| Class Schedule || Lecture Notes || Assignments || Misc. Links |
Learning
I. What is Learning?
Learning: typically defined as a relatively permenant change in behavior due to experience, and not to maturation, drugs, illness, injury, or fatigue.
II. Two Broad Classes of Learning:
A. Classical Conditioning: Learning About Stimuli as Signals
--a response naturally emitted by the organism in response to a stimulus, coming to be evoked by another sitmulus after repeated associations with the first.
Unconditioned--unlearned
Conditioned--learned
Classical conditioning takes advantage of the fact that organisms are born with a limited set of reflexes; that is, for each species, there are some specific environmental events (stimuli, or S) that reliably evoke fixed behavioral actions (responses, or R).
The term unconditional is instructive. The relationship between the unconditioned stimulus and the unconditioned response exists unconditionally--the first time, and every time, the stimulus is presented, the response reliably follows. It is in the nature of the organism to behave this way.
Classical conditioning is said to occur when a previously neutral stimulus, such as a tone or light that normally is of little interest to the organism, becomes associated with an already existing reflex to the extent that it will, by itself, evoke a response. This new reflex is said to be conditional, in that its ability to evoke a response depends upon the stimulus having been associated with a previously existing reflex.
To begin our exploration of this kind of learning (classical conditioning), let's look at an example--Pavlov's original experiment:
Periodically, a tone was sounded, followed shortly thereafter by meat powder being placed in the hungry dog's mouth (each pairing of tone/meat powder is referred to as a trial). Meat powder causes a hungry dog to salivate, whereas tones typically have little effect on salivation. The dog's salivation to meat powder is an unconditioned response--it is in-born, in that dogs do not have to learn to salivate when food is placed in their mouths.
Initially, the dog shows little responsiveness to the tone. Over time, however, the dog comes to salivate at the sounding of the tone alone. When this occurs, classical conditioning has occurred, in that a new, or conditioned response has developed. The animal behaves as though it is using the stimulus of the tone to signal that another stimulus (the meat powder) is coming.
Remember--classical conditioning is about learning to use stimuli as signals.
In describing the process by which a new reflex develops, Pavlov coined a number of terms that are still the basic terminology of conditioning:
Unconditioned Stimulus (UCS)--a stimulus that, without conditions, will elicit a predictable response.
Unconditioned Response (UCR)--a response that, without conditions, results predictably from an unconditional stimulus.
Conditioned Stimulus (CS)--a stimulus that will elicit a predictable response because of its previous pairing with a previously occurring reflex.
Conditioned Response (CR)--a predictable response to a stimulus that has influence because of its previous pairing with a previously occurring reflex.
Diagrammatically, the experimental setup can be represented in this way:
| Unconditioned Reflex | UCS |
-------> |
UCR |
| food |
|
salivation |
|
| Conditioned Reflex | CS |
-------> |
CR |
| tone |
|
salivation |
B.
Operant/Instrumental Conditioning: Learning About Responses and Outcomes
(the animal operates on the environment,
is instrumental in obtaining it's own reward)
In this kind of learning, what is learned is the relationship between a behavior that the animal emits (its response) and the outcome or consequence of that behavior.
Concepts from Operant Conditioning:
Shaping: reinforcing successive approximations of the desired behavior.
Kinds of Consequences:
Reinforcement--anything that increases the likelihood of a behavior occurring again
a. Primary--meets a primary need, like food, water, air, shelter
b. Secondary--becomes associated with a primary reinforcer through repeated pairing, and hence can also be reinforcing. Money is a good example. You cannot eat it or drink it or use it as shelter, but it can be used to get those things and hence is a powerful human reinforcer.
Two kinds of reinforcement:
Positive reinforcement--direct reward
Negative reinforcement--removal of an aversive stimulus
Punishment--anything that decreases the likelihood of a behavior occurring again
Two kinds of punishment:
Direct punishment--applying an aversive stimulus after the organism emits a behavior you wish to see decrease in frequency
Response cost--removing an appetitive stimulus after the organism emits a behavior you wish to see decrease in frequency
Biological Constraints on Learning:
I. The Problem: The Tenents of Classical Behaviorism
1. All organisms learn in the same way, and all general laws of learning apply to all creatures.
2. Learning REQUIRES reinforcement or punishment; no learning occurs without it.
3. Any combination of stimulus-response-outcome can be learned.
4. All behavior is a consequence of learning
II. Disproving the Tenents: The Evidence Against Them
1. Latent Learning
If behavor is not reinforced or punished, then according to classical learning theory, NO LEARNING SHOULD TAKE PLACE! Yet--rats that have been allowed to explore a maze without any reward run that maze faster than rats with no experience in it. This phenomenon of apparent learning in the absence of any obvious reinforcement or punishment is called latent learning.
2. Breland and Breland, "The Misbehavior of Organisms"
The Brelands trained animals for circuses, zoos, TV. Ran into some problems not predicted by the classic behaviorist theory. Ex. trained pigs to put coins in a piggy bank for a food reward. At first, no problem. Then--pigs began to root coin, toss it around, etc. Would not put it in the bank. Similar problems with racoons. If all species learn in the same way, and any old stimulus-response-consequence combination can be learned, these problems should not be arising when training animals!
3. Garcia's Work With Taste Aversion
Results, summarized:
| Stimulus | Bright-Noisy
Water |
Sweet-Tasting
Water |
||
| Consequence | Sickness |
Shock |
Sickness |
Shock |
| Outcome | Did not avoid drinking |
Avoided drinking |
Avoided drinking |
Did not avoid drinking |
If any old stimulus-response-outcome combination can be learned, there should have been no differences between these four groups in their learning to avoid drinking water!
4. Species Differences (M. Bitterman)
If the idea of classic behaviorism that the general principles apply to ALL organisms is true, and as long as the animal has the physical ability (ex. pelicans and frisbee), then there SHOULD BE NO species differences in learning styles. Well, are there species differences?
CONTRAST EFFECTS
Let's look at one assumption of the classic behaviorist traditionl. The old idea was that the larger the reward, the stronger the connection the critter would make between the stimulus, and the response (like the difference between getting paid $2.00 to wash your parents' car, and getting paid $200.00. You might wash your parents' car more often, in the latter case). Similarly, the old idea was that the more preferred the reward was, the stronger the connection would be between stimulus and response. (like getting $2.00 or 2 Hersey's kisses for washing the car).
Rats will work harder for a larger reward (a whole cookie instead of just a piece of cookie) and they will work harder for a more preferred reward (for chocolate chip cookie over rat chow).
BUT goldfish and turtles show no change in their performance as a function of quantity or quality of reward. They won't work any harder for a preferred food, or for a larger reward. For example, rats and contrast experiments show an effect of contrast on performance Ex. bran vs. sunflower seed, or 4 vs 40 sunflower seeds). Goldfish do not (4 or 40 worms, no diff). Turtles are like fish. Makes no diff to them.
MATCH VS MAXIMIZE
When offered a choice between responses with different pay-offs, animals can use one of two strategies. Let's say you can divide your time at a casino by playing a slot machine that pays off at a rate of 80%, or one that pays off at a rate of 30%. What should you do? The thing to do to MAXIMIZE your payoff is to stick with the machine that pays off 80% of the time, and do not play the other machine. Rats can maximize (so will monkeys).
The other option is to match your rate of play on the two machines to their approximate rate of payoff. That means that about 80% of the time you would play the machine that pays off 80% of the time, and about 30% of your time would be spent playing the machine that pays off 30% of the time. This strategy, while common, is not a very effective one.
Goldfish always match (never maximize), and they do so randomly. And not all tasks are the same. Ex. give a spatial task, vs. a visual task. Mammals and fish will stick with one strategy maximizing or matching) in either task. But turtles will use matching in visual tasks(fishlike), maximizing in visual tasks (ratlike).
Some Learning is More Equal Than Others: The Idea of Ecological Relevance
Pigeons trained to peck keys for a food reward will do so with an open mouth, and CANNOT BE TRAINED to do so with a closed mouth. In contrast, pigeons trained to peck a key for a water reward will do so with a closed mouth, and CANNOT BE TRAINED to do so with an open mouth. Similarly, pigeons will take longer to learn a short key peck for water, and longer to learn a long key peck for food. Why? Because pigeons peck for food with an open mouth and short pecks, while they peck for water with a closed mouth and longer pecks.
These species differences reflect different levels of organization. Some critters may be better equipped to learn one thing over another. Thus learning, just like any other process, cannot be considered apart from its organismic constraints. There are no general laws of learning that apply across the board. Rather, these principles are relational. Each instance should be considered as a specialized capability shaped by selection and understandable to us only with reference to the ecology of the organism.