At the 2004 convention of the California Association for Behavior Analysis (CalABA), I participated in a panel discussion with Janet Twyman, Greg Stikeleather, and Hank Schlinger on the interpretation of complex human behavior. At the outset of my remarks, I made a great show of setting my watch so that I should stay within my allotted time. I explained, “I bought this watch for ten dollars at WalMart. I like that it has all twelve numbers on the dial, in big black numerals. I realize that it's a piece of junk, but as my friend, Uchena Mbanefo, says, ‘De gustibus non est disputandum’ ... there's no accounting for taste.”
Since the audience was getting restive, I then began my talk about one-trial learning. In the Behavior of Organisms, Skinner (1938) reported that a feeder-trained rat, well adapted to the experimental apparatus, commonly shows an immediate change in the strength of lever pressing following a single reinforcement. His evidence for one-trial learning contrasted with the notion that there is a “learning curve” along which behavior changes on successive trials. Skinner concluded that reinforcement causes an instantaneous change in the strength of behavior; learning curves are the result of poorly controlled conditions, of experimental preparations in which concurrent contingencies contaminate the effect of the target contingency. For example, Thorndike's cats, when placed in puzzle boxes, engaged in a lot of irrelevant escape behavior that degraded the regularity of changes in strength of stepping on the treadle that opened the door. Under the special conditions of tight experimental control, we should expect to see one-trial learning.
The discrepancy between behavior under tightly controlled and loosely controlled conditions occasions no surprise. Toss a ball into the air, roll it down an inclined plane, drop it from a tower, bounce it, spin it: the predictions of our physics books will be slightly in error, for we have not controlled air resistance, temperature, friction, the elasticity of materials, and so on. We might expect, then, that one-trial learning emerges under only idealized conditions, that it is rare in nature.
But one-trial learning is commonplace in human behavior, so commonplace that we would be astonished if it did not occur. How many trials would it take you to learn that the red box is full of scorpions and the blue box is full of diamonds? Or that the switch with the red indicator light turns on the computer? Or that you got a B on the exam, not the A+ you were hoping for? Or that the weatherman predicts a blizzard tonight? An enterprising student recently gave me a small pine box containing gourmet chocolates; I had no trouble repeatedly operating the hinged lid after my first trial. We are continually learning new routines, new computer commands, new names, the route to new bathrooms, how to operate new-fangled hotel keys, and so on, and it is not uncommon for such learning to occur on a single trial under conditions that are by no means tightly controlled.
What are we to make of such examples? Are we “smarter” than rats? Another explanation is possible: Several incompatible behaviors may each be strong, but only one can be emitted at a given time. To the observer, only one response appears to be strong, for only one is emitted. However, a slight shift in controlling relations may cause another to emerge in strength. When a dominant response is put on extinction; it declines in strength just until it becomes weaker than competing responses. It loses no more strength because it is no longer emitted. It remains “just below threshold,” so to speak, in the sense that it can easily be nudged into dominance again. In familiar contexts, human behavior is commonly a cauldron of competing responses. I have argued the point in greater detail elsewhere (Palmer, 1997):
One-trial learning is commonplace in human behavior only when the target behavior is already a strong response in the subject's repertoire but happens to be weaker than other behavior in the prevailing circumstances. The behavior to be conditioned when an adult human learns that the weatherman predicts rain, or that a wall switch turns on the overhead fan, or that the soup-of-the-day is clam chowder are all responses that have been thoroughly conditioned under similar circumstances. When we listen to a weather report, we are already inclined to assert that it will rain, or, as it may be, that it will be fair. The same variables that make us tune in to the weather report in the first place increase the likelihood that we will make statements about the weather. The weather report itself merely selects some subset of those statements. When we discover a switch, we are already confident that it will turn something on. In a restaurant, “clam chowder” is one of some dozens of responses that are already weakly evoked by the setting long before the waiter recites the daily specials. The effect of the contingency of reinforcement is to slightly modify the control of the target response by the current setting so that it becomes stronger than the myriad other responses that tend to be evoked in that setting. At any moment, the potential behavior of an experienced organism can be thought of as a panorama. Reinforcement of a well-practiced response is analogous to increasing the illumination on a figure in the panorama to make it stand out from the background.
In contrast, some responses have little or no baseline strength. They have to be drawn into the panorama from scratch, as it were. When we hear someone speak in an unfamiliar tongue we usually find ourselves unable to repeat what was said. Even repeating a single word might be difficult. We may have to try many times before we get it even approximately right, particularly if it is composed of unfamiliar phonemes. When we first learn to ride a bicycle, to operate a clutch, to swim, to juggle oranges, to “walk the dog” with a yo-yo, the relevant behavior is shaped for the first time, and it commonly takes many trials before successful behavior consistently emerges. The target behavior has no baseline strength; our behavior is highly erratic at first and smoothes out over repeated trials. (pp. 276–277)
Thus one-trial learning is commonplace in human behavior because we often find ourselves in familiar circumstances where relevant behavior, even if not prepotent, is already strong. A single reinforcement can alter the delicate balance of competing responses.
But formidable puzzles remain: To illustrate the point, I interrupted my CalABA talk with a pop quiz: “How much did I pay for my watch?”
There was a chorus of responses from the audience, “Ten dollars.” Even the silent ones, when pressed, acknowledged that they knew the answer, although two people admitted ignorance because they hadn't been “paying attention.” However commonplace this performance may be, it is puzzling and requires an interpretation. It is not just another example of one-trial learning, for there was no trial—at least no trial in the usual sense. I simply emitted some verbal behavior. (“I paid ten dollars for this watch at WalMart . . . .”) There was no apparent behavior on the part of the audience and no apparent reinforcement. There was no contingency. In what sense can this be called a trial? But learning had occurred, as the results of the quiz showed. Confining oneself to a consideration of observable variables, one might conclude that mere exposure to a pattern of verbal stimuli is sufficient to effect a change in relevant behavior in verbally sophisticated subjects to a different set of verbal stimuli (i.e., “How much did I pay?”). Is this not zero-trial learning?
I continued with the quiz. “Where did I buy the watch?”
“WalMart.”
“Why do I like it?”
“It has big black numerals.”
“What did my friend say about my purchase?”
“There's no accounting for taste.”
“No, no. What did he say, literally?”
There was a long pause. Nobody seemed to know. Then Ernst Moerk, in his delightful German accent, piped up, “De gustibus non est disputandum!” Then, in an aside, “I was a Latin scholar.” Thank goodness for Ernst. If my demonstration were not to fall flat, the Latin phrase had to be unfamiliar to most of the audience but not all. It was a near miss.
“What was my friend's name?”
Nobody knew, not even the erudite Ernst Moerk, for Latin scholar and polyglot though he was, he had no repertoire with respect to Nigerian names.1