) are equally frequent. As participants acquired practice across blocks, over-random whiter noise gradually approached pink noise. 30 Kloos and Van Orden