Hebbian Learning in the Aplysia Californica, or, I Know Why the Sea Slug Sings
Abstract: We discuss the basic principles of Hebbian learning, and explore some of the research conducted with the sea slug Aplysia Californica, which provides evidence that long-term potentiation is the mechanism behind Hebbian learning.
Introduction
One major problem with studying human learning, or the learning behaviors of any higher animal, is that our brains extremely complicated, and it is often difficult to determine which causes are responsible for which effects. Our neurons are interconnected and abundant, and it can be extremely difficult to identify which particular neurons are firing at any given time.
An alternative to studying learning in complex neurological bundles like the human brain is to study learning in very simple mammals. When there are relatively few neurons, it becomes easier to study learning from a neurological point of view.
Background Info & Elaboration
Way back in the day, Donald Hebb (1949) proposed that learning was function of a strengthened connection between two neurons, as opposed to the generation of entirely new neurons. If two neurons fire simultaneously, this could strengthen the connection between two neurons. When the firing of one neuron is responsible for the firing of a second neuron, the strength of the connection between those two neurons is increased. From a neural network standpoint, we would say that the “weights” between these two nodes have increased, meaning that an activation of the first neuron is more likely to trigger the firing of the second neuron. Such increased connections are sometimes referred to as “long term potentiation”. This helps explain some conditioning responses. Unsurprisingly, this theory has come to be known as “Hebbian Learning.”
Donald Hebb's interest in the prevailing behavorist methods of his day led him to speculate that there was a similiar mechanism occuring in the brain. Hebb first began exploring associative learning at the neural level with rabbits. The experiment was simple: on each trial a tone is turned on (CS, conditioned stimulus), and 250 msec later, a puff of air to the eye (US, unconditioned stimulus) is presented. The tone and air puff terminate simultaneously. Initially, the rabbit does not respond to the tone, but when the air puff is administered, the inner eyelid called the "nicitating membrane" slides over the eye to protect it. Following repeated pairing trials of tone + puff, the nicitating membrane response begins to be elicited by the tone (CS), before the puff of air (US) is administered. Hebb speculated that the neronal pathways that transmit the air puff (US) stimuli and tone (CS) stimuli to the brain terminate on the dendrites of the pukinje cells, thus forming one processing unit. This process is summarized in the oft-quoted, "Neurons that fire together wire together", which is essentially the idea behind long-term potentiation. In Hebb's experiment, long-term potentiation is exemplifed by the increased sensitivity of the neurons responding to the tone (CS) as a function of the neuron responding to the air puff (US). Long-term potentiation is believed to be at the basis of synaptic plasticity, which is the connection of two connections, or synapses, to increase in strength. This is particularly important for learning because memory is thought to be encoded in synaptic connections. Without memory there would be no higher-level learning in the human species.
Hebbian learning is especially interesting when it comes to the concept of “neuroplasticity,” or the theory that our brains are very efficient at rewiring themselves.
The seminal work by Eric Kandel was essential for providing evidence that Donald Hebb's speculations were indeed true. However, Kandel argued that Hebbian learning does not necessarily involve an associative response from the environment. Rather, a much simplier mechanism, such as nonassociative learning, could induce Hebbian learning. Nonassociative learning refers to a change in behavior that results from repeated experience of a single stimulus or of two or more stimuli that are not temporally or spatially related. The three most commonly studied forms of nonassociative learning are: 1) habituation, 2) sensitization, and 3) dishabituation. Habituation is simply the decrease in the strenth of the behavioral reaction to a repeatedly presented stimulus. Sensitization is the increase in an animals' responsiveness to stimuli following a noxious stimulus. Finally, dishabituation is when a stimulus releases a response from habituation, usually following an unassociated noxious stimulus. Kandel demonstrated Hebbian learning across all three nonassociative forms of learning in the species, Aplysia Califonica.
The Aplysia Californica is a sea slug that has relatively few neurons (several thousand), so we have much less to look at than with more complicated systems such as the human brain. Eric Kandel has been instrumental in the demonstration of Hebbian learning through his ingenious goofing around with the sea slug (Antonov, Antonova, Kandel, & Hawkins, 2003).
Methodologies/Algorithms/Techniques
The Aplysia has a withdrawal response that can easily be exploited in order to study Hebbian learning. If the slug’s mantle (a shell-creating organ), siphon (an organ used for movement) or a gill is stimulated, the slug will reflexively try to get away. Since the Aplysia has so few neurons, we know which areas of its bodies follow which neurons. So by touching the slug and observing its response, we can predict almost exactly which neurons are being triggered.
Experiments/Results
By stimulating the neurons of the researchers can selectively trigger many of the Aplysia’s neural pathways, in order to determine what kinds of stimulation produce habituation (a decrease in response to a stimulus), and sensitization (an increase to stimuli in general as a result of a first, extremely stimulating stimulus). They exposed the mantel, gill, and siphon of the slug, and then stimulated various pathways, measuring the slug’s response via microcomputer. Castellucci & Kandel (1976) found that, if the Aplysia is given moderate stimulation over time, the Aplysia’s response to that stimulus grows weaker with each subsequent stimulus, thus demonstrating habituation. However, Castellucci et al. (1970) found that if the Aplysia is exposed to a very strong stimulus, then subsequent, moderate stimuli provoke a greater-than-normal reaction, as a result of the initial strong stimulus, thus demonstrating sensitization.
In order to determine exactly which chemicals were involved, and to demonstrate that long-term potentiation was the mechanism responsible for Hebbian learning, Anotov et al. (2003) exposed the slug to a series of neuroinhibiters, and then stimulated the slug on various pathways. One such inhibitor is the PKA inhibitor, which is thought to block chemicals important to long term potentiation. As the researchers hypothesized, the introduction of a PKA inhibitor blocked the sensitization and habituation normally observed as a result of Aplysia stimulus. Based on these results, Anotov et al. have concluded that long-term potentiation is indeed a key mechanism in Hebbian Learning.
Concluding Remarks
How much can long-term potentiation in an invertebrate tell us about Hebbian learning? On one hand, since invertebrates like the Aplysia Californica are so simple, they provide a pretty clear proving ground for neurological research. On the other hand, if the ultimate goal is to study human learning by observing animal learning, one could argue that our neurological systems are too complex to be accurately reflected in the neurology of a simple sea slug. The debate rages on.
The relationship between Hebbian learning and computational neural networks is not trite. Neural networks can display complex global patterns from the interconnections of basic processing units. At it's core, a neural network is a mathematical model for information processing, however, it can be easily equated to processes underlying true (biological) neural connectivity. For example, unsupervised learning in neural networks is analagous to Hebbian learning as described above. Please see "Fundamentals of Natural Computing: Basic Concepts, Algorithms, and Applications, L. N. de Castro, Chapman & Hall/CRC, June 2006" for more information.
REFERENCES
Anotov, Anotova, Kandel, & Hawkins (2003). Activity-Dependent presynaptic facilitation and Hebbian LTP are both required and interact during classical conditioning in Aplysia. Neuron, 37, 135-147.
Castellucci, V., & Kandel., E.R. (1976). Presynaptic facilitation as a mechanism for behavioral sensitization in Aplysia. Science, 194, 1176-1178.
Castellucci, V., Pinsker., H., Kupfermann, I., & Kandel, E.R. (1970). Neuronal mechanisms of habituation and dishabituation of the gill-withdrawal reflex in Aplysia. Science, 167, 1745-1748.
Hebb, D.O. (1949). Organization of behavior. New York: Wiley.
Kalat, J.W. (2004). Biological Psychology (8th ed.). Belmont, CA: Wadsworth/Thomson Learning.
Learning, Memory, and Plasticity
***Changes made by Nick Duran are in "bold italics"***