“One can use neurophysiological data…to…stand in opposition to many of the assumptions of cognitivism.”

Standard

…one of the most fundamental assumptions of cognitive psychology {is} the distinction between disembodied computational mechanisms of pure cognition and the physical implementation of behavior by the brain.  This distinction is very dear to modern psychology because it was instrumental in making a study of human thought and the “cognitive revolution” justifiable at a time when not much was known about brain mechanisms. Much has changed in recent decades.

Owing to the accelerating growth of neuroscience, there is now a movement to bring psychology and biology back together, a movement often called “cognitive neuroscience”.  But cognitive neuroscience is not merely a broad endeavor to understand the functional architecture of the brain.  In practice, it is an attempt to map a very particular conceptual toolbox, that of cognitivism, onto brain data.   Alternative viewpoints, such as “embodied cognition” or the “dynamical approach” a framework is based upon a distinction between two kinds of pragmatic concerns that animals face while actively interacting with a physical world: action specification and action selection.

{Cisek’s alternative model includes}

The natural world continuously presents us with many opportunities for action, and sensory information arriving from that world significantly constrains the parameters of these potential actions

  • The arrangement of surfaces and objects around one’s body constrains the possible directions of locomotion.
  • The egocentric location, orientation, and size of a graspable object constrain the possible limb configurations, hand orientations, and finger apertures required to grasp it.

In the framework, the brain automatically uses spatial information arriving from the world to begin to specify the parameters of currently available potential actions.  Of course, one cannot perform all possible actions at the same time.  Neither is it likely that the brain attempts to begin planning all interactions possible with the environment at any given moment.  Thus, there must exist mechanisms to reduce the number of potential actions and to ultimately select one for overt execution.  These decisions can also be based upon sensory information, such as the identity of objects in the world.

In this framework, behavior is seen as a constant battle between currently available opportunities for action.  In other words, the brain continuously transforms sensory information into the parameters of potential actions, while accumulating information useful for selecting one of these into overt execution.  There is no border between planning and execution systems, and even the final selected action is not completely specified as a “desired trajectory” typical of many robotics type controllers.  Instead, as suggested by both neurophysiological and psychophysical data, only the initial part of a movement is prepared and the movement then develops on-line via both external feedback through the environment and internal feedback through “forward models”.  Below, I briefly speculate on how this framework can be used to interpret neural data from parietal and frontal cortex.

In the specification-selection model:

  • spatial information is gradually transformed from a sensory format describing the world (e.g., object position)
  • into a motor format specifying potential actions (e.g., direction from hand to object) along the posterior parietal cortex. This is consistent with the characterization of the parietal cortex as part of a dorsal visual system involved in visually-guided movement.
  • As the sensorimotor transformation occurs, information for many potential actions is progressively eliminated from further processing through a series of selection mechanisms.
  • When the information is still in a sensory format, selection occurs based on sensory features such as salience or spatial location – we may call such selection a mechanism of “attention”.
  • In the early part of the visual system, attentional influences enhance information from particular regions of space while other regions are suppressed.
  • Surviving information is transformed further, and the dorsal stream diverges into separate systems concerned with different classes of actions, such as reaching in MIP, grasping in AIP, biting in VIP, and looking in LIP, each with its own idiosyncratic representation of space.
  • These representations are quite impoverished, however, with only the most salient features of the environment being represented.
  • Because the same regions are clearly implicated in early movement planning, we can say that only the most promising potential actions make it this far along the dorsal stream.
  • At this point, the expected consequences of potential actions can be used to influence further selection – we can call such selection a mechanism of “decision-making.”
  • Indeed, decision variables such as movement probability and expected payoff influence activity of parietal cells, as does behavioral context.
  • Surviving potential actions are carried to frontal regions such as the dorsal premotor cortex, where final action decisions are reflected before overt movements are generated.

A number of brain regions may provide the influences needed for the selection mechanisms described above.  Visually-guided action selection may utilize information from the ventral stream,
where cells are sensitive to stimulus features.  Such features could he used to bias selection along different areas of the parietal action specification system by enhancing promising potential actions while suppressing others, or simply by influencing the dorsal stream to fixate on behaviorally-relevant information.

Because action selection is a fundamental concern faced even by our distant ancestors, we should expect that it involves phylogenetically old structures such as the basal ganglia.  A behavioral competition in the basal ganglia may bias selection by influencing specific cortico-basal-thalamo-cortical loops.

At the same time, because action selection is also likely to have become significantly more sophisticated in the recent evolutionary history of primates, it probably also involves neural structures that are particularly developed in the great apes, such as the frontal lobes of the cerebral cortex.  Complex criteria for sophisticated selection may be processed in the prefrontal cortex, where cells are sensitive to those stimulus features relevant to response selection and where information for making decisions is accumulated.  Our exceptionally large frontal lobes may have enabled the human ability to select actions based upon increasingly complex criteria, and classical frontal syndromes, which affect a patient’s ability to select actions appropriately, illustrate what can go wrong when that ability is lost.

The theory {is a} “motor chauvinistic” perspective on neurophysiological data.  The locus of activity in their movement planning field specifies potential actions in a space of action parameters (in their case, a space of potential directions), and the amplitude at each locus reflects the influences of selection.  Thus, competing potential actions can coexist as distinct hills {spiking peaks} in the landscape of cellular activity.

  • A highly desirable action which demands great precision may be coded as a tall narrow hill, while a less desirable and more unconstrained action may be a low and wide plateau {of neuronal activity and spiking}.
  • All kinds of selection influences, from “attention” to “decision variables,” may be combined together to bias the competition between hills of activity which correspond to different potential actions.

Such a planning field model can be used to simulate the results of many of the studies…when a monkey was faced with two potential reaching actions, one of which would eventually be performed after a delay of several seconds, neural activity from the dorsal premotor cortex indicated that both movements were prepared simultaneously before one would be selected for overt execution.

That is, instead of making a cognitive decision first and then preparing action (as would be predicted by the “sense-think-act” architecture of traditional cognitivism), the monkey first specified multiple potential actions and then selected among them (as predicted by the “specification-selection” architecture).

One can predict that competition between actions in a movement planning field will be evident even in the final movement trajectory, with subtle deviations occurring when the activity bill of an unselected potential movement slightly overlaps that of the selected movement…most neural activity is not so much concerned with representing the world as with “mediating interactions with the world,” through specifying potential actions and selecting among them.  One can use neurophysiological data, traditionally interpreted from the perspective of cognitivism, to support theoretical frameworks such…which stand in opposition to many of the assumptions of cognitivism.  In fact, if we are indeed poised to witness a shift away from the disembodied computational assumptions of traditional cognitive psychology to a more embodied science of behavior, such a shift may be primarily driven by the growing literature of neurophysiological data.

http://www.cisek.org/pavel/Pubs/Cisek-Thelen.pdf

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s