Extinction and blocking of conditioned inhibition in human causal learning

Two experiments investigated extinction and blocking of a conditioned inhibitor in a human contingency learning task. Lotz and Lachnit (2009) and Melchers, Wolff, and Lachnit (2006) reported extinction of inhibition only when participants experienced outcome levels lower than those used in training. In Experiment 1, which used more neutral instructions than the previously mentioned studies, we found that extinction of inhibition occurred, whether or not participants experienced lower outcome levels. In Experiment 2, we applied this outcome manipulation to blocking of a conditioned inhibitor. We found blocking of inhibition both when participants had experience with lower outcomes and when they did not. The results of our two experiments are consistent with Rescorla and Wagner’s (1972) associative model, and inconsistent with an inferential account of causal learning (De Houwer, Beckers, & Vandorpe, 2005).     

The effect of subadditive pretraining on blocking: limits on generalization

Recent evidence indicates that prior learning about a set of cues may determine how new cues are processed. If subjects are taught that two reliable predictors of an outcome do not summate when the cues are presented together (i.e., subadditive pretraining), the subjects will tend to show a less profound blocking effect when trained with different cues. Three experiments investigated the conditions necessary for subadditive pretraining to generalize to new cues. Experiment 1 demonstrated that subadditive pretraining is less effective in reducing blocking when it is experienced in a context other than that in which the blocking training is experienced. In Experiment 2, the effectiveness of subadditive pretraining waned with time. Experiment 3 showed that subadditive pretraining is more effective when the temporal characteristics of pretraining cues are similar to those of the cues used in blocking training. These results provide information concerning the conditions under which learning will generalize from one set of cues to another.     

Generalization of causal efficacy judgments after evaluative learning

In three experiments, we examined the effect of response-outcome relations on human ratings of causal efficacy and demonstrated that such efficacy ratings transfer to novel situations through derived stimulus relations. Causal efficacy ratings were higher, and probability of an outcome given a response was lower, for a differential reinforcement of high rate schedule than for either a differential reinforcement of low rate schedule (Experiment 1) or a variable interval schedule (Experiment 2). In Experiment 3, we employed schedules that were equated for outcome probability and noted that ratings of causal efficacy and the rate of response were higher on a variable ratio than on a variable interval schedule. For participants in all three experiments, causal efficacy ratings transferred to the stimulus present during each schedule and generalized to novel stimuli through derived relations. The results corroborate the view that schedules are a determinant of both response rates and causal efficacy ratings. In addition, the novel demonstration of a mechanism of generalization of these ratings via derived relations has clinical implications.     

Learned predictiveness effects following single-cue training in humans

The results of several recent studies of human associative learning indicate that people will learn more rapidly about cues that have previously been experienced as predictive of events of significance, as compared with cues previously experienced as nonpredictive. Notably, however, these experiments have typically established this prior predictiveness by means of pretraining with multiple, simultaneously presented cues, some of which are more predictive than others. The present experiments instead investigated the influence of prior predictiveness on future learning when this predictiveness was established via pretraining with individual cues, each of which was the best available predictor of the outcome with which it was paired. Results indicate that, following this pretraining, human participants again show better learning about previously predictive cues than about previously nonpredictive cues.     

Overshadowing and blocking between landmark learning and shape learning: the importance of sex differences

Rats were trained in a triangular-shaped pool to find a hidden platform that maintained a constant relationship with two sources of information, an individual landmark and one part of the pool with a distinctive shape. In Experiment 1, shape learning overshadowed landmark learning but landmark learning did not overshadow shape learning in males, while landmark learning overshadowed shape learning but shape learning did not overshadow landmark learning in females. In Experiment 2, rats were pretrained either with the single landmark relevant or with the shape relevant, in the absence of the alternative cue. Final test trials, without the platform, revealed reciprocal blocking only in females; in males, shape learning blocked landmark learning, but not viceversa (Experiment 2a). In Experiment 2b, male rats received a longer pretraining with the single landmark relevant, and now landmark learning blocked shape learning. The results thus confirm the claim that males and females partially use different types of spatial information when solving spatial tasks. These results also agree with the suggestion that shape learning interacts with landmark learning in much the same way as does learning about any pair of stimuli in a Pavlovian conditioning experiment.     

Simulations of a modified SOP model applied to retrospective revaluation of human causal learning

Dickinson and Burke (1996) proposed a modified version of Wagner’s (1981) SOP associative theory to explain retrospective revaluation of human causal judgments. In this modified SOP (MSOP), excitatory learning occurs when cue and outcome representations are either both directly activated or both associatively activated. By contrast, inhibitory learning occurs when one representation is directly activated while the other is associatively activated. Finite node simulations of MSOP yielded simple acquisition, overshadowing, blocking, and inhibitory learning under forward contingencies. Importantly, retrospective revaluation was predicted in the form of unovershadowing and backward inhibitory learning. However, MSOP did not yield backward blocking. These predictions are evaluated against the relevant empirical evidence and contrasted with the predictions of other associative theories that have been applied to retrospective revaluation of human causal and predictive learning.     

Surprise and change: Variations in the strength of present and absent cues in causal learning

It is said that “absence makes the heart grow fonder.” But, when and why does an absent event become salient to the heart or to the brain? An absent event may become salient when its nonoccurrence is surprising. Van Hamme and Wasserman (1994) found that a nonpresented but expected stimulus can actually change its associative status—and in the opposite direction from a presented stimulus. Associative models like that of Rescorla and Wagner (1972) focus only on presented cues; so, they cannot explain this result. However, absent cues can be permitted to change their value by assigning different learning parameters to present and absent cues. Van Hamme and Wasserman revised the Rescorla-Wagner model so that the α parameter is positive for present cues, but negative for absent cues; now, changes in the associative strength of absent cues move in the opposite direction as presented ones. This revised Rescorla-Wagner model can thus explain such otherwise vexing empirical findings as backward blocking, recovery from overshadowing, and backward conditioned inhibition. Moreover, the revised model predicts new effects. For example, explicit information about the absence of nonpresented cues should increase their salience (that is, their negative α value should be larger), leading to stronger associative changes than when no explicit mention is made of cue absence. Support for this prediction is detailed in a new causal judgment experiment in which participants rated the effectiveness of different foods’ triggering a patient’s allergic reaction. Overall, these and other findings encourage us to view human causal learning from an associative perspective.     

Contiguity and covariation in human causal inference

Nearly every theory of causal induction assumes that the existence and strength of causal relations needs to be inferred from observational data in the form of covariations. The last few decades have seen much controversy over exactly how covariations license causal conjectures. One consequence of this debate is that causal induction research has taken for granted that covariation information is readily available to reasoners. This perspective is reflected in typical experimental designs, which either employ covariation information in summary format or present participants with clearly marked discrete learning trials. I argue that such experimental designs oversimplify the problem of causal induction. Real-world contexts rarely are structured so neatly; rather, the decision about whether a cause and effect co-occurred on a given occasion constitutes a key element of the inductive process. This article will review how the event-parsing aspect of causal induction has been and could be addressed in associative learning and causal power theories.     

Transfer of judgments of control to a target stimulus and to novel stimuli through derived relations

Three experiments examined the effect of response?Coutcome contingencies on human ratings of causal efficacy and demonstrated that such ratings transfer to novel situations through derived stimulus relations. Efficacy ratings generally followed the delta probability rule when positive response-outcome contingencies were employed (Experiment 1) and when some outcomes were not contingent on participants?? responses (Experiment 2). Experiment 3 employed a negative response?Coutcome contingency and manipulated performance expectancies in the task. All three groups overestimated their causal efficacy ratings. A learned helplessness effect was observed when the response?Coutcomes were uncontrollable and in the high-expectancy group when participants?? performance in the task was worse than they had expected. In all experiments, ratings transferred to a stimulus presented during the task and often generalized to novel stimuli through derived relations. These results corroborate the view that outcome probability is a determinant of causal efficacy ratings and that schedules can be employed as UCS in procedures that share characteristics of evaluative conditioning procedures.     

God does not play dice: causal determinism and preschoolers' causal inferences

Three studies investigated children's belief in causal determinism. If children are determinists, they should infer unobserved causes whenever observed causes appear to act stochastically. In Experiment 1, 4-year-olds saw a stochastic generative cause and inferred the existence of an unobserved inhibitory cause. Children traded off inferences about the presence of unobserved inhibitory causes and the absence of unobserved generative causes. In Experiment 2, 4-year-olds used the pattern of indeterminacy to decide whether unobserved variables were generative or inhibitory. Experiment 3 suggested that children (4 years old) resist believing that direct causes can act stochastically, although they accept that events can be stochastically associated. Children's deterministic assumptions seem to support inferences not obtainable from other cues.     

Two-choice,observational learning and reversal in the rat: S-S versus S-R effects

In Experiment 1, male rats were trained to press both bars in a two-choice apparatus and were then given observational training of a go/no-go discrimination in which the observed operation of two inaccessible, dissimilar bars by a hidden experimenter constituted S+ and S?. After discrimination was established, individual rats were permitted access to the two bars. Six of the seven rats consistently pressed the S+ bar on 10 test trials, but failed to reverse bar preference after observational training was reversed. In Experiment 2, nine naive males received the same observational training as in Experiment 1, but without any pretraining to press either bar. All rats pressed the S+ bar on initial test and did so consistently throughout the 10 trials. Six of these rats received reversal training of the go/no-go discrimination after the 10 test trials. As in Experiment 1, all rats failed to press the new S+ bar. However, five of six rats in another group, which received reversal trainingprior to any test trials, did reverse and press the new S+ bar. In Experiment 3, controls for possible confounding effects of overtraining trials were conducted. These manipulations had no effect; the rats tested before reversal still failed to press the S+ bar, and the rats reversed before testing all reversed or pressed the most recent S+ bar. That is, S-R learning predominated over S-S learning if active, though unreinforced, responding to a particular bar intervened. In contrast, however, a cognitive (S-S) interpretation of directed response learning was supported by the results of Experiment 4, in which the rats that learned the go/no-go discrimination without responding (only by auditory and light cues) failed to press the S+ bar consistently.     

Associative foundation of causal learning in rats

Are humans unique in their ability to interpret exogenous events as causes? We addressed this question by observing the behavior of rats for indications of causal learning. Within an operant motor–sensory preconditioning paradigm, associative surgical techniques revealed that rats attempted to control an outcome (i.e., a potential effect) by manipulating a potential exogenous cause (i.e., an intervention). Rats were able to generate an innocuous auditory stimulus. This stimulus was then paired with an aversive stimulus. The animals subsequently avoided potential generation of the predictive cue, but not if the aversive stimulus was subsequently devalued or the predictive cue was extinguished (Exp. 1). In Experiment 2, we demonstrated that the aversive stimulus we used was in fact aversive, that it was subject to devaluation, that the cue–aversive stimulus pairings did make the cue a conditioned stimulus, and that the cue was subject to extinction. In Experiments 3 and 4, we established that the decrease in leverpressing observed in Experiment 1 was goal-directed instrumental behavior rather than purely a product of Pavlovian conditioning. To the extent that interventions suggest causal reasoning, it appears that causal reasoning can be based on associations between contiguous exogenous events. Thus, contiguity appears capable of establishing causal relationships between exogenous events. Our results challenge the widely held view that causal learning is uniquely human, and suggest that causal learning is explicable in an associative framework.     

Effect of outcome valence on positive and negative patterning in human causal reasoning

The importance of configural cues and whether a situation involves beneficent or maleficent outcomes was investigated in two experiments on human causal reasoning, based on experienced causal information. Participants learned positive and negative patterning discriminations involving either beneficent or maleficent outcomes in a health-reasoning task and in a social-reasoning task. With maleficent outcomes, positive patterning was consistently easier to learn than negative patterning, a positive patterning advantage that is predicted by current associative theories and commonly taken as evidence for configural cues. However, with beneficent outcomes, the two discrimination tasks were not significantly different in ease of learning, a result not predicted by current theories. The reliable positive patterning effect found with maleficent outcomes broadens the range of conditions in which the effect can be shown in causal reasoning. The novel effect of outcome valence poses an interesting theoretical challenge for attempts to account for the relation between learning about individual cues and combinations of those cues.     

The influence of the length of causal chains on question asking and on the comprehensibility of scientific texts

This study examines question asking about causal relations when students read scientific texts. We examine the influence of the length of causal chains and the knowledge of readers. Students from grades 8 and 12 read two short paragraphs that described natural phenomena. Length of the causal chain linking cause and effect in two key sentences was manipulated. The students were instructed to ask on anything that they did not understand. The results showed, in the first place, that students ask a majority of causal antecedent questions. Second, a longer causal chain resulted in less causal questions asked by the 12th grade students, but did not reduce the number of causal questions asked by 8th grade students. Third, there was no relation between comprehensibility ratings given to the causal relations and causal question asking. According to this, an explicit comprehensibility score that taps on the metacognitive awareness of students may not be a good indicator of the level of understanding.     

A landmark blocks searching for a hidden platform in an environment with a distinctive shape after extended pretraining

In the blocking phase of three experiments, rats had to find a submerged platform beneath a spherical landmark in one corner of a triangular pool. Prior to this treatment, they were required to find the platform relative to either a sphere above it (blocking groups) or a rod attached to it (control groups). The position of the platform changed from trial to trial for the initial training. The sphere did not restrict learning about the geometric cues provided by the triangular arena in the blocking phase when 12 sessions of initial training took place in either the triangular (Experiment 1) or a circular (Experiment 3) pool. Blocking was observed, however, after 24 sessions of initial training in either the triangular (Experiment 2) or the circular (Experiment 3) pool. Thus, blocking of geometric cues by a landmark is possible after extended initial training with the blocking cue.     

A common error term regulates acquisition but not extinction of causal judgments in people

In three experiments, we used the allergist task to examine the role of error correction mechanisms in the acquisition and extinction of causal judgments in people. Consistent with existing human and animal studies, acquisition of causal judgments was influenced by the discrepancy between the allergenic outcome and that predicted by all of the cues present on a trial (the "common error" term). However, in the present experiments, we failed to detect any evidence for the use of a common error term in extinction learning: Judgments of the allergenic properties of a cue were unaffected by the predictive value of the other cues present on a trial. This asymmetry in the use of a common error term in acquisition and extinction learning is inconsistent with previous animal studies and also with most models of associative learning. However, approaches that allow learning to be specific to a particular arrangement of elemental cues (context specific and state based) offer some explanation of the observed asymmetry.     

Effects of pretraining on acquisition of novel configural discriminations in human predictive learning

In two experiments, participants acquired one of two target configural discriminations (a biconditional or negative patterning discrimination) in a predictive learning task. In Experiment 1, participants were pretrained with either a configural or an elemental discrimination; in Experiment 2, they were pretrained with a configural discrimination, an elemental discrimination, or a control discrimination that was not expected to bias them toward elemental or configural processing. In both experiments, acquisition of the target configural discriminations was faster after configural pretraining than after elemental pretraining. In addition, the negative patterning discrimination was acquired faster than the biconditional discrimination. Finally, the results of Experiment 2 were more consistent with the notion that elemental pretraining hindered acquisition of the target discriminations than with the notion that configural pretraining enhanced their acquisition. Implications of these findings are discussed.     

Competence and performance in causal learning

The dominant theoretical approach to causal learning postulates the acquisition of associative weights between cues and outcomes. This reduction of causal induction to associative learning implies that learners are insensitive to important characteristics of causality, such as the inherent directionality between causes and effects. An ongoing debate centers on the question of whether causal learning is sensitive to causal directionality (as is postulated by causal-model theory) or whether it neglects this important feature of the physical world (as implied by associationist theories). Three experiments using different cue competition paradigms are reported that demonstrate the competence of human learners to differentiate between predictive and diagnostic learning. However, the experiments also show that this competence displays itself best in learning situations with few processing demands and with convincingly conveyed causal structures. The study provides evidence for the necessity to distinguish between competence and performance in causal learning.     

Learning by teaching in immersive virtual reality – Absorption tendency increases learning outcomes

We investigated the learning outcome of teaching an agent via immersive virtual reality (IVR) in two experiments. In Experiment 1, we compared IVR to a less immersive desktop setting and a control condition (writing a summary). Learning outcomes of participants who had explained the topic to an agent via IVR were better. However, this was only the case for participants who scored high on absorption tendency. In Experiment 2, we investigated whether including social cues in the task instructions enhances learning in participants explaining a topic to an agent. Instruction manipulation affected learning as a function of absorption tendency: Low-absorption participants benefitted most from being instructed to imagine they were helping a student peer pass an upcoming test, while high-absorption participants benefitted more when they were to explain the text to a virtual agent. The findings highlight the crucial role of personality traits in learning by teaching in IVR.     

Using an analogy to model causal mechanisms in a complex text

Two experiments examined the role of analogy inthe development of a situation model of atarget passage. Specifically, we examinedwhether an analogous source text could improvecomprehension and inferencing about causalmechanisms in the target. An ``analogy group'studied a fictional pseudo-scientific passageunder the guidance of an analogous passage(also fictional) which shared causal structureswith the target. A ``statement group' studiedthe target under the guidance of an abstractstatement that characterized the target causalstructures. A ``target-only group' received noadditional material. All groups received tasksthat assessed their representation of thetarget. In Experiment 1 (N = 92) participantswere asked to list the most important objectsand relations in the passage. In Experiment 2(N = 80) participants were directly asked aboutthe target causal structures. In bothexperiments, responses of analogy subjects weremore likely to include given and inferredinformation that comprised the target causalmechanisms. Analogies in which a sourcepassage models causal mechanisms correspondingto those of a target passage appear to promotebetter representation of the details of targetcausal mechanisms, allowing a more fullydeveloped situation model of the target.