Kelso, J. A. Scott

The mobile conjugate reinforcement (MCR) paradigm, made famous by Carolyn Rovee-Collier and her colleagues (Rovee & Rovee, 1969), has long been used to study infant learning and memory. In MCR studies, the infant's foot is tethered to a mobile hanging overhead, and the mobile responds directly to the infant's kicking. Infant kicking rate triples within a few minutes of interacting with the mobile. This result was classically interpreted as evidence of reinforcement learning. Kelso and Fuchs (2016) reinterpreted it as evidence that a coordinative structure, or functional synergy, forms between infant and mobile, triggering a positive feedback loop between the two. Positive feedback is proposed to give rise to an `Aha!' moment as the (prelinguistic) infant suddenly realizes it is an agent in control of the mobile's motion. While some have theorized the realization of self as causal agent emerges from organism-environment interactions, Kelso and Fuchs (2016) developed a mathematical model of the coordination dynamics between the infant and mobile, providing mechanistic explanations for the formation of agency. The current study was the first to measure movement of the mobile and analyze how dynamics of coordination between infant and mobile relate to possible transitions from spontaneous to intentional action. Novel measures of infant and mobile dynamics were used to test model predictions. Infant activity dropped drastically in response to non-contingent mobile movement and remained suppressed at the start of infant~mobile contingency, suggesting that mobile movement triggers a qualitatively different context for infants. This finding challenges the widely held assumption that mobile movement rewards and stimulates infant movement and calls into question the sufficiency of standard contingency detection cut-offs and explanations of conjugate reinforcement learning. Assessing coordination dynamics on a fine time scale using new analytical techniques made it possible to identify moments of agentive realization. Approaching agency as a relational phenomenon allowed for detailed characterization of the infant~mobile relationship and its role in the emergence of causal agency. In addition, the results revealed a number of surprising insights into agency formation such as the critical role of inactivity for agentive discovery and the possibility of intermediary stages or quasi-agentive states.

A fundamental question in Complexity Science is how numerous dynamic processes
coordinate with each other on multiple levels of description to form a complex
whole - a multiscale coordinative structure (e.g. a community of interacting people,
organs, cells, molecules etc.). This dissertation includes a series of empirical, theoretical
and methodological studies of rhythmic coordination between multiple agents
to uncover dynamic principles underlying multiscale coordinative structures. First,
a new experimental paradigm was developed for studying coordination at multiple
levels of description in intermediate-sized (N = 8) ensembles of humans. Based
on this paradigm, coordination dynamics in 15 ensembles was examined experimentally,
where the diversity of subjects movement frequency was manipulated to induce
di erent grouping behavior. Phase coordination between subjects was found to be
metastable with inphase and antiphase tendencies. Higher frequency diversity led
to segregation between frequency groups, reduced intragroup coordination, and dispersion
of dyadic phase relations (i.e. relations at di erent levels of description).
Subsequently, a model was developed, successfully capturing these observations. The
model reconciles the Kuramoto and the extended Haken-Kelso-Bunz model (for large- and small-scale coordination respectively) by adding the second-order coupling from
the latter to the former. The second order coupling is indispensable in capturing
experimental observations and connects behavioral complexity (i.e. multistability) of
coordinative structures across scales. Both the experimental and theoretical studies
revealed multiagent metastable coordination as a powerful mechanism for generating
complex spatiotemporal patterns. Coexistence of multiple phase relations gives rise
to many topologically distinct metastable patterns with di erent degrees of complexity.
Finally, a new data-analytic tool was developed to quantify complex metastable
patterns based on their topological features. The recurrence of topological features
revealed important structures and transitions in high-dimensional dynamic patterns
that eluded its non-topological counterparts. Taken together, the work has paved the
way for a deeper understanding of multiscale coordinative structures.

Emotion and coordinated movement complimentarily depicts our social experiences.
How is motion colored? This study investigates variations in emotional responses during social
coordination. Subjects were instructed to coordinate their finger movement with a Virtual Partner
(VP), whose homologous movement was displayed as a video on the computer screen. The
partner was driven by the Haken-Kelso-Bunz equations, an empirically validated model that
captures behavioral and social coordination. It has been shown that people perceive VP as an
intentional human agent. In each of 80 trials, subjects coordinated for 8 sec inphase or antiphase
with VP, and then rated the partner’s intention (cooperation -VP intend same coordination
pattern as human-, or competition) and subjective response to a Turing test of partners’
humanness. VP cooperated for half of the time, and could change its intention in the middle of a
trial. Skin potential response (SPR) quantified the intensity of emotional responses. After
validating the SPR measurements, we compared emotional responses by coordination pattern,
cooperative~competitiveness, and humanness attribution. Subjects experienced higher emotional
responses when they believed that their partner was human. This was observed both during
coordination (ANOVA, p=0.020), and during rating (p=0.012). Furthermore during the rating
period, higher emotional responses were found for cooperative behavior (p=0.012), modulated
by VP’s change of intention and coordination pattern. This study suggests that emotional
responses are strongly influenced by features of the partner’s behavior associated with
humanness, cooperation and change of intention. Implications for mental health (e.g. autism) and
design of socially cooperative machines will be discussed.

How one behaves after interacting with a friend may not be the same as before
the interaction began What factors a ect the formation of social interactions
between people and, once formed, how do social interactions leave lasting changes on
individual behavior? In this dissertation, a thorough review and conceptual synthesis
is provided Major features of coordination dynamics are demonstrated with
examples from both the intrapersonal and interpersonal coordination literature that
are interpreted via a conceptual scheme, the causal loops of coordination dynamics
An empirical, behavioral study of interpersonal coordination was conducted to
determine which spontaneous patterns of coordination formed and whether a remnant
of the interaction ensued ("social memory") To assess social memory in dyads, the
behavior preceding and following episodes of interaction was compared In the
experiment, pairs of people sat facing one another and made continuous flexion-extension finger movements while a window acted as a shutter to control
whether partners saw each other's movements Thus, vision ("social contact") allowed
spontaneous information exchange between partners through observation Each trial consisted of three successive intervals lasting twenty seconds: without social contact
("me and you"), with social contact ("us"), and again without ("me and you")
During social contact, a variety of patterns was observed ranging from phase coupling
to transient or absent collective behavior Individuals also entered and exited social
coordination differently In support of social memory, compared to before social
contact, after contact ended participants tended to remain near each other's
movement frequency Furthermore, the greater the stability of coupling, the more
similar the partners' post-interactional frequencies were Proposing that the
persistence of behavior in the absence of information exchange was the result of prior
frequency adaptation, a mathematical model of human movement was implemented
with Haken-Kelso-Bunz oscillators that reproduced the experimental findings, even
individual dyadic patterns Parametric manipulations revealed multiple routes to
persistence of behavior via the interplay of adaptation and other HKB model
parameters The experimental results, the model, and their interpretation form the
basis of a proposal for future research and possible therapeutic applications

In this thesis the transition region between two modes of behavior is explored using a novel technique, delayed feedback, and a variety of dynamical systems measures. In a previous study, Engstrom, Kelso, and Holroyd (to appear) established the existence of a transition between anticipatory and reactive behavior in a sensorimotor coordination task as a control parameter (frequency) was varied. Here, in order to explore the hypothesis that the behavioral dynamics during this transition are intermittent in character, subjects were asked to synchronize with a metronome that was actually a delayed copy of their own response pattern. The use of delayed feedback was expected to destabilize the behavioral dynamics enough to allow the observation of hypothesized intermittent phenomena. Use of delayed feedback was shown to destabilize synchronization, resulting in the emergence of a new behavioral pattern in the transition region that exhibited complex "bursting" dynamics. Analysis revealed that this bursting behavior displays many of the characteristics common to intermittency, which supports the idea that the anticipation-reaction transition is the result of a neurobehavioral dynamical system losing stability. Living in the vicinity of instabilities may be an important mechanism for biological organisms to maintain both flexibility and stability.

The aim of this research was to study the coordinative dynamics of multijoint arm movements as a function of forearm spatial orientation. Six subjects rhythmically coordinated flexion and extension of the right elbow and wrist under the following conditions: (1) forearm supine: wrist flexion/elbow flexion and vice versa; and (2) forearm prone: wrist flexion/elbow extension and vice versa. Starting in either pattern, subjects rotated the forearm in eight 20 steps, producing 15 cycles of motion at a frequency of 1.25 Hz. Switching from pattern (1) to pattern (2) and vice versa was observed at a critical spatial orientation. The critical point depended on the direction of forearm rotation, thus revealing the hysteretic nature of the switching. En route to the transition, regardless of direction of change, critical fluctuations and critical slowing down were observed in the relative phasing between the joints. Such results provide definitive evidence that relative phase is a viable order parameter, spatial orientation a relevant control parameter and loss of stability the chief mechanism leading to observed changes in coordination.

Humans are often faced with tasks that require stabilizing inherently unstable situations. We performed four experiments to explore the nature of functional stabilization. In Experiment 1 participants balanced a pole until a time criterion was reached. The geometry, mass, and characteristic "fall time" of the pole were manipulated. Distributions of timing between pole and hand velocities showed strong action-perception coupling. When actions demonstrated a potential for failure, the period of hand oscillation correlated significantly with the "time to balance" (t bal=theta/theta.), where q is pole angle re: the vertical balance point, but not other quantities such as theta and theta. alone. This suggested that participants were attending to available t bal information during critical situations. In a model analysis and simulation, we demonstrated how discrete t bal information may be used to adjust the parameters of a controller to perform this task. In Experiment 2 participants balanced a virtual inverted pendulum under manipulations designed: (1) to decouple the mechanics of the system from its visual image; (2) to alter the mapping of perception and action; and (3) to perturb successful balancing. A replication of the correlation analysis of Experiment 1 revealed that across all conditions, significant relationships existed between visually specified t -variables and hand oscillation during critical motions of the pole. These results suggested that participants use the same t bal information to successfully stabilize both virtual and physical unstable systems, despite quite dramatic visual and mechanical transformations. In Experiments 3 and 4 we investigated how parts of the body, or individuals in a social dyad cooperate to perform a functional stabilization task. Participants balanced a pole either intermanually (using 2 separate hands) or interpersonally (2 persons each using their preferred right hand) until a time criterion was reached. Although the magnitudes of the forces exerted by each hand were different, an analysis of the timing of the forces revealed that intermanual (interpersonal) participants developed a consistent antiphase (inphase) coordination pattern. These different coordination patterns allowed for the recruitment of previously unavailable efferent and afferent connections to produce the net forces that served to stabilize the pole via theta. (see Experiment 1).

The dynamics of human sensorimotor coordination are studied at behavioral and neural levels through temporal synchronization and syncopation tasks. In experiment 1, subjects synchronized their finger movements (in-phase) with a metronome at 2.0Hz and 1.25Hz for 1200 cycles. Fluctuations of timing errors were analyzed through correlation, power spectrum analyses and Maximum Likelihood Estimation (MLE). Results indicated that the synchronization error time series was characterized by a 1/falpha type of long memory process with alpha = 0.5. Previous timing models based upon motor program or simple "central clock" ideas were reviewed to show that they could not explain such long range correlations in the synchronization task. To explore the possible cognitive origins of long range correlation, experiment 2 required subjects to synchronize (on the beat) or syncopate (off the beat) to a metronome at 1Hz using different cognitive strategies. Timing fluctuations were again found to be 1/f alpha type, with alpha = 0.5 in synchronization and alpha = 0.8 in syncopation. When subjects employed a synchronization strategy to successfully syncopate, timing fluctuations shifted toward 1/f 0.5 type. This experiment indicated that the scaling exponent in timing fluctuations was related to task requirements and specific coordination strategies. Further, they suggest that the sources of such long memory originated from higher level cognitive processing in the human brain. Experiment 3 analyzed magnetoencephalography (MEG) data associated with synchronization and syncopation tasks. Brain oscillations at alpha (8--14Hz), beta (15--20Hz) and gamma (35--40Hz) frequency ranges were shown to correlate with different aspects of the coordination behavior. Specifically, through power and coherence analyses, alpha activity was linked to sensorimotor integration and "binding", beta activity was related to task requirements (synchronization or syncopation), and gamma activity was related to movement kinematics (trajectory). These results supported the idea that the 1/f alpha type of timing fluctuations originated from collective neural activities in the brain acting on multiple time scales.

The dynamics of recruitment and suppression processes are studied in the coupled pendulum paradigm developed by Kugler and Turvey (1987). Experimentally, the main concern is whether pendulum motion in this task is purely planar. Theoretically, the main concern is whether one-dimensional phase equations developed originally by Haken, Kelso and Bunz (1985) and the symmetry breaking extension by Kelso, Delcolle and Schoner (1990), can capture the richness of the dynamics of this experimental model system. In experiment 1, subjects swung single hand-held pendulums in time with an auditory metronome whose frequency increased. Bifurcations from planar to spherical pendulum motion occurred at critical cycling frequencies. Typically, these frequencies were above the pendulum's eigenfrequency. Spectral measures showed that spherical pendulum motion was generated through the recruitment of wrist abduction and adduction. The spectral measures revealed that elbow flexion and extension was recruited as movement rate increased, presumably to stabilize pendulum motion. When recruited, both components frequency- and phase-entrained with the primary pendulum mover, wrist ulnar flexion-extension. In experiment 2, subjects swung coupled pendulums in either an in-phase or anti-phase coordinative mode as movement rate increased. Transitions between coordinative modes were not observed. Pattern stability, as defined by the variability of the phase relation between the pendulums, was not affected to any large degree by increasing movement rate. Bifurcations from planar to spherical motion emerged at critical cycling frequencies. Spectral measures demonstrated that this motion was generated by abduction and adduction of the wrist. Elbow flexion-extension motion was also recruited. The newly active components frequency- and phase-entrained with wrist ulnar flexion-extension. When the same neuromuscular components were recruited simultaneously, e.g., elbow motion in both arms, the components exhibited frequency- and phase-entrainment with the task defined pattern. The results demonstrate that recruitment processes stabilize the coordinative modes, thereby reducing the need to switch patterns. Both experiments revealed a much richer dynamics than ever observed in the coupled pendulum paradigm and question the application of one-dimensional phase equation models to the coupled pendulum paradigm.

Perception and behavior are mediated by a widely distributed network of brain areas. Our main concern is, how do the components of the network interact in order to give us a variety of complex coordinated behavior? We first define the nodes of the network, termed functional units, as a strongly coupled ensemble of non-identical neurons and demonstrate that the dynamics of such an ensemble may be approximated by a low dimensional set of equations. The dynamics is studied in two different contexts, sensorimotor coordination and multisensory integration. First, we treat movement coupled to the environment as a driven functional unit. Our central hypothesis is that this coupling must be minimally parametric. We demonstrate the experimental validity of this hypothesis and propose a theoretical model that explains the results of our experiment. A second example of the dynamics of functional units is evident in the domain of multisensory integration. We employ a novel rhythmic multisensory paradigm designed to capture the temporal features of multisensory integration parametrically. The relevant parameters of our experiment are the inter-onset interval between pairs of rhythmically presented stimuli and the frequency of presentation. We partition the two dimensional parameter space using subjects perception of the stimulus sequence. The general features of the partitioning are modality independent suggesting that these features depend on the coupling between the unisensory subsystems. We develop a model with coupled functional units and suggest a candidate coupling scheme. In subsequent chapters we probe the neural correlates of multisensory integration using fMRI and EEG. The results of our fMRI experiment demonstrate that multisensory integration is mediated by a network consisting of primary sensory areas, inferior parietal lobule, prefrontal areas and the posterior midbrain. Different percepts lead to the recruitment of different areas and their disengagement for other percepts. In analyzing the EEG data, we first develop a mathematical framework that allows us to differentiate between sources activated for both unisensory and multisensory stimulation from those sources activated only for multisensory stimulation. Using this methodology we show that the influences of multisensory processing may be seen at an early (40--60 ms) stage of sensory processing.