LOGAN'S LITERATURE AND LAWS OF LEARNING 1/1/2000 AVOIDANCE CONDITIONING: Preface Preface Introduction Avoidance Paradigms The Literature of Learning is replete with terminological misnomers. We have been most remiss in employing terms from the vernacular, trying to give them a scientific meaning while also carrying the burden of surplus meaning from their everyday contexts. Some writers introduce a new term when an old term is already available. Other terms are used in more than one context, while sometimes, the same term is used for basically different operations. Many times, a term has both an empirical and a theoretical meaning, and it is often unclear which meaning is intended. All of these terminoligical difficulties are illustrated within the domain of avoidance conditioning. A good example of mixing theoretical analyses with empirical operations is referring to the procedure in which a response could prevent the occurrence of an aversive event as "active" avoidance. This is then contrasted with "passive" avoidance, in which the organism can prevent the occurrence of an aversive event by not making a designated response. I have symbolized the former paradign as Conting(R,-S) to indicate the contingency that the second event does not occur if the first event does occur. What frequently is called passive avoidance would be symbolized Conting(-R,-S) to indicate that the second event does not occur if the first event does not occur. Such a notion is inconsistent with a systematic operational analysis. Organisms are continually not-making an indefinitely large number of responses. The so-called "passive avoidance" paradigm is actually punishment. The principal difference between the avoidance operations and the punishment operation is that the stimulus event is scheduled independently of behavior in avoidance but it is dependent on behavior in punishment. Although it is true that there is no punishment if there is no response, there is also no punishment for not making dozens of other possible responses. The fact that organisms learn not to make a punished response is due to the Conting(R,em-S) and not the reciprocal notion of Conting(-R,-em-S). This is not to prejudice the theoretical explanation of either avoidance or punishment, but paradigms should be labeled by operations, not by theoretical conceptions. It would be impossible to purge the extant literature of the term, passive avoidance, but I have made the appropriate translations in this empirical analysis. Otherwise, I would have also to change the reinforcement paradigm, which is Conting(R,em+S), to a bizarre notion of "passive non- reinforcement" symbolized Conting(-R,-em+S). Rigorous terminology is one long-term goal of the Quad-L project. Introduction Many stimulus operations are not "pure" in the sense of being completely independent of the organism's behavior. One of these interdependent operations is when a stimulus event is scheduled to occur at some point in time, but its occurrence is precluded (or postponed) by the prior performance of some designated behavior. The traditional label for this operation is avoidance conditioning. As noted by Sheffield (1972), the term "avoidance" is prejudicial in that it implies a cognitive awareness of the impending event and the operational contingency. That is to say, it suggests that the organism "knows" that some injurious event is about to happen and deliberately performs some response in order to prevent it. Actually, the generic paradigm, which I have symbolized Conting(R,-S), is indifferent as to the emotional significance of the stimulus event. Accordingly, I am reluctantly using the term "avoidance" conditioning only because it is so well established in the Literature of Learning. In a generic sense, avoidance conditioning is defined by the contingency that a response is followed by the nonoccurrence of some emotionally-negative stimulus event. But such a definition is practically meaningless because "contingency" refers only to a temporal relationship, not necessarily a causal one. Typically, the AR does indeed preclude the consequent event, but the generic paradigm must allow for adventitious outcomes. For example, a man might urinate after engaging in unprotected sex and subsequently not develop a sexually-transmitted disease. He may infer that urination is an effective avoidance response. In similar fashion, some religious ceremonies and superstitious rituals are hypothetical avoidance responses. Hence, the expression Conting(R,-S) permits but does not require an instrumental contingency. Because Conting(R,-S) implies only a temporal relationship, it would be an enormous over-generalization if taken in isolation. Every response of an organism is followed by the non-occurrence of countless events. It would be trivial in the extreme to say, for example, that every time you go inside a house, you are not struck by lightning. Going inside is an AR re lightning only in the context of a thunder storm. More generally, one needs to combine Conting(-R,S) with Conting(R,-S) to identify an avoidance paradigm. Avoidance Paradigms Historically, avoidance conditioning provided the first stumbling block to theorists who hoped to build a general theory of learning based on conditioning principles. One such principle is that of experimental extinction. Now in avoidance conditioning, as soon as the excitatory process gets strong enough to produce a response, the reinforcing event does not occur. At best, it would seem that the avoidance procedure should lead to a low-level oscillation between acquisition and extinction. In fact, in many situations, avoidance conditioning is as good or better than nonavoidance conditioning. This was dubbed the "avoidance paradox." A response followed by "nothing" is maintained rather than extinguished. Theorists scrambled to find something to reinforce an avoidance response (AR). In the quest to understand avoidance conditioning, a rather large number of variations were designed. The first distinction is whether the occurrence of the scheduled event is cued by a ready/warning signal: Cued Avoidance. The simplest cued procedure is a variation of the classical conditioning paradigm. The instrumental Conting(CR,-US) is added to Pair(S,US). If the US is aversive (emotionally negative, em-S), the CR becomes a classical avoidance response (ARc). This is a simple procedure because the US, when it occurs, is a fixed event and because the pairing operation essentially insures that the ARc will occur. (It is worth noting that there may be an element of avoidance in some classical conditioning preparations; an acid US is less aversive to a moist mouth, and an airpuff US is less aversive on a closed eyelid. Indeed, finger- withdrawal conditioning, while usually called classical, is actually an avoidance procedure if withdrawal removes the finger from the electrode. It is sometimes difficult to determine whether the avoidance contingency affects the underlying processes.) For example, consider the effect of US intensity. In the pure classical procedure, USI usually affects the magnitude of the UR. This encourages the interpretation that magnitude of the CR is conditioned per se. But in the avoidance procedure, precluding a more intense US may be more reinforcing. This effect may depend on the magnitude of CR required to qualify as an ARc. Hence, one needs to attend to procedural details in making generalizations.) When the avoidance response is not a component of the UR to the US in Pair(S,US), a trial-and-error learning process is implicated. That is to say, the organism must discover the response (ARi) that is instrumental in precluding the impending em-S. This generates at least 18 variations, 6 of which are diagrammed in the accompanying figure. The first distinction is whether, in the classcond component of the procedure, the antedating S is brief (trace conditioning) or prolonged until the US (delayed conditioning). In the ltter case, a further variation is whether the ARi terminates the ready/warning cue. Each of these three possibilities is then further distinguished by whether or not an escape response (ER) is required on those trials in which the em-S occurs. This latter case should include yet another distinction as to whether the ER is the same as the ARi. Although usually that is the case, it is operationally possible to require one response to avoid an em-S and a different response to terminate it should it occur. (For example, saying "I'm sorry" might avoid a fight, but only fighting might end a fight.) The final possibility is for the procedure to provide a distinctive exteroceptive stimulus contingent on the occurrence of an ARi. Some such feedback is inherent in those procedures in which the cue is terminated by the ARi. However, even in those procedures, avoidance conditioning may be affected by the addition of a discriminable safety signal. There are, of course, may parametric variations to these paradigms, and there is one procedural factor that deserves explicit mention. This is when the overall procedure includes occasional unavoidable USs mixed with the avoidance trials. In such situations, the cue is still a perfectly valid predictor of an impending US, but its reliability is decreased in proportion to the frequency of "free shocks." This is important because it calls attention to the relevance of the context in which avoidance conditioning occurs. No Escape R Escape R Trace Avcond Cue ___/\__________ ___/\__________ ___/\__________ ___/\__________ em-S___________/\__ _______________ __________/--\_ _______________ OR-> OR-> ARi _______________ _______/\______ _______________ ________/\____ ER _______________ _______________ ____________/\_ ______________ Delayed Avcond Cue ___/--------\__ ___/--------\__ ___/---------\_ ___/--------\__ em-S___________/\__ _______________ __________/--\_ ______________ OR-> OR-> ARi _______________ _______/\______ _______________ ________/\____ ER _______________ _______________ ____________/\_ ______________ ARi --> cue off Cue ___/--------\__ ___/----\______ ___/---------\_ ___/-----\____ em-S___________/\__ _______________ __________/--\_ ______________ OR-> OR-> ARi _______________ _______/\______ _______________ ________/\____ ER _______________ _______________ ____________/\_ ______________ Time______________> ______________> ______________> _____________> Note 1. When an ER is required, the ARi may or may not be the same response. Note 2. In all of the above paradigms, there may or may not be an exteroceptive feedback stimulus associated with the ERi. Figure showing the six basic paradigms for instrumental avoidance conditioning, where ARi is not reflexively elicited by the em-S. Noncued Avoidance. In the noncued procedures, the em-S event (usually shock) is scheduled to occur, usually on a regular basis such as once a minute. This is customarily referred to as the shock-shock interval, i.e., the rate at which the em-S would occur absent any avoidance responses. Because the avoidance response (ARo, e.g., barpress) is a freely available operant, it is operationally different from the previously described instrumental avoidance. In the noncued procedure, the ARo has schedule consequences at all times rather than just during discrete trials. In noncued avoidance, the occurrence of ARo postpones the em-S for some length of time. This is customarily referred to as the response-shock interval, i.e., duration of the safety time before another em-S occurs. (Note that the response-shock interval need not be the same as the shock-shock interval.) Accordingly, a sufficiently high rate of ARo in operant avoidance would prevent the em-S from ever occurring. A cue can be added to the basic noncued procedure. Specifically, a warning signal can be presented shortly before the em-S if an ARo has not occurred. The difference between this and the previously described instrumental procedure is that, in instrumental avoidance, intertrial responses have no schedule consequences whereas, in operant avoidance, a sufficiently high rate of ARo would preclude the occurrence of the warning signal as well as the primary aversive event. Whether this operational difference is important behaviorally is an empirical question. Failure of ARo at a sufficiently high rate results in the occurrence of the em-S, which may be of fixed duration or an ER may be required. In the latter case, the ER is usually the same as the ARo, but it need not be. It is also possible to require several responses as either the ER or the ARo. Omission training. As noted previously, the generic paradigm (conting(R,-S) is indifferent as to the emotional significance of the S. However, only if the S is emotionally negative does the term "avoidance" seem appropriate. Hence, I have chosen to follow Sheffield and adopt the term "omission" when the S is emotionally positive. Potentially, the "omission paradox" is even more perplexing than the avoidance paradox. Intuitively, it makes sense to make a response that prevents an aversive event, but what could possibly reinforce a response that prevents a scheduled rewarding event? Perhaps the underlying processes are indeed independent of the emotional significance of the contingent event. An empirical systematist is frequently confronted with procedures that challenge the system-to-date and suggest the need for a major re-analysis. Such is the case with operant omission training. I can discribe the dilemma but I can not yet reconcile the conflict. The baseline paradigm is Pair(S,US) such as a keylight preceding presentation of the grain hopper to a pigeon. Operationally, I could have (and perhaps should have) treated that as a classcond procedure because it is the Pavlovian paradigm. However, rather than recording salivation as the CR, operant keypecking is observed. This was dubbed "autoshaping" and hence I categorized it as an interaction of classical and operant conditioning. Little did I then realize that the autoshaping paradigm would become the context for many insightful studies of associative learning. Now further to complicate matters, another contingency can be added to the autoshaping paradigm...food is not delivered if an antedating keypeck occurs. This has been called "negative automaintenance" but it best fits the present schema as operant omission training. This makes for symmetry of the avoidance/omission operations. Nevertheless, the category would not be interesting were it not for the autoshaping phenomenon. None of the omission procedures has enjoyed the range of systematic variations accorded the avoidance procedures. For example, little is known about the role of exteroceptive feedback in omission contexts. Accordingly, there remains a great deal of empirical research necessary to provide a basis for a comprehensive theory of avoidance conditioning.