Classical Neural Theory (Neuralism) and its Demise

S. David Stoney, Ph. D., Dept. of Physiology, Medical College of Georgia, Augusta, GA 30912 (Emeritus),
Iced Neuron (

I. Introduction. Classical neural theory (CNT) is the conceptual counterpart to classical physical theory. It is a carry-over of concepts of mind and matter (dichotomous thinking) from the 19th century, one that ignores the actuality of a quantum universe. CNT (a.k.a. neuralism) accepts that, with one exception, matter is in all respects insentient. The exception is for special arrangements of special, electrochemically active matter in certain neurons in restricted parts of the nervous system, which are considered to be necessary and sufficient for conscious awareness. Although neuralism appears to be a kind of monism, it is, in fact, a surreptitious form of dualism: the objective, external world is represented internally by patterns of neural activity that become subjectively known. All the properties of the Cartesian mind are now misattributed to the stuff of the brain (Bennett & Hacker, 2001).

Fig. 1. Theories of Mind. Classical neural theory misascribes properties of the Cartesian mind – res cogitans – to the matter of the brain.

Here, I point out certain difficulties that the incoherent neuralist scheme has in reconciling itself with the actual response properties of the nervous system. This critique assumes that the purported formation of neural representations of the external world is via spike train (spike rate, interspike interval) encoding.

II. Classical Neural Theory’s Wounds

A. CNT needs an evolutionary miracle. Miraculously, conscious awareness "winks in" (Chalmers, 1966) for brains. The idea that sentience/awareness can develop from insentient (classical) matter is, in fact, a subtle form of Creationism.

B. Neuronal receptive fields (RFs) are broadly tuned and not very informative about the nature of objects. The neuronal RF is that part of the joint body and stimulus phase space in which a stimulus is effective for changing a neuron’s rate of discharge. Fig. 2 shows a tactile RF for a "direction-sensitive" neuron in monkey SI cortex. A mechanical (touch) stimulus drawn across the pink area activates the neuron. In CNT, the neuron is considered to "represent" the direction of movement that produces the largest response. However, there is nothing in the neuron’s response that allows it to distinguish between a lot of a 270° stimulus and a little of a 290° stimulus. Broad tuning allows for precise signaling of the fact of a stimulus event, but imprecise signaling of almost any thing else and forces CNT to account for precise representations in terms of population responses.


Fig. 2. Response field of a directionally sensitive neuron in SI cortex of a monkey. Neuron responses (spikes) are shown in the upper traces. Lower traces show the duration of each stroke. Note the broad tuning. (Kandel et al, 2000)

C. Neuronal RFs signal prehensions, not representations. The response properties of CNS neurons are entirely compatible with their signaling prehension (see Swift & Stoney, 2002) rather than representation. The two examples below demonstrate this fact.

1. Responses of a neuron with a bimodal, so-called "space" representing RF in the putamen described by Graziano & Gross, 1993. Clearly, when this neuron fires action potentials it is signaling connection (prehension) between the monkey and an object within the neuron’s RF.

Fig. 3. Responses of a bimodal "space representing" neuron in the putamen of an anesthetized monkey. A. The neuron responds 1) when an object approaches or 2) when an object touches the shaded portion of the monkey’s face. Only objects passing through the indicated region towards the left side of the monkey’s face are effective stimuli.

2. Responses of mirror neurons. Mirror neurons of the premotor cortex respond both during a particular forelimb movement as well to the sight of another individual making a similar goal directed movement. From a classical perspective, such neurons are said to "represent the goal of an action" (Rizzolati et al, 1999) or to be an instance of a more general matching system that represents "goals, emotions, body states and the like to map the same states in other individuals" (Gallese and Goldman, 1998). Well, what can certainly be said is that watching another monkey perform a goal directed action causes activation of neurons that are activated during similar goal directed actions. From a process philosophical perspective they can be said to not only signal a goal, but also the sharing (prehension) of value, i.e. sharing of a goal.

MirrorNeuron.jpg (42133 bytes)

Fig. 4. Action potentials (spikes) of a mirror neuron associated with observed and executed actions. There is neural activity when the monkey observes the experimenter grasping the piece of food and when the monkey grasps it. (From Rizzolatti & Arbib, 1998.)

D. Neurons are quite poor rate/interspike interval coders. Classically, because conducting action potentials are mostly of equal amplitude, it was assumed that neurons only encoded information in their firing rate (over some duration) or in the distribution of interspike intervals, the time between spikes. In fact, axon branch points, because of the increased electrical load they present to approaching impulses, act as a ganged, two position switches with high frequency cutoff features. At low frequency, the AP invades both branches (switch closed). However at frequencies just above the high frequency cutoff of the branch point, every other AP invades neither branch (switch periodically open) and the bandwidth of the neuron is degraded. This means that, in terms of output each neuron has a maximum effective frequency of oscillation that cannot be exceeded. Fig. 5 shows the set up for the experiments.

Fig. 5. Setup (A) and method (B) for determining impulse conduction at branch points of frog DRG neurons. B1 shows an orthodromic impulse failing to invade the myelinated segment at short interstimulus intervals (20 mV x 2 msec). Collision tests showed that when the action potential failed to invade the stem process it also failed to invade the dorsal root. B2 shows impulse failure in a nerve fiber recorded extracellularly just peripheral to the DRG (1 mV x 1 msec). (Stoney, 1990)

Branch points of slowly conducting neurons have the lowest high-frequency cutoffs, but the maximum effective firing frequency for conducted APs is significantly reduced at the branch points of even the most heavily myelinated axons (Fig. 6). (Stoney, 1990).


Results2.jpg (30684 bytes)

Fig. 6. (A) Impulse fails to invade myelinated segment and soma when frequency of orthodromic impulses is increased. (B) Comparison of maximum frequency of conducted impulses (±SE) for a sample of axons in the peripheral nerve and for axon branch points matched for conduction velocity.

Thus, with regard to conducted action potentials, axon branch points of myelinated and unmyelinated axons cause neurons to have lower maximum frequencies of firing than is typically estimated from measurement of absolute refractory periods of parent axons. It is therefore unclear that neurons will have the necessary bandwidth to represent things as required in CNT. Three alternatives have been proposed to rescue a classical brain: 1) that CNS neurons represent the world independently, i.e., that no encoded information from the periphery is necessary (Llinás & Paré, 1996); or 2) that a magical system is available in the brain to precisely measure the timing of each spike (Rieke et al, 1991; deCharms & Zador, 1998); or 3) that individual spikes somehow carry information (Rieke et al, 1991). Thus, CNT requires the brain to become more and more magical.

E. Population coding requires magical decoding: Since no neuron that specifically represents a whole object has ever been found in the brain, CNT, if it is correct, requires some sort of population coding and decoding. This sounds great except for the fact that there is, for neurons concerned with perception, no known mechanism for joining the separate information either within one or between more than one population response. In the case of neurons whose concern is with movement production, the population response is, in fact, actualized in the firing of motor units. Other than in the minds of ambitious experimenters, there is no known ‘final common path’ for neurons processing sensory information.

F. A representational system may be too slow to work: Generation of a population response requires a finite amount of time - extending into the 10’s or 100’s of msec. Such a response must reach a certain maturity prior to comparison with memory stores and other evaluative processes that must precede conscious recognition and/or utilization of the meaning of the population response pattern for control of behavior. The very short processing times for distinguishing stimuli (DeCharms, ARN) seems incompatible with such a system.

G. The concepts associated with CNT are not up to the task of understanding: Classical concepts like sensation, perception, memory, etc. are extremely fuzzy and actually do not fit well with the hard data of either physiological or cognitive neuroscience. See Vanderwolf (1994), Bennett & Hackett (1998), and O’Regan & Noë (2001).

H. CNT leads to a solipsism of the present moment: If all we really know are representations signaled by the activity of neurons in our brains, then the external world and everything we know about it is merely an inference. From this lonely perspective there is only a private "now," with no basis for temporality.

III. Conclusions

1. Classical neural theory (neuralism) appears to be mortally wounded.

2. Certainly, CNT is inadequate to the tasks of modern neuroscience and neuropsychology.

3. But how, if not via internal neural representations, do we each come to know the world?

4. See the Swift and Stoney poster for the answer.


M.R. Bennett and P.M.S. Hacker, Perception and memory in neuroscience: a conceptual analysis, Prog. Neurobiol. 65: 499-543, 2001.

D.J. Chalmers, The Conscious Mind: In Search of a Fundamental Theory, London: Oxford University Press, 1996.

R.C. deCharms and A. Zador, Neural representation and the cortical code. Ann. Rev. Neuroscience. 23: 613-47, 2000.

Vittorio Gallese and Alvin Goldman, Mirror neurons and the simulation theory of mind-reading, TICS 12: 493-501, 1998.

M.S.S. Graziano and C.G. Gross, A bimodal map of space: somatosensory receptive fields in the macaque putamen with corresponding visual receptive fields, Exp. Brain. Res. 97: 96-109, 1993.

E.R. Kandel, J.H. Schwartz, & T.M. Jessell (Editors), Principles of Neural Science, 4th Ed., New York: McGraw-Hill, 2000.

R. Llinás and D. Paré, The brain as a closed system modulated by the senses, In: The Brain-Mind Continuum: Sensory Processes, Cambridge, MA: The MIT Press, pgs. 1-18, 1996

J.K. O'Regan and A. Noë, A sensorimotor account of vision and visual consciousness. Beh. Brain Sci. 24(5), 2001.

F. Rieke, R. de Van Steveninck, R. de Ruyter, R. Van Steveninck, Spikes: Exploring the Neural Code, Cambridge, MA: The MIT Press, 1991.

G. Rizolatti et al, Resonance behaviors and mirror neurons, Arch. Ital. Biol. 137: 85-100, 1999.

G. Rizolatti and M.A. Arbib, Language within our grasp, TINS 21:188-94, 1998.

S. D. Stoney, Jr. Limitations on impulse conduction at the branch point of afferent axons in frog dorsal root ganglion, Exp. Brain Res. 80: 512-524, 1990.

A.V. Swift and S.D. Stoney, Neural bases of embodied human consciusness, Towards a Science of Consciousness, Tucson, AZ, April 8-12, 2002, Conference Res. Abstr., pg. 76.

C.H. Vanderwolf, The behavioral neurobiology of learning and memory: a conceptual reorientation, Brain Res. Revs. 19: 264-97, 1994.