- Email: [email protected]
On unpredictability as a causal factor in “learned helplessness”
LEARNING
AND
MOTIVATlON
14, 324-337 (1983)
On Unpredictability as a Causal Factor in “Learned Helplessness” J. BRUCE OVERMIER AND RICHARD M. WIELKIEWICZ University of Minnesota In two replications, two groups of dogs were exposed to a series of uncontrollable, electric shocks. For one group the shocks were preceded by a tone (i.e., Paired). For the other group the shocks were randomly related to the tones and hence unpredictable (i.e., Random). Each replication also included a third group; in the first it was exposed only to the series of tones (CS-only), while in the second, it was exposed only to a series of shocks (Shocks-only). Then, all dogs were required to learn a discriminative choice escape/avoidance task in which the required response was to lift the correct paw in the presence of each of two visual S’s to escape or avoid the shocks [(Sf - R,)/(St - R2)l. Dogs preexposed to random tones and shocks were least successful in learning the task relative to those groups which experienced either predicted shocks, only the tones, or only the shocks, which in turn did not differ from each other. These results permitted the inference that the proactive interference with choice behavior following random tone CSs and shocks was attributable to a learned irrelevance generalized with respect to CSs.
Psychologists have long studied various transfer and proactive interference phenomena to gain insight into learning processes (e.g., Underwood, 1957). In the animal literature, several such interference phenomena are well established. Some focused upon presentations of the CS alone (e.g., latent inhibition, Lubow, 1973), others studied combinations of CSs and USs (e.g., learned irrelevance, Baker & Mackintosh, 1979), while still others focused upon presentations of the US alone (e.g., US preexposure phenomenon, Randich & Lolordo, 1979). One form of proactive interference resulting from preexposure to USs has gained special attention. It is called “learned helplessness” and is of special interest because of the This research was supported in part by Grant BNS-7728161 from NSF to J.B.O. and by Grants HD-01136, HD-00098, HD-07151, and BNS 7503816 from NICHHD and NSF to the Center for Research in Human Learning of which R.M.W. was a postdoctoral fellow. The authors thank Susan Mineka, Steve Maier, and Nicholas Mackintosh for their keen comments. Requests for reprints should be sent to J. B. Overmier, Elliott Hall, Psychology Department, 75 East River Rd, University of Minnesota, Minneapolis, Minnesota 55455. 324 0023-%90/83 $3.00 Copyright All rights
0 1983 by Academic Press, Inc. of reproduction in any form reserved.
UNPREDICTABILITY
AND
LEARNED
HELPLESSNESS
325
generality of the interference across subsequently used new reinforcers presented in new situations (Altenor, Kay, & Richter, 1977; Caspy & Lubow, 1981; Williams & Maier, 1977). The learned helplessness phenomenon is observed when subjects exposed in one apparatus to uncontrollable and unpredictable electric shocks (or other noxious hedonic events) fail to learn a subsequent signaled escape/ avoidance learning task in a new apparatus. Overmier and Seligman (1967) provided an early demonstration of this interference effect, and there are now numerous replications and extensions of the phenomenon (see reviews by Maier & Seligman, 1976, and Overmier, Patterson, & Wielkiewicz, 1980). Three consequences of exposure to the uncontrollable and unpredicted shocks have been inferred (e.g., Seligman, Maier, & Solomon, 1971). Observations of general passivity and emotional unreactivity suggested that such exposure causes an emotional deficit. Failure to initiate responses in escape/avoidance tasks led to the hypothesis that prior exposure to uncontrollable shocks also produces a motivational deficit. Finally, failures to show increased probabilities of responding following a successful escape or avoidance (i.e., a reinforced response) have been the basis for inferring an associative deficit. “Learned helplessness” accounts of the three deficits comprising the proactive interference seen in these experiments have emphasized the causal role of a single factor: the uncontrollable nature of the preexposure to shock (Maier & Seligman, 1976). In fact, the function of the commonly employed “triadic design” (e.g., Seligman & Maier, 1967) is to provide an inferential basis for the assertion that uncontrollability of the pretreatment shocks causes the proactive interference syndrome. However, as pointed out by Overmier et al. (1980), most studies of the proactive interference effects of exposure to uncontrollable shock have confounded uncontrollability with unpredictability. That is, the shocks presented in the first “induction” phase are both uncontrollable and unpredictable. Therefore, the uncontrollable nature of the shocks is not necessarily the only relevant causal dimension. As an illustration, with respect to a physiological debilitation, Weiss (1971) has shown that length of gastric lesions caused by exposures to electric shocks is a joint function of both the controllability and predictability of the aversive shocks. The most gastric lesioning was caused by shocks that were both uncontrollable and unpredictable, and the least amount of lesioning was caused by exposure to shocks that were both controllable and predictable. Intermediate amounts of lesioning were produced by controllable/unpredictable and by uncontrollable/predictable electric shocks. (See also Imada & Soga, 1971, for closely related results.) To the extent that Weiss’s pattern of results may be generalized to behavioral debilitation, a clear implication is that degree of predictability may also influence the proactive interference syndrome produced by prior exposures to shocks. Indeed, it is possible
326
OVERMIER
AND
WIELKIEWICZ
that the different features of the syndrome are contributed differentially by the confounded induction factors of uncontrollability and unpredictability. Up to the present, strong tests of whether unpredictability contributes to the proactive interference seen in the learned helplessness phenomenon have not been conducted. Overmier and Seligman (1967, Experiment 3) did carry out a preliminary assessment of this by presenting subjects with uncontrollable shocks, half of which were predicted by a tone (and half of which were not!). They concluded that the proactive interference from exposure to inescapable shocks was not affected by the predictability of the shocks because the performances of the semisignaled groups did not differ from groups receiving no signals. This report may have discouraged further attempts to attribute a role of predictability in the learned helplessness proactive interference effect. Unfortunately, the experiment was a very weak test of a role for predictability because predicted and unpredicted shocks were intermixed; additionally the predictability factor was crossed with another variable (i.e., time course) and diluted by most animals not showing any interference because of this latter variable. Recently, in some exploratory work on the learned helplessness phenomenon (Overmier et al., 1980, see Fig. 6) we again looked at the effect of predicting shocks upon the degree of proactive interference. It was observed that preexposure to predicted shocks caused less proactive interference than preexposure to uncorrelated presentations of signals and shocks. As a consequence of this and review of other evidence, Overmier et al. hypothesized that the degree of proactive interference is a consequence of both the unpredictability and uncontrollability of the preshock. Furthermore, they suggested that uncontrollability of the aversive event produces the motivational, response initiation deficit, while the unpredictable nature of the aversive event contributes to and may even be responsible for the associative component of the proactive interference. What is needed, then, is a clear demonstration of a role for unpredictability in contributing to a general proactive interference. Furthermore, if it is assumed that associative learning is most sensitive to the predictability dimension, then the optimal procedure for seeing such interference would be to use a task that is especially sensitive to associative learning. One approach is suggested by the work of Irwin, Suissa, and Anisman (1980) who tested mice in a T-maze water escape task following exposures to inescapable shocks. The T-maze choice task, they argued, allowed them to separate response initiation deficit (stem escape times) from learning deficit (percentage correct). While they found significant impairment of escape behavior, they concluded that there was no effect upon choice behavior, although we note that under conditions where dzfliirentiul escape deficits were precluded (i.e., 0 delay procedures) the mice subjected to inescapable prior shocks made more choice errors. Jackson, Alexander,
UNPREDICTABILITY
AND
LEARNED
HELPLESSNESS
327
and Maier (1980) followed a similar rationale for test task selection and used a Y maze in a choice escape test task. A series of experiments suggested that exposure to uncontrollable (and unpredicted) shocks caused a learning deficit which was not related to response initiations or to vigor of behavior. This they interpreted as evidence for an associative deficit which is in addition to any motivational deficits which might also occur. They concluded that a choice task is particularly sensitive to associative interference while the “traditional” shuttlebox test is particularly sensitive to motivational deficits because locomotion is strongly related to motivation and activity while choice behavior is much less affected by such variables (viz., Pubols, 1960). Jackson et al. inferred that the subject learns that response and outcome are independent during the preexposure to shock (i.e., that shocks are uncontrollable) and that the associative interference is the result of “reduced sensitivity” to relations between responses and shock termination which, in turn, results in the lower percentage of correct responding in the Y-maze choice task. This contrasts with our hypothesis that associative deficits are caused, at least in part, by the unpredictable nature of the inescapable shock. The purpose of the present research was to explore this issue further by studying the consequences of prior exposure to predicted versus unpredicted shocks that are also uncontrollable. We chose as our dependent variable learning in a discriminative conditional choice escape/avoidance task which would allow assessment of proactive associative interference caused by preexposure to electric shocks. Our question was whether predictability of the uncontrollable shocks would influence the degree of proactive interference. METHOD
Subjects The subjects were 18 adult mongrel dogs ranging from 32 to 60 cm tall at the shoulder, obtained from the University of Minnesota Animal Hospital. They were housed in individual cages with free access to food and water throughout the experiment. Design In each replication, nine dogs were unsystematically assigned to one of three groups of equal size. These groups differed in their treatment only during the induction phase. The groups were (1) a group that received uncontrollable electric shocks, the occurrences of which were fully predicted by auditory warning signals (Paired), (2) a group that received the same uncontrollable shocks and auditory signals but uncorrelated with one another (Random), (3) a control group that in replication 1 received no shocks but did receive the signals (CS-only), and in replication 2 received the same shocks but received no signals (Shocks-only). The
328
OVERMIER
AND
WIELKIEWICZ
three groups of replication 1 parallel those of the triadic design used to identify uncontrollability as a causal variable, except that here the effort is to identify whether or not degree of predictability is a causal factor in the proactive interference with escape/avoidance learning. Apparatus
Different apparatus units were used in the first (induction) and second (test) phases of this experiment. Inescapable shocks were presented while the dog was suspended in a rubberized cloth hammock which hung from a metal frame. The dog’s legs hung below the body through four holes in the hammock and were secured with sponge-padded ropes tied to the metal frame. Shock was delivered via pairs of electrodes fastened to each of the subject’s rear legs. Each electrode pair consisted of two stainless-steel disks coated with electrode paste and mounted approximately 6 cm apart in a single piece of Plexiglas. The shock source for both pairs of electrodes was a 600 Vat variable transformer, with current applied through 50 kohm of resistance. The shock level across the subject was 4.5 mA for the entire study. The tones employed in the induction phase were produced by a pair of Bud code oscillators at about 85 dB (SPL). The tone frequencies were impure with the predominant frequency of the low tone equal to about 270 Hz, and the predominant frequency of the high tone about 1950 Hz. The escape/avoidance test apparatus consisted of two wooden platforms mounted parallel to the floor on a steel frame. The subject’s rear legs passed through two holes in a harness of rubberized cloth and then to a platform where they were secured with sponge-padded ties. The dog’s front paws rested freely upon another platform which consisted of two hinged, frosted Plexiglas pedals, 19.7 x 28.1 cm, located next to each other and separated by a vertical black barrier, 28.1 x 16-30 cm (depending upon the height of the dog), which was mounted between the pedals. Black stripes oriented longitudinally were on the left pedal and stripes oriented transversely were on the right pedal. Each pedal had a 28-Vdc light mounted under it and was spring loaded so that a microswitch was closed when the dog’s paw was lifted upward off the pedal (i.e., flexion). The movement of the dog was restricted by a clear Plexiglas collar which surrounded the front and sides of its neck. A standard dog collar was around the dog’s neck and secured to the frame. A piece of rubberized cloth ran from one side of the Plexiglas collar, around the back of the dog’s neck, to the other side of the Plexiglas collar. Finally, the distance from the collar to the pedals could be adjusted for each subject. In this configuration, the animal could freely move either front leg but could not exit from the apparatus. Signal lights were mounted behind two fixed 21.5 x 17.3-cm Plexiglas panels located 0.5 m in front of the subject’s head and 0.5 m on either side. The left side panel was longitudinally striped and the right side panel was transversely striped.
UNPREDICTABILITY
AND LEARNED
HELPLESSNESS
329
The induction hammock and the choice apparatus were each located inside a large, white, sound-attenuating cubicle with a shielded houselight directly above the animal. White noise was continuously presented in the cubicle at 75 dB (SPL). Control and recording equipment were located outside the cubicle. Procedures Adaptation. Two 50-min sessions of adaptation to the escape/avoidance test apparatus were conducted with both the white noise and houselight on. At the beginning of the first session of adaptation, the dog was placed in the apparatus which was then adjusted to fit the animal. The animal was then left alone; at the end of 20 min the animal’s rear legs were shaved to assure good electrode contact, the apparatus was readjusted if necessary, and the two electrode pairs were taped to the dog’s rear legs. The subject was then left alone for about 30 additional minutes. For second session of adaptation the dog was secured in the apparatus with electrodes attached for 50 min. Occasionally, the apparatus was readjusted when observations indicated that the dog was uncomfortable. Induction. In this preexposure phase, all subjects rested in the full hammock. Each session began with the onset of the houselight. Subjects in the Paired-shocks group and Random-shocks group were exposed to a series of electric shocks, each of 4-set duration presented at varying intervals averaging 2 min and ranging from 1 to 3 min. A random half of the shocks were delivered to the left rear leg and the remaining shocks were delivered to the right rear leg. Shocks delivered to the right leg were pulsed ten times per second while those to the left leg were unpulsed. (Two different kinds of shocks were employed because this facilitates response acquisition in our choice test task. See Overmier, Bull, & Trapold, 1971.) For subjects in a Paired group, each shock was preceded by a 5-set auditory signal which then continued until shock termination, For subjects in replication 1, shocks to the left rear leg were preceded by the high tone and shocks to the right rear leg were preceded by the low tone. For subjects in replication 2, high tones preceded shocks to the right leg and low tones preceded the shocks to the other leg. Each subject in the Random group was matched with a subject in the Paired group and thus experienced the same number of shocks in the same temporal order and with the same temporal spacing as the partner subject in the Paired group. However, for all dogs in the Random-shocks group, the high and low tones were presented on a random schedule, independent of the schedule for shocks. That is, tones and shocks were uncorrelated in this group, and the shocks were unpredictable both as to time of occurrence and place of delivery.
330
OVERMIER
AND
WIELKIEWICZ
Each subject in the CS-only group of replication 1 was exposed to the same series of high and low tones experienced by a partner subject in the Paired group (replication 1). However, subjects in the CS-only group did not experience any electric shocks in the induction phase. Each subject in the Shocks-only group of replication 2 was exposed to the same series of shocks experienced by a partner subject in the Paired group (replication 2). However, the subjects in the Shocks-only group did not experience any tones in the induction phase. The experiment was run in two systematic replications (Sidman, 1960). The replications differed in the number of shocks per induction session (40 versus 20) and the number of sessions of preexposure to shock (6 versus 12). There were no other differences between replications. Testing. The day following the last preexposure session was the first of two daily sessions of discriminative choice escape/avoidance training. In this testing phase, the subjects were presented with the following problem: Each 15set trial began with the onset of a 10 Hz flashing light (equal on-off cycle) located directly under the left front paw (Sy) or right front paw (SF). If (Sp) was presented, the animal was required to lift its left front paw within 5 set to avoid pulsed shock to its right rear leg. If (Sy) was presented, the animal was required to lift its right front paw within 5 set to avoid shock to its left rear leg. If during the 5-set CS-US interval the initial response of the animal was correct, then the trial was immediately terminated and an intertrial interval was initiated. If the initial response was wrong, a correction procedure allowed the animal to switch responses but now the criterion was lift and hold for the remainder of the trial. If the dog did not perform an initial correct avoidance within 5 set of the beginning of the trial, shock was scheduled for the final 10 set of the trial except during any period the correct response was being performed. During the final 10 set of each trial, the flashing SD light and the shock were turned off whenever the dog made the correct response. Each correct response during a trial was accompanied by extra feedback events of (1) a 0.5-set offset of the white noise and (2) onset of the side panel light. Correct responses did not include those times when the dog lifted both front paws. The interval between trials averaged 2 min and ranged from 1 to 3 min. Each test session was composed of ten (SF) and ten (Sy) trials. Thus, in this test of learning a discriminative choice escape/avoidance task, if a subject’s initial anticipatory response was correct it avoided shocks; if the response was not correct a correction procedure allowed the subject to find the correct response but now it had to maintain it so as to prevent delivery of scheduled aversive stimuli. The requirement or response persistence here for escape/avoidance is copied from Skipin’s and Vinnik’s modification of the method of Petropavlovskii as described in Skipin, Ivanova, Kozlovskaya, & Vinnik (1969).
UNPREDICTABILITY
AND LEARNED
HELPLESSNESS
331
RESULTS The main dependent variable was the amount of time per 15set trial during which the subject was not performing the correct response (wrong time). This was averaged across the three subjects in each group of each replication for each test day. Learning the task, then, was reflected by a decline in mean wrong time. These data are presented in Fig. 1 for all groups as a function of blocks of the 20 discriminative choice training trials of each test day. The data summarized in Fig. 1 were subjected to an analysis of variance in which the factors were Replications (1,2) x Induction Treatments (Paired, Random, Control) x Test Days (1,2) with repeated measures on dogs over the two test days. All statistical tests were conducted at the .05 level of confidence. It appears that all groups showed learning across days (i.e., decreasing mean wrong times), Days F(1, 12) = 12.61, and that the two replications produced substantially the same patterns of results. The ANOVA confirms the similarity of results across replications. Replications and all interactions with replications were not sign&cant, F’s < 1.3. The absence of differences between replications was additionally confirmed by follow-up comparisons on each treatment across replications, all three t’s < 1.20. In particular among these, note that the CS-only was not significantly different from Shocks-only, t < 0.01. The induction treatment conditions did, however, have different effects upon discriminative choice escape/avoidance performance, F(2, 12) = 7.02. The Random groups were significantly impaired relative to the Paired groups, t(12) = 3.40, and relative to the Control groups, t(12) = REPLICATION
1
REPLICATION
2
, I 1 CHOICE
PAIRED RANDOM I 2 RESPONSE
0 .
CS ONLY 4 SHK ONLYA I I I 1 2 ACQUISITION DAYS
FIG. 1. Daily mean wrong time (time during daily trials during which dogs failed to make the correct response) for each group on each of 2 days of test phase training on the discriminative choice escape/avoidance task.
332
OVERMIER
AND WIELKIEWICZ
2.79.’ However, the Paired groups and the Control groups did not differ reliably, t < 1. To ascertain whether the between treatments effects seen in mean wrong times (Random > Paired = Control) are not an artifact of differential response initiation, latencies to first responses, correct and incorrect, on each trial were analyzed as a function of Replications x Treatments x Test Days, paralleling the analysis of the primary dependent variable. Latencies on the second test day were shorter than on the first day, F(1, 12) = 8.05, reflecting learning in a second way. However, central to our concern here, latencies did not differ as a function of the induction treatments; treatments and all interactions involving treatments F’s were not significant. DISCUSSION The pattern of empirical effects in the present experiment is simple and straightforward: Prior exposure to tones and shocks presented on random and independent schedules interferes with subsequent learning to make correct choice response to visual SDs in an escape/avoidance choice task. This interference is ameliorated by having the tones predict the shocks. These results confirm the earlier preliminary observation (Overmier et al., 1980) that preexposure to unpredictable shocks caused more proactive interference than preexposure to predictable shocks. Compared to the CS-only (no shocks) reference level, only preexposure to inescapable shocks and uncorrelated signals produced proactive interference. That prior exposure to predictable, inescapable shocks had no interfering effect is at first glance surprising from the traditional “learned helplessness” perspective. But recall that the primary deficit observed in the traditional shuttlebox task is one of reduced initiations of the effortful response of jumping over a barrier. Here, we purposefully chose a less effortful task and a measure based upon discriminative choice behavior with the intent of detecting associative interference rather than initiation interference. Indeed, all our animals made several responses on nearly every test trial. Thus, our data primarily reflect the facility with which the dogs learned the (Sf - R,)/(Sf - R2) discriminative choice problem. The latency data confirm that we were successful in designing a learning task in which the criterion response was sufficiently easy not to be influenced by the response initiation deficit obtained in more effortful tasks. (See Glazer & Weiss, 1976, for another demonstration that the initiation of appropriately selected instrumental acts may be unimpaired by exposures to uncontrollable shocks which commonly produce learned ’ Although these between treatment contrasts were dictated a priori by our hypothesis. some will be comforted to know that these contrasts are confirmed using Newman-Keuls procedures, q, = 4.77 and q2 = 3.90.
UNPREDICTABILITY
AND
LEARNED
HELPLESSNESS
333
helplessness.) Hence, the impaired learning of our conditional choice task by the Random groups reflected in the mean wrong times is attributable to differential choosing behavior resulting from pretreatment with the independent presentations of CS and US. While the present results are unique within the “learned helplessness” literature, they may be related to other phenomena documented in the Pavlovian conditioning literature. Mackintosh (1973) and Baker (1976) have published demonstrations of a phenomenon that has been called ‘ ‘learned irrelevance. ” In these experiments, four groups of rats were preexposed to either uncorrelated presentations of CS (tone or noise) and shock, shock-alone, CS-alone, or no stimulus. The subjects were then tested for acquisition of response suppression when the same CS now signaled electric shocks and was presented while the animal was responding freely for appetitive reinforcers (water and food). During this test, the group preexposed to uncorrelated CS shock presentations was slowest in acquiring conditioned suppression. The attentional account of these results offered was that those rats exposed to the uncorrelated presentations of CS and shocks learned that these two particular events were irrelevant to each other. Mackintosh (1973) further demonstrated that this irrelevance learning was specific to the US used in the preexposure phase. In contrast to Mackintosh’s (1973) and Baker’s (1976) experiments in which exposure to uncorrelated presentations of a stimulus and shock interfered with later acquisition of conditioned suppression in the presence of the same cue stimulus, the present finding was that uncorrelated presentations of one CS and shocks interfered with acquisition of shockmotivated responding to distinctly different cue stimuli. Thus, the present finding is of a generalized learned irrelevance effect which transcends the CS conditions of the original preexposure to shock. Since tones were employed as the CSs in the first phase of the present experiment and lights were employed as the discriminative stimuli in the test phase, it is not likely that simple stimulus generalization could account for these effects. Thus, while Mackintosh showed that the learned irrelevance was specific with respect to the US, our results suggest that there is a general associative interference with respect to CSs. Related experiments by Wickens, Tuber, Nield, and Wickens (1977) previously hinted at the existence of a generalized irrelevance effect. Wickens et al. tested the effect of prior exposure to a random and uncorrelated CS and US upon later Pavlovian conditioning employing either the same CS and US combination or a different CS and different US. These they compared (across experiments) to a group preexposed to pairings of a CS and US and later Pavlovianly conditioned with a different CS and different US. They found that these three groups in the second Pavlovian conditioning phase took significantly different numbers
334
OVERMIER
AND
WIELKIEWICZ
of sessions to reach criterion. The group preexposed to random, independent presentations of the CS and US later to be used in the Pavlovian pairing was the slowest to learn; the group preexposed to random, independent presentations of a CS and US, but ones different from those paired in the Pavlovian phase was the fastest. This pattern is consistent with the assumption that some proportion of the learned irrelevance effect is specific to the particular CS and US presented in the preexposure phase, but also that there is a generalized learned irrelevance contribution as well. Thus, the Wickens et al. (1977) experiments provide a Pavlovian conditioning analog to the present study. In both cases, prior exposure to random signals and shock caused significantly slower acquisition than prior exposure to paired signals and shock-an effect which transcended the particular CS used during the preexposure phase. Let us also consider two other potential accounts of our results: (a) blocking and (b) reduced effective reinforcer. These two accounts have in common that they rely upon the empirical phenomenon that exposing an organism to a US has the effect of slowing subsequent conditioning when that same US is now signaled by a CS (Kamin, 1961; see Randich & LoLordo, 1979, for review). They differ in the theoretical mechanism invoked. According to the blocking account, conditioning to any CS in the test phase is blocked by the excitatory strength already conditioned to the context (or background) stimuli during the US preexposure phase (e.g., Rescorla & Wagner, 1972). Since the US is already well predicted by the context, the CS cannot acquire any additional excitatory strength. On this account, however, the contextual stimuli should not become so excitatory during the preexposure phase if the US presentations were signaled by some CS. The signaling would “protect” the context from becoming excitatory so that later the context would not block the association of the US with some new CS. Thus, the blocking account would predict the pattern of results we obtained in replication 1: Random-shocks groups were slower to learn than the Paired-shocks group and the CS-only group, which in turn might not differ. The effective reinforcer account (McAllister, McAllister, & Douglass, 1971) suggests that the interference with learning in our avoidance test would obtain because prior exposures to unsignaled USs would result in conditioning of fear to the background (or the intertrial interval). Thus, in the avoidance training phase, when the dogs performed successful escape or avoidance responses during the discriminative stimuli, they would be returned to the background stimulus conditions which would continue to elicit fear, hence minimizing any potential reinforcement for the correct response. Such disruption of the reinforcement process hypothesized by two-process theory would not occur if the background stimuli were less fear eliciting-as would be the case when there had
UNPREDICTABILITY
AND
LEARNED
HELPLESSNESS
335
been no prior US exposures (as in our CS-only group) or when any US preexposures had been signaled (as in our Paired-shocks group). This means that the effective reinforcement theory might also account for the pattern of results obtained in replication 1. Replication 2 was included to provide a test for these blocking and reduced reinforcer effectiveness accounts. Replication 2 compared learning by groups exposed to Paired, Random, and Shock-only groups. It is important to note that neither of these accounts would predict differential outcomes between our Random group (which received random CSs and independent shocks) and our Shock-only group, while a general learned irrelevance account would predict impairment only in the former. Moreover, a learned irrelevance account would suggest that our CS-only group (replication 1) and our Shocks-only group (replication 2) should acquire equally well in our discriminative choice avoidance test task (see Mackintosh, 1973). Note that the test performance of the Shocks-only group was almost identical to that of the CS-only group and that both were superior to the Random group. This pattern contravenes both the blocking and the effective reinforcer accounts of our results but is consistent with a generalized learned irrelevance account. At the present time, our understanding of these phenomena remains tentative. Our data show that preexposure to Random CSs and shocks interferes with condkional choice escape/avoidance learning more than preexposure to shocks-only, CS-only, or predicted-shocks. This effect is apparently independent of the response initiation deficit of the learned helplessness phenomenon seen with motorintensive non-choice tasks. It does parallel the learned irrelevance phenomenon. Our observation extends those instances of learned irrelevance reported by Mackintosh (1973) and by Tomie, Murphy, Fath, and Jackson (1980), however, because the present results demonstrate the general nature of the effects in transcending CSs, so long as the US is held constant over phases. One potentially problematic feature of our data is the absence of difference between our two control groups, CS-only and US-only. While this absence of difference allows us to identify the generalized learned irrelevance phenomenon, it raises a question as to why the US exposures to the US-only control do not cause an impairment in associative learning; after all these USs were unpredictable as well. Moreover, it does not allow us to choose between the report of Irwin et al. (1980) and that of Jackson et al. (1980) because under one unpredictable US condition, Random CSs and USs, we found impairment while under another unpredictable US condition, US-only, we did not. The latter suggests that associative interferences heretofore reported may have reflected primarily some response initiation deficit as Alloy and Seligman (1979) have suggested. That is not the case here. Alternatively, there may be more than one kind of associative proactive interference the causal factors of which may differ.
OVERMIER
336
AND WIELKIEWICZ
A goodly amount of additional research is needed before a clear picture of proactive inference phenomena will emerge. We need more integrative research which brings together proactive interference phenomena such as generalized learned helplessness and generalized learned irrelevance. Baker (1976) provides an example of one such tack. The present data provide another and are consistent with the suggestion that the proactive interference phenomenon seen in the multifaceted learned helplessness syndrome may in fact be multidimensional in terms of its causal factors as well as in terms of its consequent features. Indeed, it has always seemed to us that one weakness of the “learned helplessness” theory was its reliance upon one single experimental factor and its resultant cognitive state as a single causal entity of three effects (emotional, incentive motivational, and associative). This feature of the theory requires that whenever the three are independently assessed, they ought to show substantial covariation. Unfortunately this has not always been so. The present experiment provides evidence that the learned helplessness syndrome may well be the product of multiple interacting causal factors which are typically confounded but may under appropriate experimental conditions be isolated for study. One of these is generalized learned irrelevance. REFERENCES Alloy, L. B., & Seligman, M. E. P. On the cognitive component of learned helplessness and depression. In G. Bower (Ed.). The psychology of learning and motivation. New York: Academic Press, 1979. Vol. 13, pp. 219-276. Altenor, A., Kay, E., & Richter, M. The generality of learned helplessness in the rat. Learning
and Motivation,
1977, 8, 54-62.
Baker, A. G. Learned irrelevance and learned helplessness: Rats learn that stimuli, reinforcers, and responses are uncorrelated. Journal of Experimental Psychology: Animal Behavior Processes, 1976, 2, 130-141. Baker, A. G., & Mackintosh, N. J. Preexposure to the CS alone, US alone, or CS and US uncorrelated: Latent inhibition, blocking by context, or learned irrelevance? Learning and Motivation, 1979, 10, 278-294. Caspy, T., & Lubow, R. E. Generality of US preexposure effects: Transfer from food to shock or shock to food with and without same response requirements. Animal Learning & Behavior, 1981, 9, 524-532. Glazer, H. I., & Weiss, J. M. Long-term interference effect: An alternative to “learned helplessness.” Journal of Experimental Psychology: Animal Behavior Processes, 1976, 2, 202-213. Imada, H., & Soga, M. The CER and BEL as a function of predictability and escapability of an electric shock. Japanese Psychological Research, 1971, 13, 115-122. Irwin, J., Suissa, A., & Anisman, H. Differential effects of inescapable shock on escape performance and discrimination learning in a water escape tank. Journal of Experimental Psychology: Animal Behavior Processes, 1980, 6, 21-40. Jackson, R. L., Alexander, J. H., & Maier, S. F. Learned helplessness, inactivity, and associative deficits: Effects of inescapable shock on response choice escape learning. Journal of Experimental Psychology: Animal Behavior Processes, 1980, 6, l-20. Kamin, L. J. Apparent adaptation effects in the acquisition of a conditioned emotional response. Canadian Journal of Psychology, 1961, 15, 176-188. Lubow, R. E. Latent inhibition. Psychological Bulletin; 1973, 79, 398-407.
UNPREDICTABILITY
AND LEARNED
HELPLESSNESS
337
Mackintosh, N. J. Stimulus selection: Learning to ignore stimuli that predict no change in reinforcement. In R. A. Hinde &J. Stevenson-Hinde (Eds.), Construints on learning. London/New York: Academic Press, 1973. Pp. 75-96. Maier, S. F., & Seligman, M. E. P. Learned helplessness: Theory and evidence. Journal of Experimental Psychology: General, 1976, 105, 3-46. McAllister, W. R., McAllister, D. E., & Douglass, W. K. The inverse relationship between shock intensity and shuttlebox avoidance learning in rats: A reinforcement explanation. Journal
of Comparative
and Physiological
Psychology,
1971, 74, 426-433.
Overmier, J. B., Bull, J. A., III, & Trapold, M. A. Discriminative cue properties of different fears and their role in response selection in dogs. Journal of Comparative and Physiological
Psychology,
1971. 76, 478-482.
Overmier, J. B., Patterson, J., & Wielkiewicz, R. M. Environmental contingencies as sources of stress in animals. In S. Levine & H. Ursin (Eds.), Coping and he&h. New York: Plenum, 1980. Pp. l-38. Overmier, J. B., & Seligman, M. E. P. Effects of inescapable shock upon subsequent escape and avoidance responding. Journal of Comparative and Physiological, 1967. 63, 28-33. Pubols, B. H., Jr. Incentive magnitude, learning, and performance in animals. Psychological Bulletin, 1960, 57, 89-115. Randich, A., & LoLordo, V. M. Associative and nonassociative theories of the UCS preexposure phenomenon: Implications for Pavlovian conditioning. Psychological Bulletin, 1979, 86, 523-548. Rescorla, R. A., & Wagner, A. R. A theory of Pavlovian conditioning. In A. H. Black & W. F. Prokasy (Eds.), Classical conditioning II: Current research and theory. New York: Appleton-Century-Crofts, 1972. Pp. 64-99. Seligman, M. E. P., & Maier, S. F. Failure to escape traumatic shock. Journal of Experimental Psychology, 1967, 74, l-9. Seligman, M. E. P., Maier, S. F., & Solomon, R. L. Unpredictable and uncontrollable aversive events. In F. R. Brush (Ed.), Aversive conditioning and learning. New York: Academic Press, 1971. Sidman, M. Tactics of scient$c research. New York: Basic Books, 1960. Skipin, G. V., Ivanova, I. G., Kozlovskaya, I. B., & Vinnik, R. L. Defense conditioning and avoidance. In M. Cole and I. Maltzman (Eds.), Handbook of contemporary Societ psychology. New York: Basic Books, 1969, Pp. 785-810. Tomie, A., Murphy, A. L., Fath, S., &Jackson, R. L. Retardation of autoshaping following pretraining with unpredictable food: Effects of changing context between pretraining and testing. Learning and Motivarion, 1980, 11, 117-134. Underwood, B. J. Interference and forgetting. Psychological Review, 1957, 64, 49-60. Weiss, J. M. Effects of coping behavior in different warning signal conditions on stress pathology in rats. Journal of Comparative and Physiological Psychology, 1971, 77, l-13. Wickens, D. D., Tuber, D. S., Nield, A. F., & Wickens, C. Memory for the conditioned response: The effects of potential interference introduced before and after original conditioning. Journal of Experimental Psychology: General, 1977, 106, 47-70. Williams, J. L., & Maier, S. F. Transsituational immunization and therapy of learned helplessness in the tat. Journal of Experimental Psychology: Animal Behavior Processes, 1977, 3, 240-253. Received May 12, 1983 Revised September 13, 1983
AND
MOTIVATlON
14, 324-337 (1983)
On Unpredictability as a Causal Factor in “Learned Helplessness” J. BRUCE OVERMIER AND RICHARD M. WIELKIEWICZ University of Minnesota In two replications, two groups of dogs were exposed to a series of uncontrollable, electric shocks. For one group the shocks were preceded by a tone (i.e., Paired). For the other group the shocks were randomly related to the tones and hence unpredictable (i.e., Random). Each replication also included a third group; in the first it was exposed only to the series of tones (CS-only), while in the second, it was exposed only to a series of shocks (Shocks-only). Then, all dogs were required to learn a discriminative choice escape/avoidance task in which the required response was to lift the correct paw in the presence of each of two visual S’s to escape or avoid the shocks [(Sf - R,)/(St - R2)l. Dogs preexposed to random tones and shocks were least successful in learning the task relative to those groups which experienced either predicted shocks, only the tones, or only the shocks, which in turn did not differ from each other. These results permitted the inference that the proactive interference with choice behavior following random tone CSs and shocks was attributable to a learned irrelevance generalized with respect to CSs.
Psychologists have long studied various transfer and proactive interference phenomena to gain insight into learning processes (e.g., Underwood, 1957). In the animal literature, several such interference phenomena are well established. Some focused upon presentations of the CS alone (e.g., latent inhibition, Lubow, 1973), others studied combinations of CSs and USs (e.g., learned irrelevance, Baker & Mackintosh, 1979), while still others focused upon presentations of the US alone (e.g., US preexposure phenomenon, Randich & Lolordo, 1979). One form of proactive interference resulting from preexposure to USs has gained special attention. It is called “learned helplessness” and is of special interest because of the This research was supported in part by Grant BNS-7728161 from NSF to J.B.O. and by Grants HD-01136, HD-00098, HD-07151, and BNS 7503816 from NICHHD and NSF to the Center for Research in Human Learning of which R.M.W. was a postdoctoral fellow. The authors thank Susan Mineka, Steve Maier, and Nicholas Mackintosh for their keen comments. Requests for reprints should be sent to J. B. Overmier, Elliott Hall, Psychology Department, 75 East River Rd, University of Minnesota, Minneapolis, Minnesota 55455. 324 0023-%90/83 $3.00 Copyright All rights
0 1983 by Academic Press, Inc. of reproduction in any form reserved.
UNPREDICTABILITY
AND
LEARNED
HELPLESSNESS
325
generality of the interference across subsequently used new reinforcers presented in new situations (Altenor, Kay, & Richter, 1977; Caspy & Lubow, 1981; Williams & Maier, 1977). The learned helplessness phenomenon is observed when subjects exposed in one apparatus to uncontrollable and unpredictable electric shocks (or other noxious hedonic events) fail to learn a subsequent signaled escape/ avoidance learning task in a new apparatus. Overmier and Seligman (1967) provided an early demonstration of this interference effect, and there are now numerous replications and extensions of the phenomenon (see reviews by Maier & Seligman, 1976, and Overmier, Patterson, & Wielkiewicz, 1980). Three consequences of exposure to the uncontrollable and unpredicted shocks have been inferred (e.g., Seligman, Maier, & Solomon, 1971). Observations of general passivity and emotional unreactivity suggested that such exposure causes an emotional deficit. Failure to initiate responses in escape/avoidance tasks led to the hypothesis that prior exposure to uncontrollable shocks also produces a motivational deficit. Finally, failures to show increased probabilities of responding following a successful escape or avoidance (i.e., a reinforced response) have been the basis for inferring an associative deficit. “Learned helplessness” accounts of the three deficits comprising the proactive interference seen in these experiments have emphasized the causal role of a single factor: the uncontrollable nature of the preexposure to shock (Maier & Seligman, 1976). In fact, the function of the commonly employed “triadic design” (e.g., Seligman & Maier, 1967) is to provide an inferential basis for the assertion that uncontrollability of the pretreatment shocks causes the proactive interference syndrome. However, as pointed out by Overmier et al. (1980), most studies of the proactive interference effects of exposure to uncontrollable shock have confounded uncontrollability with unpredictability. That is, the shocks presented in the first “induction” phase are both uncontrollable and unpredictable. Therefore, the uncontrollable nature of the shocks is not necessarily the only relevant causal dimension. As an illustration, with respect to a physiological debilitation, Weiss (1971) has shown that length of gastric lesions caused by exposures to electric shocks is a joint function of both the controllability and predictability of the aversive shocks. The most gastric lesioning was caused by shocks that were both uncontrollable and unpredictable, and the least amount of lesioning was caused by exposure to shocks that were both controllable and predictable. Intermediate amounts of lesioning were produced by controllable/unpredictable and by uncontrollable/predictable electric shocks. (See also Imada & Soga, 1971, for closely related results.) To the extent that Weiss’s pattern of results may be generalized to behavioral debilitation, a clear implication is that degree of predictability may also influence the proactive interference syndrome produced by prior exposures to shocks. Indeed, it is possible
326
OVERMIER
AND
WIELKIEWICZ
that the different features of the syndrome are contributed differentially by the confounded induction factors of uncontrollability and unpredictability. Up to the present, strong tests of whether unpredictability contributes to the proactive interference seen in the learned helplessness phenomenon have not been conducted. Overmier and Seligman (1967, Experiment 3) did carry out a preliminary assessment of this by presenting subjects with uncontrollable shocks, half of which were predicted by a tone (and half of which were not!). They concluded that the proactive interference from exposure to inescapable shocks was not affected by the predictability of the shocks because the performances of the semisignaled groups did not differ from groups receiving no signals. This report may have discouraged further attempts to attribute a role of predictability in the learned helplessness proactive interference effect. Unfortunately, the experiment was a very weak test of a role for predictability because predicted and unpredicted shocks were intermixed; additionally the predictability factor was crossed with another variable (i.e., time course) and diluted by most animals not showing any interference because of this latter variable. Recently, in some exploratory work on the learned helplessness phenomenon (Overmier et al., 1980, see Fig. 6) we again looked at the effect of predicting shocks upon the degree of proactive interference. It was observed that preexposure to predicted shocks caused less proactive interference than preexposure to uncorrelated presentations of signals and shocks. As a consequence of this and review of other evidence, Overmier et al. hypothesized that the degree of proactive interference is a consequence of both the unpredictability and uncontrollability of the preshock. Furthermore, they suggested that uncontrollability of the aversive event produces the motivational, response initiation deficit, while the unpredictable nature of the aversive event contributes to and may even be responsible for the associative component of the proactive interference. What is needed, then, is a clear demonstration of a role for unpredictability in contributing to a general proactive interference. Furthermore, if it is assumed that associative learning is most sensitive to the predictability dimension, then the optimal procedure for seeing such interference would be to use a task that is especially sensitive to associative learning. One approach is suggested by the work of Irwin, Suissa, and Anisman (1980) who tested mice in a T-maze water escape task following exposures to inescapable shocks. The T-maze choice task, they argued, allowed them to separate response initiation deficit (stem escape times) from learning deficit (percentage correct). While they found significant impairment of escape behavior, they concluded that there was no effect upon choice behavior, although we note that under conditions where dzfliirentiul escape deficits were precluded (i.e., 0 delay procedures) the mice subjected to inescapable prior shocks made more choice errors. Jackson, Alexander,
UNPREDICTABILITY
AND
LEARNED
HELPLESSNESS
327
and Maier (1980) followed a similar rationale for test task selection and used a Y maze in a choice escape test task. A series of experiments suggested that exposure to uncontrollable (and unpredicted) shocks caused a learning deficit which was not related to response initiations or to vigor of behavior. This they interpreted as evidence for an associative deficit which is in addition to any motivational deficits which might also occur. They concluded that a choice task is particularly sensitive to associative interference while the “traditional” shuttlebox test is particularly sensitive to motivational deficits because locomotion is strongly related to motivation and activity while choice behavior is much less affected by such variables (viz., Pubols, 1960). Jackson et al. inferred that the subject learns that response and outcome are independent during the preexposure to shock (i.e., that shocks are uncontrollable) and that the associative interference is the result of “reduced sensitivity” to relations between responses and shock termination which, in turn, results in the lower percentage of correct responding in the Y-maze choice task. This contrasts with our hypothesis that associative deficits are caused, at least in part, by the unpredictable nature of the inescapable shock. The purpose of the present research was to explore this issue further by studying the consequences of prior exposure to predicted versus unpredicted shocks that are also uncontrollable. We chose as our dependent variable learning in a discriminative conditional choice escape/avoidance task which would allow assessment of proactive associative interference caused by preexposure to electric shocks. Our question was whether predictability of the uncontrollable shocks would influence the degree of proactive interference. METHOD
Subjects The subjects were 18 adult mongrel dogs ranging from 32 to 60 cm tall at the shoulder, obtained from the University of Minnesota Animal Hospital. They were housed in individual cages with free access to food and water throughout the experiment. Design In each replication, nine dogs were unsystematically assigned to one of three groups of equal size. These groups differed in their treatment only during the induction phase. The groups were (1) a group that received uncontrollable electric shocks, the occurrences of which were fully predicted by auditory warning signals (Paired), (2) a group that received the same uncontrollable shocks and auditory signals but uncorrelated with one another (Random), (3) a control group that in replication 1 received no shocks but did receive the signals (CS-only), and in replication 2 received the same shocks but received no signals (Shocks-only). The
328
OVERMIER
AND
WIELKIEWICZ
three groups of replication 1 parallel those of the triadic design used to identify uncontrollability as a causal variable, except that here the effort is to identify whether or not degree of predictability is a causal factor in the proactive interference with escape/avoidance learning. Apparatus
Different apparatus units were used in the first (induction) and second (test) phases of this experiment. Inescapable shocks were presented while the dog was suspended in a rubberized cloth hammock which hung from a metal frame. The dog’s legs hung below the body through four holes in the hammock and were secured with sponge-padded ropes tied to the metal frame. Shock was delivered via pairs of electrodes fastened to each of the subject’s rear legs. Each electrode pair consisted of two stainless-steel disks coated with electrode paste and mounted approximately 6 cm apart in a single piece of Plexiglas. The shock source for both pairs of electrodes was a 600 Vat variable transformer, with current applied through 50 kohm of resistance. The shock level across the subject was 4.5 mA for the entire study. The tones employed in the induction phase were produced by a pair of Bud code oscillators at about 85 dB (SPL). The tone frequencies were impure with the predominant frequency of the low tone equal to about 270 Hz, and the predominant frequency of the high tone about 1950 Hz. The escape/avoidance test apparatus consisted of two wooden platforms mounted parallel to the floor on a steel frame. The subject’s rear legs passed through two holes in a harness of rubberized cloth and then to a platform where they were secured with sponge-padded ties. The dog’s front paws rested freely upon another platform which consisted of two hinged, frosted Plexiglas pedals, 19.7 x 28.1 cm, located next to each other and separated by a vertical black barrier, 28.1 x 16-30 cm (depending upon the height of the dog), which was mounted between the pedals. Black stripes oriented longitudinally were on the left pedal and stripes oriented transversely were on the right pedal. Each pedal had a 28-Vdc light mounted under it and was spring loaded so that a microswitch was closed when the dog’s paw was lifted upward off the pedal (i.e., flexion). The movement of the dog was restricted by a clear Plexiglas collar which surrounded the front and sides of its neck. A standard dog collar was around the dog’s neck and secured to the frame. A piece of rubberized cloth ran from one side of the Plexiglas collar, around the back of the dog’s neck, to the other side of the Plexiglas collar. Finally, the distance from the collar to the pedals could be adjusted for each subject. In this configuration, the animal could freely move either front leg but could not exit from the apparatus. Signal lights were mounted behind two fixed 21.5 x 17.3-cm Plexiglas panels located 0.5 m in front of the subject’s head and 0.5 m on either side. The left side panel was longitudinally striped and the right side panel was transversely striped.
UNPREDICTABILITY
AND LEARNED
HELPLESSNESS
329
The induction hammock and the choice apparatus were each located inside a large, white, sound-attenuating cubicle with a shielded houselight directly above the animal. White noise was continuously presented in the cubicle at 75 dB (SPL). Control and recording equipment were located outside the cubicle. Procedures Adaptation. Two 50-min sessions of adaptation to the escape/avoidance test apparatus were conducted with both the white noise and houselight on. At the beginning of the first session of adaptation, the dog was placed in the apparatus which was then adjusted to fit the animal. The animal was then left alone; at the end of 20 min the animal’s rear legs were shaved to assure good electrode contact, the apparatus was readjusted if necessary, and the two electrode pairs were taped to the dog’s rear legs. The subject was then left alone for about 30 additional minutes. For second session of adaptation the dog was secured in the apparatus with electrodes attached for 50 min. Occasionally, the apparatus was readjusted when observations indicated that the dog was uncomfortable. Induction. In this preexposure phase, all subjects rested in the full hammock. Each session began with the onset of the houselight. Subjects in the Paired-shocks group and Random-shocks group were exposed to a series of electric shocks, each of 4-set duration presented at varying intervals averaging 2 min and ranging from 1 to 3 min. A random half of the shocks were delivered to the left rear leg and the remaining shocks were delivered to the right rear leg. Shocks delivered to the right leg were pulsed ten times per second while those to the left leg were unpulsed. (Two different kinds of shocks were employed because this facilitates response acquisition in our choice test task. See Overmier, Bull, & Trapold, 1971.) For subjects in a Paired group, each shock was preceded by a 5-set auditory signal which then continued until shock termination, For subjects in replication 1, shocks to the left rear leg were preceded by the high tone and shocks to the right rear leg were preceded by the low tone. For subjects in replication 2, high tones preceded shocks to the right leg and low tones preceded the shocks to the other leg. Each subject in the Random group was matched with a subject in the Paired group and thus experienced the same number of shocks in the same temporal order and with the same temporal spacing as the partner subject in the Paired group. However, for all dogs in the Random-shocks group, the high and low tones were presented on a random schedule, independent of the schedule for shocks. That is, tones and shocks were uncorrelated in this group, and the shocks were unpredictable both as to time of occurrence and place of delivery.
330
OVERMIER
AND
WIELKIEWICZ
Each subject in the CS-only group of replication 1 was exposed to the same series of high and low tones experienced by a partner subject in the Paired group (replication 1). However, subjects in the CS-only group did not experience any electric shocks in the induction phase. Each subject in the Shocks-only group of replication 2 was exposed to the same series of shocks experienced by a partner subject in the Paired group (replication 2). However, the subjects in the Shocks-only group did not experience any tones in the induction phase. The experiment was run in two systematic replications (Sidman, 1960). The replications differed in the number of shocks per induction session (40 versus 20) and the number of sessions of preexposure to shock (6 versus 12). There were no other differences between replications. Testing. The day following the last preexposure session was the first of two daily sessions of discriminative choice escape/avoidance training. In this testing phase, the subjects were presented with the following problem: Each 15set trial began with the onset of a 10 Hz flashing light (equal on-off cycle) located directly under the left front paw (Sy) or right front paw (SF). If (Sp) was presented, the animal was required to lift its left front paw within 5 set to avoid pulsed shock to its right rear leg. If (Sy) was presented, the animal was required to lift its right front paw within 5 set to avoid shock to its left rear leg. If during the 5-set CS-US interval the initial response of the animal was correct, then the trial was immediately terminated and an intertrial interval was initiated. If the initial response was wrong, a correction procedure allowed the animal to switch responses but now the criterion was lift and hold for the remainder of the trial. If the dog did not perform an initial correct avoidance within 5 set of the beginning of the trial, shock was scheduled for the final 10 set of the trial except during any period the correct response was being performed. During the final 10 set of each trial, the flashing SD light and the shock were turned off whenever the dog made the correct response. Each correct response during a trial was accompanied by extra feedback events of (1) a 0.5-set offset of the white noise and (2) onset of the side panel light. Correct responses did not include those times when the dog lifted both front paws. The interval between trials averaged 2 min and ranged from 1 to 3 min. Each test session was composed of ten (SF) and ten (Sy) trials. Thus, in this test of learning a discriminative choice escape/avoidance task, if a subject’s initial anticipatory response was correct it avoided shocks; if the response was not correct a correction procedure allowed the subject to find the correct response but now it had to maintain it so as to prevent delivery of scheduled aversive stimuli. The requirement or response persistence here for escape/avoidance is copied from Skipin’s and Vinnik’s modification of the method of Petropavlovskii as described in Skipin, Ivanova, Kozlovskaya, & Vinnik (1969).
UNPREDICTABILITY
AND LEARNED
HELPLESSNESS
331
RESULTS The main dependent variable was the amount of time per 15set trial during which the subject was not performing the correct response (wrong time). This was averaged across the three subjects in each group of each replication for each test day. Learning the task, then, was reflected by a decline in mean wrong time. These data are presented in Fig. 1 for all groups as a function of blocks of the 20 discriminative choice training trials of each test day. The data summarized in Fig. 1 were subjected to an analysis of variance in which the factors were Replications (1,2) x Induction Treatments (Paired, Random, Control) x Test Days (1,2) with repeated measures on dogs over the two test days. All statistical tests were conducted at the .05 level of confidence. It appears that all groups showed learning across days (i.e., decreasing mean wrong times), Days F(1, 12) = 12.61, and that the two replications produced substantially the same patterns of results. The ANOVA confirms the similarity of results across replications. Replications and all interactions with replications were not sign&cant, F’s < 1.3. The absence of differences between replications was additionally confirmed by follow-up comparisons on each treatment across replications, all three t’s < 1.20. In particular among these, note that the CS-only was not significantly different from Shocks-only, t < 0.01. The induction treatment conditions did, however, have different effects upon discriminative choice escape/avoidance performance, F(2, 12) = 7.02. The Random groups were significantly impaired relative to the Paired groups, t(12) = 3.40, and relative to the Control groups, t(12) = REPLICATION
1
REPLICATION
2
, I 1 CHOICE
PAIRED RANDOM I 2 RESPONSE
0 .
CS ONLY 4 SHK ONLYA I I I 1 2 ACQUISITION DAYS
FIG. 1. Daily mean wrong time (time during daily trials during which dogs failed to make the correct response) for each group on each of 2 days of test phase training on the discriminative choice escape/avoidance task.
332
OVERMIER
AND WIELKIEWICZ
2.79.’ However, the Paired groups and the Control groups did not differ reliably, t < 1. To ascertain whether the between treatments effects seen in mean wrong times (Random > Paired = Control) are not an artifact of differential response initiation, latencies to first responses, correct and incorrect, on each trial were analyzed as a function of Replications x Treatments x Test Days, paralleling the analysis of the primary dependent variable. Latencies on the second test day were shorter than on the first day, F(1, 12) = 8.05, reflecting learning in a second way. However, central to our concern here, latencies did not differ as a function of the induction treatments; treatments and all interactions involving treatments F’s were not significant. DISCUSSION The pattern of empirical effects in the present experiment is simple and straightforward: Prior exposure to tones and shocks presented on random and independent schedules interferes with subsequent learning to make correct choice response to visual SDs in an escape/avoidance choice task. This interference is ameliorated by having the tones predict the shocks. These results confirm the earlier preliminary observation (Overmier et al., 1980) that preexposure to unpredictable shocks caused more proactive interference than preexposure to predictable shocks. Compared to the CS-only (no shocks) reference level, only preexposure to inescapable shocks and uncorrelated signals produced proactive interference. That prior exposure to predictable, inescapable shocks had no interfering effect is at first glance surprising from the traditional “learned helplessness” perspective. But recall that the primary deficit observed in the traditional shuttlebox task is one of reduced initiations of the effortful response of jumping over a barrier. Here, we purposefully chose a less effortful task and a measure based upon discriminative choice behavior with the intent of detecting associative interference rather than initiation interference. Indeed, all our animals made several responses on nearly every test trial. Thus, our data primarily reflect the facility with which the dogs learned the (Sf - R,)/(Sf - R2) discriminative choice problem. The latency data confirm that we were successful in designing a learning task in which the criterion response was sufficiently easy not to be influenced by the response initiation deficit obtained in more effortful tasks. (See Glazer & Weiss, 1976, for another demonstration that the initiation of appropriately selected instrumental acts may be unimpaired by exposures to uncontrollable shocks which commonly produce learned ’ Although these between treatment contrasts were dictated a priori by our hypothesis. some will be comforted to know that these contrasts are confirmed using Newman-Keuls procedures, q, = 4.77 and q2 = 3.90.
UNPREDICTABILITY
AND
LEARNED
HELPLESSNESS
333
helplessness.) Hence, the impaired learning of our conditional choice task by the Random groups reflected in the mean wrong times is attributable to differential choosing behavior resulting from pretreatment with the independent presentations of CS and US. While the present results are unique within the “learned helplessness” literature, they may be related to other phenomena documented in the Pavlovian conditioning literature. Mackintosh (1973) and Baker (1976) have published demonstrations of a phenomenon that has been called ‘ ‘learned irrelevance. ” In these experiments, four groups of rats were preexposed to either uncorrelated presentations of CS (tone or noise) and shock, shock-alone, CS-alone, or no stimulus. The subjects were then tested for acquisition of response suppression when the same CS now signaled electric shocks and was presented while the animal was responding freely for appetitive reinforcers (water and food). During this test, the group preexposed to uncorrelated CS shock presentations was slowest in acquiring conditioned suppression. The attentional account of these results offered was that those rats exposed to the uncorrelated presentations of CS and shocks learned that these two particular events were irrelevant to each other. Mackintosh (1973) further demonstrated that this irrelevance learning was specific to the US used in the preexposure phase. In contrast to Mackintosh’s (1973) and Baker’s (1976) experiments in which exposure to uncorrelated presentations of a stimulus and shock interfered with later acquisition of conditioned suppression in the presence of the same cue stimulus, the present finding was that uncorrelated presentations of one CS and shocks interfered with acquisition of shockmotivated responding to distinctly different cue stimuli. Thus, the present finding is of a generalized learned irrelevance effect which transcends the CS conditions of the original preexposure to shock. Since tones were employed as the CSs in the first phase of the present experiment and lights were employed as the discriminative stimuli in the test phase, it is not likely that simple stimulus generalization could account for these effects. Thus, while Mackintosh showed that the learned irrelevance was specific with respect to the US, our results suggest that there is a general associative interference with respect to CSs. Related experiments by Wickens, Tuber, Nield, and Wickens (1977) previously hinted at the existence of a generalized irrelevance effect. Wickens et al. tested the effect of prior exposure to a random and uncorrelated CS and US upon later Pavlovian conditioning employing either the same CS and US combination or a different CS and different US. These they compared (across experiments) to a group preexposed to pairings of a CS and US and later Pavlovianly conditioned with a different CS and different US. They found that these three groups in the second Pavlovian conditioning phase took significantly different numbers
334
OVERMIER
AND
WIELKIEWICZ
of sessions to reach criterion. The group preexposed to random, independent presentations of the CS and US later to be used in the Pavlovian pairing was the slowest to learn; the group preexposed to random, independent presentations of a CS and US, but ones different from those paired in the Pavlovian phase was the fastest. This pattern is consistent with the assumption that some proportion of the learned irrelevance effect is specific to the particular CS and US presented in the preexposure phase, but also that there is a generalized learned irrelevance contribution as well. Thus, the Wickens et al. (1977) experiments provide a Pavlovian conditioning analog to the present study. In both cases, prior exposure to random signals and shock caused significantly slower acquisition than prior exposure to paired signals and shock-an effect which transcended the particular CS used during the preexposure phase. Let us also consider two other potential accounts of our results: (a) blocking and (b) reduced effective reinforcer. These two accounts have in common that they rely upon the empirical phenomenon that exposing an organism to a US has the effect of slowing subsequent conditioning when that same US is now signaled by a CS (Kamin, 1961; see Randich & LoLordo, 1979, for review). They differ in the theoretical mechanism invoked. According to the blocking account, conditioning to any CS in the test phase is blocked by the excitatory strength already conditioned to the context (or background) stimuli during the US preexposure phase (e.g., Rescorla & Wagner, 1972). Since the US is already well predicted by the context, the CS cannot acquire any additional excitatory strength. On this account, however, the contextual stimuli should not become so excitatory during the preexposure phase if the US presentations were signaled by some CS. The signaling would “protect” the context from becoming excitatory so that later the context would not block the association of the US with some new CS. Thus, the blocking account would predict the pattern of results we obtained in replication 1: Random-shocks groups were slower to learn than the Paired-shocks group and the CS-only group, which in turn might not differ. The effective reinforcer account (McAllister, McAllister, & Douglass, 1971) suggests that the interference with learning in our avoidance test would obtain because prior exposures to unsignaled USs would result in conditioning of fear to the background (or the intertrial interval). Thus, in the avoidance training phase, when the dogs performed successful escape or avoidance responses during the discriminative stimuli, they would be returned to the background stimulus conditions which would continue to elicit fear, hence minimizing any potential reinforcement for the correct response. Such disruption of the reinforcement process hypothesized by two-process theory would not occur if the background stimuli were less fear eliciting-as would be the case when there had
UNPREDICTABILITY
AND
LEARNED
HELPLESSNESS
335
been no prior US exposures (as in our CS-only group) or when any US preexposures had been signaled (as in our Paired-shocks group). This means that the effective reinforcement theory might also account for the pattern of results obtained in replication 1. Replication 2 was included to provide a test for these blocking and reduced reinforcer effectiveness accounts. Replication 2 compared learning by groups exposed to Paired, Random, and Shock-only groups. It is important to note that neither of these accounts would predict differential outcomes between our Random group (which received random CSs and independent shocks) and our Shock-only group, while a general learned irrelevance account would predict impairment only in the former. Moreover, a learned irrelevance account would suggest that our CS-only group (replication 1) and our Shocks-only group (replication 2) should acquire equally well in our discriminative choice avoidance test task (see Mackintosh, 1973). Note that the test performance of the Shocks-only group was almost identical to that of the CS-only group and that both were superior to the Random group. This pattern contravenes both the blocking and the effective reinforcer accounts of our results but is consistent with a generalized learned irrelevance account. At the present time, our understanding of these phenomena remains tentative. Our data show that preexposure to Random CSs and shocks interferes with condkional choice escape/avoidance learning more than preexposure to shocks-only, CS-only, or predicted-shocks. This effect is apparently independent of the response initiation deficit of the learned helplessness phenomenon seen with motorintensive non-choice tasks. It does parallel the learned irrelevance phenomenon. Our observation extends those instances of learned irrelevance reported by Mackintosh (1973) and by Tomie, Murphy, Fath, and Jackson (1980), however, because the present results demonstrate the general nature of the effects in transcending CSs, so long as the US is held constant over phases. One potentially problematic feature of our data is the absence of difference between our two control groups, CS-only and US-only. While this absence of difference allows us to identify the generalized learned irrelevance phenomenon, it raises a question as to why the US exposures to the US-only control do not cause an impairment in associative learning; after all these USs were unpredictable as well. Moreover, it does not allow us to choose between the report of Irwin et al. (1980) and that of Jackson et al. (1980) because under one unpredictable US condition, Random CSs and USs, we found impairment while under another unpredictable US condition, US-only, we did not. The latter suggests that associative interferences heretofore reported may have reflected primarily some response initiation deficit as Alloy and Seligman (1979) have suggested. That is not the case here. Alternatively, there may be more than one kind of associative proactive interference the causal factors of which may differ.
OVERMIER
336
AND WIELKIEWICZ
A goodly amount of additional research is needed before a clear picture of proactive inference phenomena will emerge. We need more integrative research which brings together proactive interference phenomena such as generalized learned helplessness and generalized learned irrelevance. Baker (1976) provides an example of one such tack. The present data provide another and are consistent with the suggestion that the proactive interference phenomenon seen in the multifaceted learned helplessness syndrome may in fact be multidimensional in terms of its causal factors as well as in terms of its consequent features. Indeed, it has always seemed to us that one weakness of the “learned helplessness” theory was its reliance upon one single experimental factor and its resultant cognitive state as a single causal entity of three effects (emotional, incentive motivational, and associative). This feature of the theory requires that whenever the three are independently assessed, they ought to show substantial covariation. Unfortunately this has not always been so. The present experiment provides evidence that the learned helplessness syndrome may well be the product of multiple interacting causal factors which are typically confounded but may under appropriate experimental conditions be isolated for study. One of these is generalized learned irrelevance. REFERENCES Alloy, L. B., & Seligman, M. E. P. On the cognitive component of learned helplessness and depression. In G. Bower (Ed.). The psychology of learning and motivation. New York: Academic Press, 1979. Vol. 13, pp. 219-276. Altenor, A., Kay, E., & Richter, M. The generality of learned helplessness in the rat. Learning
and Motivation,
1977, 8, 54-62.
Baker, A. G. Learned irrelevance and learned helplessness: Rats learn that stimuli, reinforcers, and responses are uncorrelated. Journal of Experimental Psychology: Animal Behavior Processes, 1976, 2, 130-141. Baker, A. G., & Mackintosh, N. J. Preexposure to the CS alone, US alone, or CS and US uncorrelated: Latent inhibition, blocking by context, or learned irrelevance? Learning and Motivation, 1979, 10, 278-294. Caspy, T., & Lubow, R. E. Generality of US preexposure effects: Transfer from food to shock or shock to food with and without same response requirements. Animal Learning & Behavior, 1981, 9, 524-532. Glazer, H. I., & Weiss, J. M. Long-term interference effect: An alternative to “learned helplessness.” Journal of Experimental Psychology: Animal Behavior Processes, 1976, 2, 202-213. Imada, H., & Soga, M. The CER and BEL as a function of predictability and escapability of an electric shock. Japanese Psychological Research, 1971, 13, 115-122. Irwin, J., Suissa, A., & Anisman, H. Differential effects of inescapable shock on escape performance and discrimination learning in a water escape tank. Journal of Experimental Psychology: Animal Behavior Processes, 1980, 6, 21-40. Jackson, R. L., Alexander, J. H., & Maier, S. F. Learned helplessness, inactivity, and associative deficits: Effects of inescapable shock on response choice escape learning. Journal of Experimental Psychology: Animal Behavior Processes, 1980, 6, l-20. Kamin, L. J. Apparent adaptation effects in the acquisition of a conditioned emotional response. Canadian Journal of Psychology, 1961, 15, 176-188. Lubow, R. E. Latent inhibition. Psychological Bulletin; 1973, 79, 398-407.
UNPREDICTABILITY
AND LEARNED
HELPLESSNESS
337
Mackintosh, N. J. Stimulus selection: Learning to ignore stimuli that predict no change in reinforcement. In R. A. Hinde &J. Stevenson-Hinde (Eds.), Construints on learning. London/New York: Academic Press, 1973. Pp. 75-96. Maier, S. F., & Seligman, M. E. P. Learned helplessness: Theory and evidence. Journal of Experimental Psychology: General, 1976, 105, 3-46. McAllister, W. R., McAllister, D. E., & Douglass, W. K. The inverse relationship between shock intensity and shuttlebox avoidance learning in rats: A reinforcement explanation. Journal
of Comparative
and Physiological
Psychology,
1971, 74, 426-433.
Overmier, J. B., Bull, J. A., III, & Trapold, M. A. Discriminative cue properties of different fears and their role in response selection in dogs. Journal of Comparative and Physiological
Psychology,
1971. 76, 478-482.
Overmier, J. B., Patterson, J., & Wielkiewicz, R. M. Environmental contingencies as sources of stress in animals. In S. Levine & H. Ursin (Eds.), Coping and he&h. New York: Plenum, 1980. Pp. l-38. Overmier, J. B., & Seligman, M. E. P. Effects of inescapable shock upon subsequent escape and avoidance responding. Journal of Comparative and Physiological, 1967. 63, 28-33. Pubols, B. H., Jr. Incentive magnitude, learning, and performance in animals. Psychological Bulletin, 1960, 57, 89-115. Randich, A., & LoLordo, V. M. Associative and nonassociative theories of the UCS preexposure phenomenon: Implications for Pavlovian conditioning. Psychological Bulletin, 1979, 86, 523-548. Rescorla, R. A., & Wagner, A. R. A theory of Pavlovian conditioning. In A. H. Black & W. F. Prokasy (Eds.), Classical conditioning II: Current research and theory. New York: Appleton-Century-Crofts, 1972. Pp. 64-99. Seligman, M. E. P., & Maier, S. F. Failure to escape traumatic shock. Journal of Experimental Psychology, 1967, 74, l-9. Seligman, M. E. P., Maier, S. F., & Solomon, R. L. Unpredictable and uncontrollable aversive events. In F. R. Brush (Ed.), Aversive conditioning and learning. New York: Academic Press, 1971. Sidman, M. Tactics of scient$c research. New York: Basic Books, 1960. Skipin, G. V., Ivanova, I. G., Kozlovskaya, I. B., & Vinnik, R. L. Defense conditioning and avoidance. In M. Cole and I. Maltzman (Eds.), Handbook of contemporary Societ psychology. New York: Basic Books, 1969, Pp. 785-810. Tomie, A., Murphy, A. L., Fath, S., &Jackson, R. L. Retardation of autoshaping following pretraining with unpredictable food: Effects of changing context between pretraining and testing. Learning and Motivarion, 1980, 11, 117-134. Underwood, B. J. Interference and forgetting. Psychological Review, 1957, 64, 49-60. Weiss, J. M. Effects of coping behavior in different warning signal conditions on stress pathology in rats. Journal of Comparative and Physiological Psychology, 1971, 77, l-13. Wickens, D. D., Tuber, D. S., Nield, A. F., & Wickens, C. Memory for the conditioned response: The effects of potential interference introduced before and after original conditioning. Journal of Experimental Psychology: General, 1977, 106, 47-70. Williams, J. L., & Maier, S. F. Transsituational immunization and therapy of learned helplessness in the tat. Journal of Experimental Psychology: Animal Behavior Processes, 1977, 3, 240-253. Received May 12, 1983 Revised September 13, 1983