Collective intelligence notes on Turvey's "Coordination"
Turvey’s “Coordination” paper (pdf) is very useful for understanding motor behavior as collective intelligence. I reread it recently; here are some notes and thoughts.
Early in his career Bernstein [wikipedia] had argued that the kinematic and dynamic aspects of movements, and the functional expediencies of these aspects, were grossly undervalued in accounts of coordination. For him, the problem in studying movement was one of understanding the control of a complex kinematic system. Recognizing that inertia, reactive forces, and initial postural conditions combine with active muscle forces in producing movements, Bernstein ruled out any straightforward, unambiguous relation between the nervous impulses innervating movements and the movements themselves. In so doing, he underscored the essential formative and steering roles of the information available to perceptual systems. More specifically, he saw the basic problem of coordination as that of mastering the many degrees of freedom involved in a particular movement--of reducing the number of independent variables to be controlled.
Reducing the number of independent variables to be controlled is something the price system does. If you want to get a particular outcome, you can just move prices around without having to micromanage the billions of economic agents and all the complex activity going on in the economy. Bernstein’s idea that there is no straightforward relationship between particular nervous behavior and particular movements foreshadows analogous discoveries in both psychology and morphogenesis: a particular firing of neurons in the brain has no unambiguous psychological implications; a paritcular genetic code has no unambiguous morphogenetic implications.
Reflexes, for example, were not elements of coordinated actions for Bernstein but, rather, elementary coordinated actions and, therefore, part of the problem of coordination rather than contributors to its solution.
An updated version of the same idea has recently been independently advanced in neuroscience, corroborating Berstein’s argument. Reflexes are constructed just like everything else.
Each and every movement comprises a state space of many dimensions; the problem of coordination, therefore, is that of compressing such high-dimensional state spaces into state spaces of very few dimensions.
The problem is analogous in morphogenesis: there are 36 trillion cells, too many degrees of freedom to deal with. Bioelectricity solves the problem by reducing the control problem to a relatively low-dimensional space: the space of the parameters of the shared model.
The kinematic (spatiotemporal) details of any coordinated state are not determined at the outset, in a single step by a single subsystem. The details are contributed gradually, by many subsystems working together.
This is obviously analogous to economics and morphogenesis. Economic activity isn’t planned at the outset, and morphogenesis isn’t genetically encoded; instead, the relevant agents figure it out as they go.
It is possible to configure (a) different degrees of freedom (e.g., body segments and joints) in the same way to achieve the same purpose and (b) the same degrees of freedom in different ways to achieve different purposes. These abilities, in conjunction with distributed construction, imply that a "plan" for a coordinated act is probably defined functionally in an abstract manner that refers neither to body segments nor to actual motions. The referents appear to be relations among properties that are relatively few in number, realizable in all body segments, and capable of generating many different motions.
This is relational realism/polycomputing/multicausality. Plans are collections of prediction signals that are only relationally meaningful to each other, and which together only constitute a plan in the ordinary sense with respect to a collection of agents and an environment.
The subsystems that compose a biological movement system are relatively autonomous; each relates to the "external medium " of surrounding subsystems according to a locally defined and simple expedient, for example, minimal interaction. With each subsystem satisfying its own local and particular expedient, multiple subsystems cooperate to generate "desired" states of affairs at the system level without "knowing" that they are doing so. There may well be a method of cooperation that is independent of the particular bodily coordination to be achieved. That is, cooperation among autonomous subsystems may be solved once and for all by very general principles common to all biological movement systems.
This is the most important passage for a collective intelligence connection: the subsystems of motor behavior are autonomous agents optimizing within an environment that consists in large part of other agents. Each subsystem is self-interested, but because of the way each connects to the others, they achieve group-level outcomes together. And this coordination method isn’t just limited to biological movement but also applies to morphogenesis and all economic systems.
Subsystems of a biological movement system have standard behaviors or generate standard functions. When a subsystem is active a "ballpark" of states, a family of functions, is established. In order to produce a variant of the coordination suited to current circumstances, the activated subsystems must be adjusted or tuned. The tuning is done by other relatively autonomous subsystems, for example, feedback from receptors signaling current limb positions.
More on the autonomy of subsystems: they do what they’re going to do, and controlling them is about manipulating their connections so that when they do what they want, they end up doing what you want.
In a biological movement system with its distributed control, any subsystem (or set of subsystems) in the executive role will not know the actual outputs of the subordinate sub-systems from which it composes an act. This follows, in part, from the ballpark/tuning distinction. To achieve a given end, the executive might always activate a given subsystem, even though the subsystem's actual behavior cannot be anticipated by the executive because of the subsystem's tuning, which varies from one occasion to the next. Moreover, the subsystem activated by the executive might not be the actual subsystem that does the job. Different subsystems can satisfy a piece of an executively specified function and will be interconverted by low-level automatic processes according to a local expediency of which subsystem gets the job done most simply. Executive knowledge is, therefore, only approximate, referring to large classes of tunings, subsystems, and low-level processes in general, that seem equivalent because they can be used to achieve the same purpose. On the positive side, these equivalence classes may identify systematically behaving units of very few degrees of freedom in terms of which intentions (goals, plans) can be framed for coordinating the behaviors of very many degrees of freedom.
Very Hayekian feel to this; on the biological side, any imagined genetic executive would run into the same issue.
A group of muscles spanning several different joints, and capable of contracting independently of each other, can become functionally linked so as to behave as a single task-specific unit. Such linkages may comprise the most primitive independently governable actuators of movement. They are characterized by a pronounced standardization that lessens the degrees of freedom requiring executive control. During movements that vary in speed and power, the internal mechanical degrees of freedom of a muscle linkage relate among themselves and with respect to the time frame in a fixed fashion. Muscle linkages are also characterized by task-specific flexibility. If perturbed temporarily, all of the internal mechanical degrees of freedom and not just those at the site of the perturbation--readjust immediately in such a way that the task goal is preserved.
In retrospect, this sure reads like groping toward the concept of a cognitive glue.
There’s much more of interest in this paper. But I think this so far shows that the science of motor behavior was at one point not far from the collective intelligence view in morphogenesis.