Feature

Without learning, a species could not adapt and endure. To study this key process, experimental psychologists have for more than half a century sorted their knowledge into either Column A, classical conditioning, or Column B, instrumental (operant) conditioning.

But what if there's only one column?

In January's Journal of Experimental Psychology: Animal Behavior Processes (JEP: APB, Vol. 30, No. 1), John Donahoe, PhD, a researcher in the psychology and neuroscience programs at the University of Massachusetts at Amherst and Rocío Vegas, PhD, formerly a doctoral student in Donahoe's laboratory and now a professor at the Central University of Venezuela, present experimental evidence consistent with a unified principle embracing both approaches, gathered from close study of conditioning in pigeons.

If the unified principle appears to explain learning better, theorists may start bridging the long-standing divergence between Pavlov's classical conditioning and Thorndike's operant conditioning. A better explanation would advance psychological science--and could help clinicians gain fresh insight into how people learn, with possible implications for training, extinguishing phobias and more, some observers say.

In addition, the research is good basic science in that it gives "a more thorough description of what's going on in learning," says psychologist Charles Catania, PhD, a veteran behavior analyst at the University of Maryland Baltimore County.

Learning Theory 101

Researcher Donahoe has been assessing the unified reinforcement principle for 20 years. Now, along with co-author Vegas, he claims to have found that the critical relation in classical conditioning is not an association in time between two different stimuli, but rather between a stimulus and behavior.

As a quick refresher, classical conditioning, first described by Russian physiologist Ivan Pavlov, takes place when one stimulus that is reliably paired with a second stimulus begins to be responded to as though it were the second stimulus. In the classic example Pavlov described in dogs, salivating can become conditioned to a new, neutral stimulus (the conditioned stimulus, for example, a buzzer) that is reliably paired with food (the unconditioned stimulus). Soon the buzzer will elicit salivating (now a conditioned response) all by itself. The dog will learn to salivate when it hears a buzzer, even if food is no longer present.

By comparison, instrumental conditioning, first investigated by American psychologist Edward Thorndike, concerns what happens when a behavior (the instrumental response or operant) is immediately followed by an unconditioned stimulus, as when lever-pressing is followed by food. What happens is that the animal is more likely to press the lever the next time it sees the lever. In operant conditioning, the food is usually called a reinforcer instead of an unconditioned stimulus, and the response elicited by the reinforcer is commonly ignored.

Donahoe says that over time, "Each procedure developed its own technical vocabulary and its own rich set of phenomena to investigate." However, he was not convinced that different conditioning procedures meant different learning processes.

Initially, he doubted whether, in nature, two different processes made sense. Psychologist James Kehoe, PhD, explains why. "In evolution, it is more likely that a single learning process, undoubtedly with considerable diversity in implementation, would be the case," says Kehoe, an expert on classical conditioning at the University of New South Wales.

Donahoe, together with his then students José Burgos (now at the University of Guadalajara) and David Palmer (now at Smith College), tested his ideas using computer-simulated neural networks. They found that the simulated networks "learned" in ways consistent with a unified principle, whether they used either classical or operant procedures. In addition, the unified principle accounted for important aspects of a wide range of learned behaviors, including acquisition, extinction, conditioned reinforcement, discrimination, generalization, faster reacquisition following extinction, timing and revaluation. But would these simulations hold up in the lab?

To test his ideas, Donahoe needed to find a procedure that would allow him to remove the hallmark of the classical conditioning procedure--that the unconditioned stimulus and the unconditioned response occur very close together in time.

Enter the pigeons

So he developed a method that, using digital technology and remote sensors, enables researchers to study learning more closely by separating the temporal relation of the conditioned stimulus to the unconditioned stimulus from its relation to the unconditioned response; for example, the relation of a buzzer to food from its relation to salivation. He searched for a procedure where the unconditioned response did not follow so close in time after the unconditioned stimulus.

Donahoe found his late-blooming response--to test the unified principle--in pigeons. When a pigeon's oral cavity is injected with water, the bird will, after a brief but significant delay, swallow and make rapid tongue movements as part of the swallowing reflex. In this throat-movement response, the delay between water injection (unconditioned stimulus) and throat movement (unconditioned response) allowed Donahoe and Vegas to disentangle relationships in learning.

To make a long story very short, the researchers studied 66 pigeons in four experiments. Thanks to the pigeons' delayed response, they observed learning with the key variables of unconditioned stimulus and unconditioned response clearly separated.

In the core experiment, two different stimuli, water injections and a colored light (the conditioned stimulus), preceded the swallowing response in either of two orders. The pigeons learned to respond with throat movements to the visual stimulus no matter which stimulus (conditioned or unconditioned) came first--as long as the visual stimulus preceded the throat-movement response. Both orders resulted in conditioning.

Donahoe and Vegas conclude that the association of the conditioned stimulus with the unconditioned response, not with the unconditioned stimulus, is fundamental to learning. Thus with both classical and instrumental conditioning, it is the relation of the conditioned stimulus to subsequent behavior that produces learning.

Says Donahoe, "Our view is that Pavlov and Thorndike identified two different procedures for the study of learning, but that the two procedures have different outcomes that are, nonetheless, expressions of the same fundamental conditioning process."

New pecking order?

Given the weight of a century of research, the report is bound to generate discussion. JEP: APB Editor Nicholas Mackintosh, PhD, of the University of Cambridge, views the findings as a valuable re-conceptualization of Pavlovian conditioning, but he doesn't think they have "much to say about the distinction, or lack thereof, between Pavlovian [classical] and operant conditioning." Kehoe says that at the least the evidence of a unified principle is a useful reminder that the law of association applies equally to classical and operant conditioning.

Applying this abstract notion to real-live brains, Kehoe notes that the Donahoe-Vegas finding supports neural models in which inputs (stimuli) get mapped onto outputs (behaviors). He says the research suggests that in the brain, "learning will be based on changes in chains of neurons that connect a conditioned-stimulus input to a conditioned-response output."

It could take a while for the findings to filter into accepted theory and practice. Donahoe's early reflection is that psychologists who wish to help people acquire a desired behavior should assess both the unconditioned and instrumental responses and make sure they don't interfere with one another during the conditioning process.

Kehoe adds that Donahoe's work suggests that learning by association would benefit from overt rehearsal--for example, saying a name aloud when you first meet someone, or actively organizing and writing summaries when studying complex material.

As for clinical applications, such as treating learned fear, Catania has a qualified endorsement. He says, "Every new finding doesn't give us a new procedure, but [the unified principle] makes it easier to account for the generalization of fear."

For now, Donahoe prefers to stay in the realm of experimental analysis, saying, "Clinical practice must be tailored to the particulars of the conditioning history of the client. What a unified reinforcement principle provides is a guide to the formulation of better clinical practices and an explanation of how existing practices work."

Rachel Adelson is a science writer in Raleigh, N.C.

Further Reading

Read more about John Donahoe's research on learning at www.umass.edu/neuro/faculty/files/donahoe.html.