“Neural coding” is a popular metaphor in neuroscience, where objective properties of the world are communicated to the brain in the form of spikes. Here I argue that this metaphor is often inappropriate and misleading. First, when neurons are said to encode experimental parameters, the neural code depends on experimental details that are not carried by the coding variable (e.g., the spike count). Thus, the representational power of neural codes is much more limited than generally implied. Second, neural codes carry information only by reference to things with known meaning. In contrast, perceptual systems must build information from relations between sensory signals and actions, forming an internal model. Neural codes are inadequate for this purpose because they are unstructured and therefore unable to represent relations. Third, coding variables are observables tied to the temporality of experiments, whereas spikes are timed actions that mediate coupling in a distributed dynamical system. The coding metaphor tries to fit the dynamic, circular, and distributed causal structure of the brain into a linear chain of transformations between observables, but the two causal structures are incongruent. I conclude that the neural coding metaphor cannot provide a valid basis for theories of brain function, because it is incompatible with both the causal structure of the brain and the representational requirements of cognition.

To investigate neural mechanisms of movement, physiologists have analyzed the activity of task-related neurons in behaving animals. The relative onset latencies of neural activity have been scrutinized for evidence of a functional hierarchy of sequentially recruited centers, but experiments reveal that activity changes occur largely in parallel. Neurons whose activity covaries with movement parameters have been sought for evidence of explicit coding of parameters such as active force, limb displacement, and behavioral set. Neurons with recognizable relations to the task are typically selected from a larger population, ignoring those cells with complex relations to the task and unmodulated cells. Selective interpretations are also used to support the notion that different motor regions perform different motor functions; again, current evidence suggests that units with similar properties are distributed over widely different regions.

These coding issues are reexamined for premotoneuronal (PreM) cells, whose correlational links with motoneurons are revealed by spike-triggered averages. PreM cells are recruited over long times relative to their target muscles; they show diverse response patterns relative to the muscle force they produce; functionally disparate PreM cells such as afferent fibers and descending corticomotoneuronal and rubromotoneuronal cells can exhibit similar patterns. Neural mechanisms have been further elucidated by neural network simulations of sensorimotor behavior; the pre-output hidden units typically show diverse response patterns in relation to their target units. Thus, studies in which both the activity and the connectivity of the same units are known reveal that units with both simple and complex relations to the task contribute significantly to the output. This suggests that the search for explicit coding may be diverting us from understanding distributed neural mechanisms that operate without literal representations.

During normal vision, when subjects attempt to fix their gaze on a small stimulus feature, small fixational eye movements persist. We have recorded the impulse activity of single neurons in primary visual cortex (V1) of macaque monkeys while their fixational eye movements moved the receptive-field activating region (AR) over and around a stationary stimulus. Three types of eye movement activation were found. (1) Saccade cells discharged when a fixational saccade moved the AR onto the stimulus, off the stimulus, or across the stimulus. (2) Position/drift cells discharged during the intersaccadic (drift) intervals and were not activated by saccades that swept the AR across the stimulus without remaining on it. To activate these neurons, it was essential that the AR be placed on the stimulus and many of these cells were selective for the sign of contrast. They had smaller ARs than the other cell types. (3) Mixed cells fired bursts of activity immediately following saccades and continued to fire at a lower rate during intersaccadic intervals. The tendency of each neuron to fire transient bursts or sustained trains of impulses following saccades was strongly correlated with the transiency of its response to stationary flashed stimuli. For one monkey, an extraretinal influence accompanying fixational saccades was identified. During natural viewing, the different eye movement classes probably make different contributions to visual processing. Position/drift neurons are well suited for coding spatial details of the visual scene because of their small AR size and their selectivity for sign of contrast and retinal position. However, saccade neurons transmit information that is ambiguous with respect to the spatial details of the retinal image because they are activated whether the AR lands on a stimulus contour, or the AR leaves or crosses the contour and lands in another location. Saccade neurons may be involved in constructing a stable world in spite of incessant retinal image motion, as well as in suppressing potentially confusing input associated with saccades.

This study is concerned with how information about the direction of visual motion is encoded by motion-sensitive neurons. Motion-sensitive neurons are usually studied using stimuli unchanging in speed and direction over several seconds. Recently, it has been suggested that neuronal responses to more naturalistic stimuli cannot be understood on the basis of experiments with constant-motion stimuli (de Ruyter van Steveninck et al., 1997). We measured the variability and information content of spike trains recorded from directional neurons in the nucleus of the optic tract (NOT) of the wallaby, Macropus eugenii, in response to constant and time-varying motion. While the NOT forms part of the mammalian optokinetic system, we have shown previously that the responses of its directional neurons resemble those of insect H1 in many respects (Ibbotson et al., 1994). We find that directional neurons in the wallaby NOT respond with lower variability and higher rates of information transmission to time-varying stimuli than to constant motion. The difference in response variability is predicted by an inhomogeneous Poisson model of neuronal spiking incorporating an absolute refractory period of 2 ms during which no subsequent spike can be fired. Refractoriness imposes structure on the spike train, reducing variability (de Ruyter van Steveninck & Bialek, 1988; Berry & Meister, 1998). A given refractory period has a greater impact when firing rates are high, as for the responses of NOT neurons to time-varying stimuli. It is in just these cases that variability in experimentally observed spike trains is lowest. Thus, differences in response variability do not necessarily imply that different models are required to predict neuronal responses to constant- and time-varying motion stimuli.

It is worthwhile to search for forms of coding, processing, and learning common to various cortical regions and cognitive functions. Local cortical processors may coordinate their activity by maximizing the transmission of information coherently related to the context in which it occurs, thus forming synchronized population codes. This coordination involves contextual field (CF) connections that link processors within and between cortical regions. The effects of CF connections are distinguished from those mediating receptive field (RF) input; it is shown how CFs can guide both learning and processing without becoming confused with the transmission of RF information. Simulations explore the capabilities of networks built from local processors with both RF and CF connections. Physiological evidence for synchronization, CFs, and plasticity of the RF and CF connections is described. Coordination via CFs is related to perceptual grouping, the effects of context on contrast sensitivity, amblyopia, implicit influences of color in achromotopsia, object and word perception, and the discovery of distal environmental variables and their interactions through self-organization. Cortical computation could thus involve the flexible evaluation of relations between input signals by locally specialized but adaptive processors whose activity is dynamically associated and coordinated within and between regions through specialized contextual connections.