There must be some physiological changes associated with learning and responsible for long term memories. Currently the popular place to look for such changes is in the central nervous system (brain and spinal cord). Since the central nervous system (CNS) appears to be responsible for perceiving and classifying stimuli and for sending out motor directives, it is a logical locus for learning, which involves such stimuli and responses. A few of the theoretical approaches will be discussed here under the following classifications: (a) neuronal-synaptic models, (b) RNA-protein models, (c) glial models, and (d) non-connectionistic theories.
The main building block of the CNS is the neuron.: the nerve cell and its processes (dendrites and axon). All information that passes through the CNS, relevant to incoming stimuli and outgoing responses, as well as all mediation within the CNS, is carried by neurons and passed from neuron to neuron. Within the neuron, information is transmitted electrically (as the result of flows of ions through the neuron’s surrounding membrane). Pribram (1971a, p. 15) distinguishes two types of electrical effects in the neuron: nerve impulse unit discharges and graded slow potential changes. The first type, the nerve impulse unit discharge, occurs when the excitation coming into a neuron exceeds some value, called the threshold. The neuron fires, and a wave of electrical activity (action potential) moves down the neuron toward the next neuron. The second type of electrical activity, graded slow potential changes, refers to the small, fluctuating electrical potentials within a neuron. These slow potential changes are more sensitive to non-neuronal influences, such as from the chemicals surrounding the neuron.
Neurons may influence each other if the neurons are simply close enough together so that the electrical-chemical activity of one interacts with and influences the electrical-chemical activity of the other. Such a connection between neurons, which does not involve any specific structures, is called an ephaptic junction. Perhaps more important for learning is the synapse, a specific structure for transmitting the activity of one neuron to the next. When a neuron fires, the action potential moves through the neuron to the synapse, where it activates the synaptic knob. This structure of the synapse contains a number of small sacs of chemicals called synaptic vesicles. When the neuron is fired these vesicles move within the synaptic knob to the space between the neuron and the next neuron; this space is known as the synaptic cleft. The chemicals, released from the vesicles into the synaptic cleft, travel across to the next neuron. Therefore these chemicals are called transmitter substances, for they transmit chemical changes from one neuron to another. Depending on the type of chemical it is and the nature of the synapse, the transmitter substance will produce either an excitatory or an inhibitory effect on the next neuron. This second neuron will then fire only when the sum of excitatory effects it receives from many different neurons minus the sum of the inhibitory effects exceeds the threshold for that neuron.
Figure 2—1 shows a schematic of an idealized neuron. All neurons have many dendrites which primarily receive information from other neurons via synapses, a cell body, and a single axon with many branches that carry the information to the next neurons. (For more detail on neurons see Eccies, 1957, 1964; Ochs, 1965.)
There are about 10 billion neurons in the human brain, each of which communicates with hundreds of other neurons. The number of synapses on any neuron may be in the thousands. Eccies (1965, p. 18) suggests that there could be as many as 10,000 synapses on a single neuron. This great complexity of interactions is one of the reasons that neuronal systems are popular theoretical bases for behavior and learning. A general problem is to explain how the nervous system can be flexible enough for continuous learning, but inflexible enough to store memories for the lifetime of the organism. Another problem is that there is probably no regeneration of neurons in the CNS of a mature organism; once a neuron dies, it is not replaced. Thus neuronal theories of learning must be careful about the types of growth postulated to occur in the
Most neuronal-synaptic theories of learning (e.g., Eccles, 1964, 1965; Hebb, 1949) assume that learning is based on physiological changes that permit some neurons to fire other neurons more easily. Currently most such theories assume the change to occur in the synapse between neurons. For example, it might be postulated that with learning, the relevant neurons grow closer together at the synapses so that it is easier for the transmitter substances to cross the synaptic cleft and affect the next neuron. Or a theory might be based on learning making the relevant synaptic transmission more effective, perhaps due to some chemical change in the synapse.
Thus according to a very oversimplified neuronal-synaptic model of learning, when a person learns to associate a particular song with the image of the recording artist singing it, learning involves something like the following: There is a set of neurons corresponding to hearing the song and another set corresponding to the image of the singer. Since these two sets of neurons are fired simultaneously (you see the singer as you hear the song), neuronal-synaptic changes occur between the neurons that connect the two sets of stimulated neurons. Now when one set of neurons is fired (you hear the song on the radio), there is a tendency for the other set to fire (you get a visual image of the singer). Such a theory is a connectionistic theory, since it is based on specific connections between specific neurons.
Eccles (1964, 1965) and other synaptic theorists have drawn heavily from a phenomenon called posttetanic potentiation, first observed by Lloyd (1949). This refers to an effect of very rapid stimulation (tetanic stimulation) on a simple nervous preparation. For example, a common preparation is a reflex pathway in the spinal cord of an animal. Here the firing of one set of neurons (the dorsal root fibers) has a tendency to fire some associated neurons (the ventral root fibers). If tetanic stimulation at a rate of about 300 stimulations per second is applied to the first set of neurons, it results in an enhanced ability (i.e., lowered threshold) of the first set of neurons to fire the second set. This enhanced ability lasts for several minutes after the tetanic stimulation is terminated, and thus the effect is called posttetanic potentiation (PTP). PTP is then suggested to be a model of how learning might occur by synaptic changes following continued stimulation.
At first Eccles suggested that the effects of tetanic stimulation were due to a swelling or growth of the neuron in the area of the synapse. However, as Grossman (1967, p. 837) argues, “the proposed swelling does not, per se, increase the size or number of the vesicles containing the humoral transmitter substance which presumably is responsible for the propagation of impulses across the synapse.” In light of this, Eccles by the mid- 1960’s was putting more emphasis on changes in the chemical transmission mechanism of the synapse.
There are a number of problems in using PTP as a model of learning. First is the fact that the changes are usually only seen after about 10,000 stimulations. But learning may require only one trial. Thus it is necessary to postulate some process (perhaps the reverberatory circuits discussed in Chapter 4) that is started by a single learning trial, but keeps going until the neurons are fired 10,000 times. It may be that there is no such process that naturally produces PTP in intact animals. A second problem is that the effect of PTP is only temporary, making it necessary to add other mechanisms to explain long term memories. A third problem is that the effects of PTP may not be specific enough to explain learning. That is, PTP appears to involve all the synapses of the fired neurons, which means an increased tendency to fire the hundreds or thousands of neurons that might synapse with the tetanic stimulated neuron. This effect then is probably too diffuse for learning. Learning more probably would involve only synaptic changes between a few select neurons, and thus only a few of the synapses of a neuron. To accomplish this more limited specificity some theorists (e.g., Hebb, 1949) assume that for learning to occur and produce the required synaptic changes both neurons to be associated must be fired together.
Although most synaptic theories talk about learning in terms of positive changes such as growth at the synapse, learning might also involve negative changes such as a shrinkage in contact area at the synapse. One paper (Rosenzweig, Mollgaard, Diamond, Bennett, 1972) suggests that the “structural synaptic changes underlying learning and memory storage need not be of a single form, but rather may take one or more of these four forms: (a) increase in number of synapses, (b) increase in size of contact areas, (c) decrease in number, and (d) decrease in size of contact.”
Deutsch (1968, 1971) has done a series of experiments designed to investigate chemical changes in the synapses that might underlie learning. He investigated those synapses in which the transmitter substance appeared to be acetyicholine (ACh). When released, ACh leaves the synaptic vesicles of one neuron, crosses the synaptic cleft to the next neuron, and has an excitatory effect on this second neuron. Then the ACh is broken down by the enzyme acetylcholinesterase (AChE). This prevents too much ACh from building up and interfering with future synaptic transmission. Deutsch used a variety of drugs, including drugs that interfere with ACh activity by blocking the receptor sites that ACh stimulates, drugs that produce similar effects as ACh but are resistant to being broken down by AChE, and drugs that inhibit the functioning of AChE. By manipulating the dosages of different drugs that enter the synapses and different learning variables, Deutsch mapped out the types of synaptic changes that correlate with learning. A problem with any such research that utilizes drugs is that the effect of the drug may not be due to the reason suspected but may merely be an artifact of some other effect of the drug. The drug may be producing some side effect that the researcher is not aware of and thus may produce misleading results.
From his research Deutsch concluded that learning involves changes in the efficiency with which synapses transmit messages. He suggests that this increased transmission efficiency is due to an increase in the sensitivity of the receptor neuron to respond to the transmitter substance. This sensitivity increases for some time after initial learning . and then declines. The rate of initial increase depends on the amount of learning. The decline in sensitivity may correspond to forgetting.
Besides neuronal-synaptic models, another approach to the physiology of learning and memory has centered around specific molecules that may store learned information. First let us consider an oversimplified model of how proteins are produced: In the nucleus of a cell is DNA (deoxyribonucleic acid), which contains the genetic information. DNA is a long chain in the form of a double helix, similar to a spiral staircase. A major constituent of the DNA is a base (a nitrogenous base) which comes in one of four types. DNA molecules differ in the particular sequence of these bases that occur in a long string along the DNA. Since the sequence of these bases probably determines which proteins will eventually be produced, including genetic characteristics, the sequence is called the ge netic code.
With the DNA as a pattern or template, a form of RNA (ribonucleic acid), called messenger RNA, is formed in the nucleus. Having received the genetic code about which proteins to produce, the messenger RNA moves into the cytoplasm of the cell. Here the messenger RNA synthesizes proteins from amino acids brought to it by a different type of RNA called transfer RNA. These proteins are complex molecules that are the basis for all living cells.
Although the above DNA-RNA-Protein sequence is fairly well established (although in a much more complex form), the relation to learning is quite speculative. Since behavior depends on what neurons fire and in what patterns, physiological theories of learning must include variables related to neuronal firing. There are basically two such relationships to be considered here. First, some of the proteins produced will affect the cell metabolism of the neurons and hence their firing pattern and/or they might affect some other aspect of neural transmission such as by directly affecting the transmitter substance. Second, there is evidence, discussed below, that neural firing affects the synthesis and type of RNA. These relationships are shown in Figure 2—2.
A theory of learning then might be as follows: A learning experience results in certain neurons firing. This neural activity affects RNA (or perhaps DNA), which in turn alters the type of protein synthesis, which then affects future firing of the neuron. Probably no one holds such a simple theory, but many theories have something similar to this at their core. DNA models, RNA models, and protein models will now be discussed separately as possible explanations of memory storage. Memories, however, may be more complex than any of these proposed storage systems. Memory storage may involve a number of different components, such as RNA plus proteins plus neuronal chemistry.
DNA is seldom considered as the site of memory storage, for two reasons. First, there is some question as to whether DNA ever changes or is influenced by neural activity, which appears to be a necessary prerequisite for memory storage. The second reason is that if DNA were to be changed it might hinder DNA’s prime function, the eventual production of very specific proteins. The cells might be able to afford to use some RNA and proteins for memory storage, but perhaps not the number that would result from changes in the DNA.
There have, however, been some DNA theories of learning. Gaito (1963) suggested that DNA might be the seat of memory. He dealt with the above problems by suggesting that the DNA of nerve cells might be different from the DNA usually studied by the biochemists (such as from the liver and pancreas). Gaito, however, later abandoned this theory. Griffith and Mahler (1969) have proposed what they call a DNA-ticketing theory of memory. They suggest that the sequence of nucleotides that compose DNA may be modified by enzyme activity which may produce changes in the bases away from the original four types. These base changes then alter the function of the DNA and the type of protein synthesized, which may then affect the firing of the neuron.
Most of the research and theories, however, have centered around RNA and related proteins (Booth, 1967; Gaito & Bonnett, 1971; Glassman, 1969; Gurowitz, 1969; Hornet al., 1973; Ungar, 1970). There is an abundance of RNA in brain cells, and this RNA is very responsive to neural activity. That is, if a particular area of the brain is activated, either artificially stimulated directly or because the animal is involved in a task utilizing this brain area, there is an increase in the amount of RNA synthesized in this area. As the stimulation becomes excessive there may be a decrease in the RNA. Pevzner (1966) and Gaito and Bonnett (1971) provide the following summary: With moderate stimulation (sound, electrical, vestibular) there is an increase in RNA in the appropriate brain area. With excessive stimulation (revolver shot, fatigue, electroshock) there is a decrease in the amount of RNA. Similar relationships should also hold for the related proteins (Gaito & Bonnett, 1971).
But a simple increase in the amount of RNA is probably not specific enough to justify RNA as a memory molecule. Rather it seems necessary to look for qualitatively different changes in RNA molecules as a function of the specific learning experiences. This is what Hyden attempted to do in a series of classic experiments.
In the first experiment (Hyden & Egyhazi, 1962) an experimental group of rats was trained to walk up a wire at a 45-degree angle to the floor. The behavior involved in such a task depends on the vestibular nucleus, a group of neurons concerned with balance. A second group of rats, the functional control group, was slowly rotated in a centrifuge to control for the vestibular stimulation the first group experienced. A third group, the no-treatment control group, was not given any special treatment. Following the treatments, an analysis on all rats was made of the RNA in the nuclei of cells (Deiter’s cells) in the vestibular nucleus. For the first two groups, the experimentals and the functional controls, but not for the no-treatment controls, the vestibular stimulation resulted in an increase in the amount of RNA. But there were differences in the relative amounts (ratios) of the different bases making up the RNA of the two groups. Climbing the wire resulted in the production of RNA with ratios of the bases making up the RNA different from those of the RNA resulting from the centrifuge spinning. This suggests that different learning experiences might be coded in terms of the bases of the
Associated with neurons are non-neural cells called glia or glial cells. A later experiment (Hyden & Egyhazi, 1963) showed that the types of RNA changes found in the neurons of the vestibular nucleus could also be found in the glia associated with these neurons.
Although very important and suggestive, there are a number of problems with these experiments. First, it is not clear whether the experimental rats can really be said to be learning to climb the wire, for they learned to do it very quickly, with some of the rats performing perfectly on the first trial. Second, the experimental groups and the functional control groups may not be equated on the amount of vestibular stimulation. Actively walking up a wire may produce a different amount of stimulation of the vestibular nucleus than is produced when the rats are passively being rotated. Third, a sensory nucleus, such as the vestibular nucleus, seems an improbable site for storing memories. Finally, the different effects observed disappear after 24 hours. Hence we need some other mechanism to account for long term memories.
Some of the above problems were avoided in another experiment (Hyden & Egyhazi, 1964). Rats were first trained to reach for food with one paw. Then they were forced to use the other paw to get the food (transfer of handedness). Studies of those parts of the cortex associated with the different paws showed changes in the bases of the RNA with the transfer of handedness.
Over the years Hyden has suggested a number of models relating his experiments to the molecular basis of learning and memory. He has described one of his more recent models as follows:
At learning, a sequence of events leads to a fixation of memory: information-rich modulated frequencies, field changes, transcription into messenger RNA in both neuron and glia, synthesis of proteins in the neuron, give a biochemical differentiation of the neuron-glia unit in millions, a readiness to respond on a common type of stimulus.
At retrieval, it is the simultaneous occurrence of the three variables: electrical patterns, the transfer of RNA from glia to neurons, and the presence of the unique proteins in the neuron, which decide whether the individual neuron will respond or not. (Hyden, 1970, p. 116.)
Gaito and Bonnett (1971) have questioned whether Hyden’s experiments show that learning results in any changes in the base ratios of RNA or whether it is merely an artifact of increasing the quantity of one type of RNA, with a particular base ratio, over other types of RNA. In other words, learning results in an increase in total RNA in the relevant brain areas, which increase may involve different RNA with different base ratios. Perhaps learning experiences cause a greater production of one type of RNA than of another. Since Hyden’s analyses generally involve pooling all the RNA from one area, the changes in base ratios of RNA may be due to the differential production of one type of RNA rather than to a change of the base ratios of RNA within any particular RNA molecule. Gaito and Bonnett conclude that there currently is “no conclusive evidence to indicate that qualitatively different RNA and/or protein species are synthesized during learning and other behaviors.”
If RNA is related to learning in some way, then perhaps injections of RNA could facilitate learning. Cameron and Solyom (1961) reported that they could improve learning and retention in senile humans by giv ing them yeast RNA. Later reports, although often somewhat confusing and contradictory, seemed to support this general finding. However, there are a number of problems with this research (cf. Sweet, 1969).
The main problem is that the effect may simply be due to a general stimulant effect. The RNA (perhaps through the intermediary of serum uric acid) might simply improve performance, not learning, by increasing the amount of overall stimulation. If this is true, then the effect could be produced by a whole range of stimulant drugs unrelated to RNA. This seems to be the case.
Even if the effect is somewhat specific to RNA, it may not be the RNA itself but some of its breakdown constituents (e.g., nucleotides) that produce the effect. Perhaps learning and memory optimally require a certain level of supply of nucleotides, and some senile humans go below this level. Thus by providing the extra needed nucleotides (or whatever constituents), learning and memories are improved. If this is the case then such injections might help only those people whose level of supply has fallen below the ideal level and would not help people with normal levels.
In 1955 Thompson and McConnell reported the conditioning of planaria (flatworms). Weak electric shock causes the planaria to contract their bodies. When the shock was paired with a light, the planaria eventually learned to contract to the light alone. (This is a form of respondent conditioning to be discussed in Chapter 5.) Later McConnell (1962), using the planaria, began the controversial memory transfer experiments. These experiments involve training one animal on a specific task and then seeing if the animal’s memory of this task can be biochemically transferred to some extent to another animal. McConnell began by conditioning some planaria to contract to light. These planaria, called donors, were then chopped into small pieces and fed to untrained cannibal planaria, the recipients. Control recipients were fed untrained donors. All recipients then were conditioned to contact to the light. The recipients that ate trained donors learned faster than recipients that ate untrained donors, making more correct responses from the first day of conditioning. This suggested to McConnell that perhaps some of the memory of the donor was biochemically transferred to the receiver. It appeared that RNA might be the molecule responsible for the transfer.
These experiments evoked considerable controversy (cf. McConnell & Shelby, 1970). Some people could not replicate the effects, and there were debates about experimental procedures. A popular argument is that it is not learning that is transferred, but only sensitization. For example, Hartry and associates (1964) reported that they could get the same cannibalism effect if the donor had any of a number of treatments such as exposure to light alone. Thus they argued that all that is transferred is a sensitivity to respond to certain stimuli. Therefore the reason the experimental recipients learn faster is that they have a sensitivity to simply respond more to the stimuli involved in the conditioning task.
In addition to arguments about Hartry’s procedures, there are problems with the general sensitization argument. First, learning has been defined in this book as including sensitization. Thus, by this definition, it makes no sense to say that no learning is transferred, only sensitization. Second, as will be seen in one example below, there are many transfer experiments that suggest that something more complex and specific than sensitization is being transferred.
In the mid-1960’s Danish and American investigators began reporting memory transfer studies with rats. In one of the first studies Fjerdingstad and associates (1965) trained rats in a two-alley runway to approach a light. After training, these donors were sacrificed, their brains removed, and the RNA extracted from their brains. This RNA was then injected into naive recipient rats. Control recipients either received no RNA or received RNA from naive rats. All recipients were then trained on the light approach task, with the experimental recipients learning faster than either set of controls.
Albert (1966) trained rats on an avoidance task where the rats on cue had to escape from one compartment to another to avoid shock. The rats were trained with the cortex of one hemisphere or half of the brain made temporarily non-functional through a procedure called spreading depression. Spreading depression occurs when the cortex is stimulated by certain forms of electrical, mechanical, or chemical stimulation which inhibit or depress the normal electrical activity of the cortex. This depression effect spreads from the point of stimulation through the cortex of that hemisphere. In Albert’s experiment the spreading depression restricted the cortical part of the memory to form only in the unaffected hemisphere. After training, Albert extracted part of the cortex of the rats. For the experimental group he extracted the area that appeared to be involved in the memory of the avoidance task. For the control rats he extracted a different part of the cortex. The RNA from each cortical area was extracted and injected back into the cavity around the stomach area of the same rat from which it was taken. The spreading depression was then switched from the untrained hemisphere to the trained hemisphere, and the rats were retrained on the avoidance task. Thus each rat had to relearn the avoidance with the half of the brain that was nonfunctional during the first learning, plus each rat had some of his own RNA injected back into him. The results of the experiment were that the experimental rats, who received RNA from a trained area of cortex, relearned much faster than the control rats.
Braud and Braud (1972) trained donor rats to choose the larger of two circles. The rats’ brains were then extracted, homogenized, and injected into recipient rats. The recipients were then given a choice between two circles, the larger of the two circles the donors had been trained on plus a still larger circle. The recipients, without training, had a significantly greater tendency to choose the larger of the two circles presented to them. Control rats showed no significant preference. This suggests that responding to a relationship between stimuli (always pick the larger) can be transferred biochemically. Sensitization to specific stimuli cannot account for this result. Perhaps there is sensitization to the larger-smaller relationship, but this is a more complex form of sensitization than most sensitization explanations usually suggest.
Many experimenters have not been able to replicate the memory transfer experiments and/or do not believe the reported results. Luttges and associates (1966), for example, reported unsuccessful attempts at demonstrating memory transfer. Also, when radioactively labeled RNA was injected into the stomach cavity, where RNA is often injected in transfer studies, there was no evidence that any significant amount got to the brain. Debates about why transfer studies are not replicated generally center around subtleties of training procedures and methods of extracting the active material.
The issue of the active material is fairly complex. Most of the RNA extracts used in transfer studies are impure, containing DNA and proteins, and perhaps peptides and other substances which might account for the transfer effect. Thus the effect may have nothing to do with RNA but rather may depend on some “impurity.” Perhaps the effect is based on some constituent of RNA after RNA is broken down. If one of these alternatives is true, it may not matter whether RNA molecules can reach the brain from the stomach cavity.
Protecting the brain is a functional barrier called the blood brain barrier, which keeps some substances from passing from the blood into the brain. It is often suggested that RNA molecules are too big to pass through this barrier, and that RNA injected into the body of an animal will not get to the brain. But it is not clear whether the blood brain barrier will always stop RNA (Albert, 1966).
A conceptual problem for memory transfer experiments is to explain how the RNA molecules (or whatever the active ingredient is), after reaching the brain of the recipient rat, are able to find and affect exactly those neurons that are related to the specific behavior or tendency being transferred. This is particularly difficult in light of the fact that although the brains of two different rats show a number of similarities of organization, they are also very different in many respects.
Best (1968) has proposed the following theory to account for the problems just discussed: Each genetically determined neuron is identified with a specific code which is part of its DNA. This identifier can be coded into other molecules, such as messenger RNA and Polypeptide molecules, which then affect the efficacy of synapses by interacting with similarly coded proteins. Thus each molecule can only facilitate the one neuron which it matches in code. Transfer effects, then, depend on the injected molecules affecting (perhaps indirectly) specific neurons with matching codes.
Recently many theorists have been giving more emphasis to proteins as possible memory molecules. An early protein model of memory was suggested by Katz and Haistead (1950). According to this model, learning results in proteins that are capable of reproducing themselves. Some of these replicas then become part of the membrane of the neuron (“membrane organization”), which facilitates the neuron’s ability to affect adjacent neurons. Memory traces were assumed to be composed of evolving protein lattices interrelating various neurons. In terms of things learned since 1950 we now see that Katz and Halstead made a false assumption about proteins serving as templates for the reproduction of other proteins. However, with changes the theory might still be viable.
The more recent experiments have primarily utilized drugs, such as puromycin, that are assumed to disrupt protein synthesis but which do not interfere with RNA synthesis. If such a drug given to an animal after a learning experience disrupts the memory of the experience, this suggests that the memory is stored in proteins rather than in RNA.
The first series of such experiments was begun by Flexner and his associates in 1962 (Flexner et al., 1962). Mice were trained to avoid shock in a Y-maze. After learning they were injected with puromycin, which appeared to disrupt the memories. Similar effects of protein blocking agents on learning have been reported for goldfish (Agranoff, 1967).
Later, however, Flexner and Flexner (1968) reported that the puromycin-disrupted memories recover over time, and that the recovery can be facilitated with injections of saline. This recovery suggests that the effects of puromycin are more on the expression of the memory than on the storage per se. Or, as Booth (1970, p. 25) suggested, the original effects of puromycin might have been on retrieval or on the subjects’ ability to be frightened, or both.
Unfortunately, the effects of drugs such as puromycin are not as clean as often assumed, but are often quite diverse. More information on the direct and indirect effects of such drugs is necessary for our understanding of research such as the above.
DNA, RNA, and proteins, with their complex sequences of bases and other constituents, appear to have enough complexity and storage space to store learned information. But there are some general problems:
RNA and most proteins are quite short lived, the average life span for a protein being about 8 hours. Therefore any theory that attempts to explain long term memory in terms of RNA and/or proteins requires some type of system that perpetuates those types of molecules or other physiological changes that are assumed to underlie long term memory.
It also may be that there are several different stages of learning, each with different properties and physiological substrates. RNA and proteins may be involved in only one stage of learning in a sequence of stages.
Associated with neurons are non-neural cells called glia. In man’s brain there are 5 to 10 times as many glial cells as neurons. Glia provide structural support, aid in the metabolism of the neurons, help maintain the water—salt balance, and aid the transmission from the blood in the capillaries to the neurons. By affecting the metabolism of the neuron, glia can affect the firing rate of the neuron, and thus might be involved in learning. However, neuronal theories are in vogue, and most theorists emphasize neurons instead of the more numerous glia. Some exceptions are noted in the following paragraphs.
As mentioned above, Hyden and Egyhazi (1963) found RNA changes in glia similar to those found in the related neurons. In Hyden’s theorizing he generally considers the neuron and its glia as a metabolic and functional unit involved in learning.
Galambos (1961) also saw the glia and neurons working together as functional units. He suggested that the glia might be “genetically charged to organize and program neuron activity” and thus provide the basis for memory. To a large extent, then, the neurons “merely execute the instructions glia give them.” According to this model, glia receive impulses, organize them somehow, and in some way give order to the neural activity. To Galambos the relationship of glia to the brain is like that of a program to a computer. Galambos also suggests that glia might be capable of some electrical activity, such as slow wave activity.
Other approaches to the study of the physiology of learning have emphasized the electrical properties of the nervous system. Electrodes on the skull of the subject, as with an electroencephalogram (EEG), can measure gross electrical activity of the brain. Such measures are good for identifying general states of the organism, such as how aroused it is. More localized information can be obtained by lowering fine electrodes into the brain itself. These electrodes can then measure the electrical activity of a small group of neurons or even of a single neuron. By recording under what situations neurons fire and the pattern of their firing we can attempt to map out the function of the neurons and the changes in them that correlate with learning. Electrodes in the brain can also be used to electrically (or chemically) stimulate and fire specific neurons in order to study their function.
Using such procedures, a variety of studies (John, 1961; Morrell, 1961) have investigated phenomena such as the following: the conditioning of specific brain waves as measured by the EEG, changes in the electrical activity of specific brain areas with different stages of learning, the effects on learning of the electrical stimulation of specific brain areas, and the effects on learning of electrical currents of different polarity.
E. Roy John and his co-workers (John, 1961, 1972) ran a series of learning experiments in which they used lights that flickered at specific frequencies. They chose frequencies within the range to which the brain can respond. That is, when a flickering light is presented to an organism, groups of neurons in the brain, particularly in the visual system, may tend to fire in a pattern that corresponds to the flicker frequency. By investigating the neurons that fire to a specific pattern, an attempt is made to discover what information is being carried by these neurons.
For example, in one experiment (John et al., 1969) cats were trained on a task in which they discriminated between two frequencies of light flicker. For one frequency (V1) the cats pressed the right lever for food. For the other frequency (V2) the cats pressed the left lever to avoid being shocked. After the cats were well trained, they were exposed to a novel stimulus (V3) with a frequency midway between V1 and V2. Electrical activity (averaged evoked potentials) was recorded from various brain areas such as the visual system (visual cortex, lateral geniculate). These recordings showed that specific and different wave shapes were elicited by V1 and V2. When V3 was first introduced, the cat generalized and made the response appropriate to V1 or V2. The wave pattern recorded during this presentation of V3 was thus similar to that recorded during V1 or V2, depending on which response was made. If the cat, when presented with V3, pressed the right lever (the response learned to V1), the electrical activity to V3 would be similar to that of V2. If the cat pressed the left lever, the V3 activity would correspond to the activity caused by V2.
It thus appears that the pattern activity of the visual system to the novel stimulus V3 depends in part on whether the cat “interprets” the stimulus as being an example of V1 or V2. The experimenters concluded that “the shape of the evoked potential released by a novel stimulus during generalization is not solely determined by the actual physical stimulus but contains an endogenous component which varies depending upon the meaning attributed by the animal to the signal.”
On the basis of experiments like the one above, John (John 1967, 1912) has proposed a statistical configuration theory of learning. Following a learning experience the relevant neurons are involved in a coherent pattern of activity. Neurons always have some baseline activity, but the learning causes the relevant neurons to now fire in a coherent pattern. This activity is assumed to result in a common change in cellular chemistry, which increases the probability that the next time the neurons fire they will display the coherent pattern. Memory, then, is stored as the probability of coherence of neural firing. A memory is recorded in terms of electrical patterns from many neurons, like the patterns that flicker- frequency V3 first elicited in the generalization study. According to this theory, “remembering” and “thinking” are subjective experiences corresponding to the release of the electrical wave-shape representing a specific memory. An important point is that the memory requires a particular pattern of activity, but does not necessarily require the participation of any specific neurons. All that is needed is enough of the neurons to produce sufficient patterned activity for memory.
John’s theory is an example of a non-connectionistic theory since it does not depend on specific connections, as, for example, do most neural- synaptic theories. A non-connectionistic theory of learning is one whose postulated physiological base does not depend on fixed specific neuronal or molecular connections. Most such theories, often in the tradition of Karl Lashley, explain learning in terms of a physiological organization with interchangeable constituents. Pribram (197la, p.9), in discussing his approach to non-connectionistic theories, assumes that “certain interactions important to the organization of behavior and the psychological processes occur in brain tissue, and that these interactions cannot be specified solely in terms of permanent associative connections among neurons.” Non-connectionistic theories do not disregard synaptic relationships, as such relationships obviously impose some order and restrictions on the brain. Non-connectionistic theories merely argue that learning cannot be reduced to a specific set of synaptic connections between specific neurons, but rather involves a more general organization.
A number of arguments are offered in favor of non-connectionistic theories. One set of arguments hinges on the apparent plasticity of sensory and response systems. A person can easily learn to respond to a relationship between stimuli (e.g., always choosing the smaller of two things) regardless of which stimuli are presented on each trial. A person who has learned to write with his right hand can still write to a certain extent if a pencil is put into his mouth, even though this is an entirely different response system. These examples illustrate that people don’t learn simple associations (or connections) between specific stimuli and responses. Perhaps, then, the generalized nature of learning is easier explained by non-connectionistic theories.
Another argument relates to the often reported recovery of function following brain ablation (i.e., situations where parts of the brain have been destroyed or removed). Following such an operation or accident, the organism may lose some function that was associated with the part of the brain that was destroyed. With time, the organism often regains this function. If the function depends on some past learning, the recovery of function suggests that the memory was stored in a number of different places and/or that the memory was stored in a non-connectionistic fashion and could afford the loss of some specific neurons.
Pribram (1969, 1971a, 1971b) has offered an interesting non-connectionistic model of the brain based on parallels with a form of photography called holography. One way to do holography is to take a light source from a laser and split it in two. One part goes directly to a photographic film, while the other part is reflected from the object being photographed onto the film. The film then contains a record, called a hologram, of the interference patterns of the two beams of light. If the hologram is then illuminated with a similar light, it will recreate the object photographed. The recreation is extremely lifelike, with definite three dimensional aspects to it. When looking at a scene from a hologram, the viewer, by moving his head, can see the photographed objects from different angles, including looking around and behind objects.
Pribram suggests that images and memories may be stored and produced by processes similar to those of holography: Neural activity produces momentary wave fronts which result in interactions between different wave fronts. These interactions, or interference patterns, might be recorded by some biochemical change in a manner similar to a hologram.
An interesting property of holography is that almost the entire original scene can be reproduced from just a piece of the hologram. If memories are stored as holograms, this phenomenon might correspond to the brain’s ability to retrieve a whole item of information from any of a number of different pieces of the information. One of the mysteries of human memory is how we can retrieve information so quickly on the basis of just a few related cues.
Another aspect of holography is that an enormous amount of information can be stored in a small area, with the same part of a hologram involved in entirely different information systems. If the brain works in this way, this would explain how we have the flexibility and information storage necessary for learning.
All of the different models discussed above (neuronal-synaptic models, RNA-protein models, glial models, and non-connectionistic theories) are currently very speculative and probably too simplified to completely account for the complex process of learning. (It should be noted, however, that these models are much more intricate than was presented here.) That is, although many of the models are quite complex, they are probably not as complex as the physiological processes underlying learning. But this is a relatively new and incomplete field of
study that is rapidly advancing. Future research findings may help us to integrate the different models, add new constructs, and gradually evolve a model for the physiology of learning and memory.
Between 1917 and 1950 Lashley searched for a place where a memory trace or engram was localized. His general approach was to train experimental animals on specific learning tasks and then to make localized cuts or ablations in the cortex. Cortical destruction in some areas could disrupt the learning or retention of a specific activity (e.g., destruction of parts of the visual cortex might impair some visual tasks). But Lashley was looking for areas where his destruction of the brain would disrupt an engram — a learned association — and not merely simple activities. In 1950 he reported his failure to find such an engram by his procedures (Lashley, 1950).
Lashley suggested that the reason for his failure to find an engram was due to the multiple representation of memories. That is, memories might be stored in more than one place, so that destruction of one memory site would not eliminate all copies of the memory from storage. Another interpretation is that memory systems are too complex to be reflected by the surgical and experimental procedures used by Lashley. Perhaps more extensive cortical and subcortical destruction or more sensitive tasks might have shown more positive results.
Evidence in favor of localized memory traces has been offered by Penfield (Penfield, 1954; Penfield & Jasper, 1954). Penfield’s experiments involved the electrical stimulation of the temporal lobe (along the side of the head) of the cortex of humans. The subjects would then report what subjective experience they had when electrically stimulated in different parts of this cortex. Penfield found that many stimulations seemed to produce recollection of old, and often forgotten, memories. The memory might be a specific visual image or the hearing of a specific sound, or both. These memories included all of the original associations and emotions attached to the memories. The effect of the stimulation seemed to include in the memory everything the person was aware of at the time of original learning. Stimulation of the same area of cortex on successive days always produced the same general experience. The subject could, however, by shifting his attention, pick out slightly different details each time.
This suggests that the brain may work something like a tape recorder, except that where the tape recorder records only sound on the tape, the brain records sound, visual images, emotions, and all other inputs on its “tape.” Memories then, whether occurring naturally or because of electrical stimulation of the brain, involve playing back some part of the brain’s tape.
There are some complexities, however, in interpreting Penfield’s findings. First, although Penfield produced memories by stimulating the cortex, most memories probably involve subcortical components (those parts of the brain beneath the cortex). For example, when a memory elicits an emotion in a person, it is hard to understand neurophysiologically how such an effect could occur without participation of subcortical areas known to be involved in the particular emotion. A second complication is that the electrical stimulation might not be stimulating specific memory traces, but may be stimulating some more general mechanism which in turn produces the memory.
The brain of man is divided into two halves, or hemispheres. The left hemisphere influences the right side of the body, whereas the right hemisphere is more concerned with the left side of the body. The two hemispheres are almost mirror images of each other in appearance and function. Most functions served by some area in one hemisphere also occur in a similar place in the other hemisphere. Thus the brain is highly redundant, which has obvious advantages in terms of protection from the results of brain injury. We might expect then that memory systems might also be redundant between hemispheres.
There are some exceptions to the functional symmetry of the hemispheres. One is that in the more complex animals, such as monkeys and man, one hemisphere is stronger, or dominant to the other hemisphere. In most humans the left hemisphere is dominant to the right. Hence there are more right-handed people than left-handed because the left hemisphere primarily controls the right side of the body.
Another major difference between the hemispheres is that in most humans, after the first couple of years of life, the “speech center” of the brain is primarily localized in the left hemisphere. Thus damage to this area, such as from a stroke, is harder to recover from because of the lack of equivalent functioning in the right hemisphere.
The hemispheres differ in a number of other functions, many of which favor one hemisphere or the other. The reader is referred to the articles by Kimura (1973) and Ornstein (1973) for further information in this area.
Many nerve fibers, and bundles of nerve fibers called commissures, connect the two hemispheres. The corpus callosum, or great cerebral commissure, is the main commissure interconnecting the hemispheres. It appears that a major function of the corpus callosum is the transmission of information from one hemisphere to the other. From the perspective of learning this raises the following questions: Let us assume, as many theorists do, that equivalent memory traces are often laid down in the two different hemispheres. Does this mean that memories are laid down in both hemispheres simultaneously or is the memory laid down in one hemisphere and then transferred (as by the corpus callosum) to the other hemisphere, or is learning a combination of both these processes? Perhaps which of these alternatives occurs depends on the nature or difficulty of the learning task. The process of information going from one hemisphere to the other is called interhemispheric transfer.
The most popular way of studying interhemispheric transfer utilizes the procedures of spreading depression, described earlier, in which waves of depressed electrocortical activity are induced to spread through the cortex of one hemisphere. There are several advantages to spreading depression: First, it is an easy way to make the cortex of a hemisphere largely non-functional, particularly with respect to learning. Second, it stays fairly well restricted to the hemisphere in which it is induced, leaving the other hemisphere normal. Third, the effect is reversible, dissipating over time. Although spreading depression primarily affects the cortex, it also affects some subcortical areas (e.g., the thalamus and hypothalamus). It should also be kept in mind that although spreading depression may disrupt interhemispheric transfer at the cortical level, it may have little effect on any transfer at the subcortical level.
The type of experiment used on interhemispheric transfer is as follows: An animal is trained on a learning task with one hemisphere (the left, for example) depressed by spreading depression. If the animal is later tested on the same task with the left hemisphere functional and the right one depressed, it will behave as if it had never learned the task, for the learning was restricted to the right hemisphere which is now nonfunctional. However, if after the learning with the left hemisphere depressed the animal is allowed a few trials on the task with both hemispheres functional, the animal when later tested with the right hemisphere depressed shows retention of the learning. It seems that the trials with both hemispheres functional allowed the information from the right hemisphere to transfer to the left hemisphere.
Russell and Ochs (1963) trained rats to press a bar under unilateral spreading depression (applied to one side or hemisphere). They found evidence for some interhemispheric transfer if they allowed the rat one rewarded trial with both hemispheres functional. Just letting the rat sit with both hemispheres functional did not result in any transfer. They also reported no transfer if the rat was not rewarded on the one trial (other studies have not shown the necessity for the reward).
Schneider (1967) has questioned the interpretations of the interhemispheric transfer studies and has proposed an alternative stimulus generalization explanation. He argues that unilateral spreading depression provides very pronounced stimuli to which the response is learned. The apparent loss of memory when spreading depression is shifted to the other hemisphere, then, is simply due to the large shift in stimuli. Schneider suggests that learning is not affected, because it could be stored subcortically. For example, suppose a rat with spreading depression on the left hemisphere is trained to turn left in a T-maze. The effects of the spreading depression are that the right side of the rat’s body feels differently and provides different sensory cues (e.g., from paralyzed muscles). So the rat in learning to turn left in the maze may be learning to turn in the direction opposite this different side of its body. If spreading depression is now shifted to the right hemisphere, causing the left side of the body to feel different, the rat might now turn to the right, away from the strange side of the body. Its turning right, when it was “supposed to” turn left, might look like a memory failure when, in fact, it is only due to a change in the stimuli.
Schneider and Ebbesen (1967) redid the Russell and Ochs study, but included another group. All rats were trained to bar-press with unilateral spreading depression. Then the first group, as in the Russell and Ochs study, was given a single rewarded trial with neither hemisphere depressed. The second group, however, received the rewarded trial with the trained hemisphere depressed. Both groups were then tested for bar-pressing with the trained hemisphere depressed. Although both groups showed a greater tendency to bar-press following the one rewarded trial than they did before this trial, the increase was more for the second group (the one with the trained hemisphere depressed). This fits Schneider’s theory, for the second group had more experience (the rewarded trial) pressing the bar in the presence of the stimuli resulting from shifting the spreading depression from one hemisphere to another.
As we can see from the preceding examples, the conclusions to be drawn from interhemispheric transfer studies are still uncertain. Schneider’s studies have shown the important stimulus aspects of spreading depression and how some of the effect can be explained in terms of stimulus generalization. In this sense spreading depression is another variable affecting the organism’s state as in state-dependent learning. But, as originally suggested, spreading depression might also disrupt the formation of cortical memory traces. Future research will have to separate the different effects of spreading depression to clarify exactly how it works.
If the corpus callosum is involved in transmitting information from one hemisphere to the other, a simple way of investigating this is by severing the corpus callosum. Such preparations, called split-brain studies, generally also involve severing other connections, such as other commisures and sometimes the optic chiasm. Subcortical connections, and thus possible subcortical transfer, is left intact.
The idea of split-brain studies is to separate the two hemispheres so that they can be trained separately. Since the left half of the visual field feeds the right hemisphere and the right half feeds the left hemisphere, the experimenter can control what information gets to which hemisphere by where in the visual field he presents the information. If, because the corpus callosum is cut, the hemispheres cannot exchange information, it is as if the hemispheres are two brains learning independently. Although most of the earlier experiments were done on infrahumans, the discussion below will deal only with humans, describing some of the work done by R. W. Sperry and his associates (Gazzaniga, 1967; Sperry, 1968).
The main reason for cutting the corpus callosum in humans is that this surgery appears to decrease the spread and occurrence of epileptic seizures. The operation does not produce any noticeable change in the person’s temperament, personality, or general intelligence. However, split-brain humans do appear to have some deficits in short term memory; they fatigue more quickly in mental tasks, and they often favor the right side of the body (the side controlled by the dominant hemisphere).
Generally the two hemispheres of the split-brain human learn the same things. As both eyes usually see about the same thing and both ears hear about the same thing, generally both hemispheres receive about the same input. It is only in special situations where the hemispheres can be given different inputs that differences can be observed.
Since in humans speech-related functions are generally and primarily localized in the left hemisphere, information presented only to the right hemisphere cannot be responded to verbally. For example, if a split-brain human is shown a picture of a pencil in his left visual field (which feeds to the right hemisphere) and then asked what he saw, he will verbally insist he saw nothing. For the left hemisphere with the speech functions has no information of anything being seen. However, if the person is asked to point with his left hand to what he saw, he will point to a pencil. For although the right hemisphere can’t “speak,” it can point a finger. Or if a funny picture is presented to the right hemisphere, the person may laugh but not be able to verbalize what is funny.
Similarly, consider a split-brain human whose left hemisphere sees (in the right visual field) a question mark, while the right hemisphere sees a dollar sign. If the person is now asked to draw with his left hand what he saw (keeping his left hand out of sight while he draws) his left hand draws a dollar sign. If he is then asked what he just drew, he will say that he drew a question mark.
The split-brain studies suggest that the two hemispheres of a split- brain person can learn independently. Later it will be suggested that there might even be two independent streams of consciousness. The studies also suggest that a function of the corpus callosum is to transfer learned information from one hemisphere to the other.
The split-brain human, if given two simple tasks, each of which can be handled by a single hemisphere, can do two different tasks at the same time. With more complex tasks it is often better if both hemispheres work together.
One hemisphere may try to cue the other hemisphere in on the correct response. In one task a red or green light was presented to the subject’s right hemisphere and the subject then had to verbally (i.e., using his left hemisphere) guess what color he saw. The left hemisphere verbally guessed a response (at a chance level x) even though the right hemisphere presumably knew the correct answer. After a while, however, it appeared that whenever the right hemisphere would hear the left hemisphere say an incorrect guess, the right hemisphere would make the person frown and shake his head. Perceiving the head motion, the left hemisphere would realize that it had made a mistake and would change its answer.
The first two chapters of this book have tried to provide some idea of what learning is and have indicated possible physiological mechanisms underlying learning. In the chapters that follow we will trace the role of learning from input to output. We will start with a description of how learning affects what we perceive in our environment (Perception and Learning). This perceived information is then traced through several stages (Information Holding Mechanisms) on the way to storage. After investigating some mechanisms of learning (Stimulus Contiguity, Feedback) we will show how learning is expressed in the personality of the person (Personality). Finally we will discuss the role of consciousness in learning and speculate on what role consciousness actually serves in man’s behavior (Behavior, Cognitions, and Consciousness).
There must be some physiological changes that are associated with the learning process and that underlie memory storage. Most current researchers look for such changes within the central nervous system, for it is this network of neurons that is largely responsible for processing information and producing behavior. Four categories of physiological theories of learning were discussed: neuronal-synaptic models, RNA -protein models, glial models, and non-connectionistic theories.
The major assumption of neuronal-synaptic models is that learning involves a change in the neurons which makes it easier for some neurons to influence the neural transmission of the neurons with which they have connections or synapses. Such models are popular because neurons are the units of the central nervous system which apparently underlie behavior and because the billions of neurons in the human brain are interwoven in a vastly complex web which seems intricate enough to account for the complexity of human learning and behavior.
RNA molecules are involved in the production of proteins, some of which affect neural transmission. Changes in RNA molecules and proteins have been shown to be correlated with learning. Thus proponents of RNA -protein models of learning suggest that memories are stored in specific molecules of RNA or proteins or both. One set of controversial experiments offered in support of RNA models is the memory transfer experiments. These involve training one set of animals on a task, extracting their RNA and injecting it into a second set of animals, and demonstrating that some memory was transferred with the RNA. Currently there are many complications with this type of research, and theories of the RNA-protein models and exactly how they work are still in dispute.
Glial cells are cells that support the activity and functioning of the neurons, and in this capacity they can affect the firing activity of neurons. A glial theory of learning, then, would postulate that memories are stored in glial cells and are expressed in terms of the glial cells’ influence on neurons. Essentially no-one advocates a glial model of learning, although the glial cells have been incorporated into other physiological models. Ultimately a physiological explanation of learning will probably involve some combination of neurons, glia, RNA, and proteins.
Non-connectionistic theories of learning postulate that learning is not based on specific connections between specific neurons, but rather depends on coherent patterns of electrical activity of groups of neurons. Thus memory is recorded in terms of an electrical pattern. One advantage to non-connectionistic theories is that they allow for the possibility of specific neurons being destroyed without necessarily impairing memory, so long as other neurons can still produce the required patterned activity.
Considerable research has been directed toward identifying exactly where memory traces are stored. Some animal research has involved selectively destroying parts of the brain in an attempt to obliterate memory traces. Such attempts have not been successful in localizing memory traces, suggesting to some researchers that specific memories may be stored in several different places. Electrical stimulation of some areas of the human brain has caused the subjects to recall specific memories, perhaps because the stimulation somehow activated a memory trace. There is also evidence suggesting that information stored in one hemisphere of the brain may be transferred via nerve fibers to the other hemisphere, where the information is then also stored. Severing these nerve fibers produces a split brain where the two hemispheres learn things independently to some degree.
Deutsch, J. A. (ed.) The Physiological Basis of Memory. New York: Academic Press, 1973.
Grossman, S. P. A Textbook of Physiological Psychology. New York: Wiley, 1967.
Gurowitz, E. M. The Molecular Basis of Memory. Englewood Cliffs, N.J.: Prentice-Hall, 1969.
John, E. R. Mechanisms of Memory. New York: Academic Press, 1967.
Ungar, G. (ed.) Molecular Mechanisms in Memory and Learning. New York: Plenum Press, 1970.