Disrupted prediction-error signal in psychosis: evidence for an associative account of delusions.

Delusions are maladaptive beliefs about the world. Based upon experimental evidence that prediction error-a mismatch between expectancy and outcome--drives belief formation, this study examined the possibility that delusions form because of disrupted prediction--error processing. We used fMRI to determine prediction-error-related brain responses in 12 healthy subjects and 12 individuals (7 males) with delusional beliefs. Frontal cortex responses in the patient group were suggestive of disrupted prediction-error processing. Furthermore, across subjects, the extent of disruption was significantly related to an individual's propensity to delusion formation. Our results support a neurobiological theory of delusion formation that implicates aberrant prediction-error signalling, disrupted attentional allocation and associative learning in the formation of delusional beliefs.

Neuronal correlates of familiarity-driven decisions in artificial grammar learning.

It has been proposed on the basis of behavioural data that grammaticality judgments in implicit artificial grammar learning paradigms are largely driven by priming based on fragment familiarity. A prediction that follows from this account is that neural deactivation, a common correlate of repetition priming, should be observed for grammatical compared to ungrammatical stimuli. We conducted an event-related fMRI study to investigate neuronal correlates of such fragment-based priming. In a study phase, participants performed a short-term memory task on a series of strings of pseudofont characters. Scanning was performed in a subsequent test phase in which participants classified new strings as either grammatical or ungrammatical. Test strings differed systematically from training strings in terms of exemplar and fragment similarity. Behaviourally, participants classified strings as grammatical based on fragment familiarity. Differential activity was evident during string classification as reduced activity in left lateral occipital complex and bilateral lingual gyri for strings with high fragment familiarity compared to strings with low fragment familiarity. Thus, consistent with the hypothesis, neuronal facilitation in extrastriate occipital regions may constitute one basis of implicit grammaticality decisions based on fragment priming.

Responses of human frontal cortex to surprising events are predicted by formal associative learning theory.

Learning depends on surprise and is not engendered by predictable occurrences. In this functional magnetic resonance imaging (fMRI) study of causal associative learning, we show that dorsolateral prefrontal cortex (DLPFC) is associated specifically with the adjustment of inferential learning on the basis of unpredictability. At the outset, when all associations were unpredictable, DLPFC activation was maximal. This response attenuated with learning but, subsequently, activation here was evoked by surprise violations of the learned association. Furthermore, the magnitude of DLPFC response to a surprise event was sensitive to the relationship that had been learned and was predictive of subsequent behavioral change. In short, the physiological response properties of right DLPFC satisfied specific predictions made by associative learning theory.

Amnesia and the declarative/nondeclarative distinction: a recurrent network model of classification, recognition, and repetition priming.

A key claim of current theoretical analyses of the memory impairments associated with amnesia is that certain distinct forms of learning and memory are spared. Supporting this claim, B. J. Knowlton and L. R. Squire found that amnesic patients and controls were indistinguishable in their ability to learn about and classify strings of letters generated from a finite-state grammar, but that the amnesics were impaired at recognizing the training strings. We show, first, that this pattern of results is predicted by a single-system connectionist model of artificial grammar learning (AGL) in which amnesia is simulated by a reduced learning rate. We then show in two experiments that a counterintuitive assumption of this model, that classification and recognition are functionally identical in AGL, is correct. In three further simulation studies, we demonstrate that the model also reproduces another type of dissociation, namely between recognition memory and repetition priming. We conclude that the performance of amnesic patients in memory tasks is better understood in terms of a nonselective, rather than a selective, memory deficit.

Abstractionist and processing accounts of implicit learning.

Five experiments evaluated the contributions of rule, exemplar, fragment, and episodic knowledge in artificial grammar learning using memorization versus hypothesis-testing training tasks. Strings of letters were generated from a biconditional grammar that allows different sources of responding to be unconfounded. There was no evidence that memorization led to passive abstraction of rules or encoding of whole training exemplars. Memorizers instead used explicit fragment knowledge to identify the grammatical status of test items, although this led to chance performance. Successful hypothesis-testers classified at near-perfect levels by processing training and test stimuli according to their rule structure. The results support the episodic-processing account of implicit and explicit learning.

The effect of mental practice on performance in a sequential reaction time task.

The effect of mental practice on performance in a dot-location RT task was investigated. Participants (N = 40) were required either to mentally practice, physically practice, or do no practice on an RT task in which the signals appeared in a repeating sequence. Correct mental practice, as opposed to incorrect mental practice and no practice, was predicted to have a positive (enhancing) effect on performance of the RT task. Despite previous evidence that mental rehearsal does enhance performance in many perceptual-motor tasks, neither correct nor incorrect mental rehearsal affected subsequent sequence learning; that is, no mental practice effect was observed. That surprising result is discussed in terms of motivational, psychoneuromuscular, separate memory systems, and transfer-appropriate processing explanations of mental practice.

Is causal induction based on causal power? Critique of Cheng (1997).

The authors empirically evaluate P. W. Cheng's (1997) power PC theory of causal induction. They reanalyze some published data taken to support the theory and show instead that the data are at variance with it. Then, they report 6 experiments in which participants evaluated the causal relationship between a fictitious chemical and DNA mutations. The power PC theory assumes that participants' estimates are based on the causal power p of a potential cause, where p is the contingency between the cause and the effect normalized by the base rate of the effect. Three of the experiments used a procedure in which causal information was presented trial by trial. For these experiments, the power PC theory was contrasted with the predictions of the probabilistic contrast model and the Rescorla-Wagner theory. For the remaining 3 experiments, a summary presentation format was employed to which only the probabilistic contrast model and the power PC theory are applicable. The power PC theory was unequivocally contradicted by the results obtained in these experiments, whereas the other 2 theories proved to be satisfactory.

Evaluating the relationship between explicit and implicit knowledge in a sequential reaction time task.

Can knowledge underlying a simple perceptual-motor skill be unconscious? Three experiments (a) trained participants on a 4-choice reaction time (RT) task in which the stimulus on each trial was determined by a repeating 12-element sequence and (b) studied the extent to which participants' knowledge of this sequence was implicit, that is, unavailable for conscious access. Participants proved via an indirect test to have acquired knowledge of the sequence, because their RTs increased when the sequence was changed. To evaluate whether this knowledge was consciously accessible, participants were asked to perform an "objective" free-generation or recognition test of sequence knowledge. Results show that sequence knowledge is fully accessible on these objective tests. Moreover, it is demonstrated in this procedure that old-new recognition cannot be explained by unconscious attribution of perceptual-motor fluency. The question is raised whether distinct implicit (procedural) and explicit (declarative) forms of knowledge are acquired when participants learn a perceptual-motor skill.

Resistance to interference in human associative learning: evidence of configural processing.

In 4 experiments the authors used 2-stage designs to study susceptibility to interference in human discrimination learning. The experiments used a food allergy task. In Experiment 1, participants were presented with a discrimination in Stage 1 in which Food A predicted an allergy outcome (A-->O). In Stage 2, when combined with Food B, Food A predicted the absence of the allergy (B-->O, AB-->no O). In the test phase, Food A was found to have retained its Stage 1 association with the allergy despite the potentially interfering Stage 2 trials. In Experiment 2, a discrimination between 2 compounds (AB-->O, CD-->no O) remained intact despite subsequent complete reevaluation of the elements, (A-->no O, B-->no O, C-->O, D-->O); in Experiments 3 and 4, a discrimination between 2 pairs of elements (A-->O, B-->O, C-->no O, D-->no O) remained intact despite subsequent complete reevaluation of the AB and CD compounds, (AB-->no O, CD-->O). These experiments yielded evidence of remarkable resistance to interference in human discrimination learning. The results are at variance with the predictions of J. M. Pearce's (1987, 1994a) configural theory of associative learning.

Abstraction processes in artificial grammar learning.

Four experiments explored the extent the extent to which abstract knowledge may underlie subjects' performance when asked to judge the grammaticality of letter strings generated from an artificial grammar. In Experiment 1 and 2 subjects studied grammatical strings instantiated with one set of letters and were then tested on grammatical and ungrammatical strings formed either from the same or a changed letter-set. Even with a change of letter-set, subjects were found to be sensitive to a variety of violation of the grammar. In Experiments 3 and 4, the critical manipulation involved the way in which the training strings were studied: an incidental learning procedure was used for some subjects, and others engaged in an explicit code-breaking task to try to learn the rules of the grammar. When strings were generated from a biconditional (Experiment 4) but not from a standard finite-state grammar (Experiment 3), grammaticality judgements for test strings were independent of their surface similarity to specific studied strings. Overall, the results suggest that transfer in this simple memory task is mediated at least to some extent by abstract knowledge.

Causal order does not affect cue selection in human associative learning.

Waldmann and Holyoak (1992) presented evidence in support of the claim that cue selection does not emerge in "diagnostic" human learning tasks in which the cues are interpretable as effects and the outcomes as the causes of those effects. Waldmann and Holyoak argued that this evidence presents a major difficulty for associationist theories of learning and instead supports a "causal model" theory. We identify a number of flaws in Waldmann and Holyoak's experimental procedures and report three new experiments designed to test their claim. In Experiment 1, cue selection was observed regardless of causal order and regardless of whether the cues were abstractly or concretely specified. In Experiments 2 and 3, cue selection was again observed when subjects predicted causes from effects. We conclude that our results are consistent with simple associationist theories of learning but contradict Waldmann and Holyoak's causal model theory.

Is human learning rational?

We can predict and control events in the world via associative learning. Such learning is rational if we come to believe that an associative relationship exists between a pair of events only when it truly does. The statistical metric delta P, the difference between the probability of an outcome event in the presence of the predictor and its probability in the absence of the predictor tells us when and to what extent events are indeed related. Contrary to what is often claimed, humans' associative judgements compare very favourably with the delta P metric, even in situations where multiple predictive cues are in competition for association with the outcome. How do humans achieve this judgmental accuracy? I argue that it is not via the application of an explicit mental version of the delta P rule. Instead, accurate judgements are an emergent property of an associationist learning process of the sort that has become common in adaptive network models of cognition. Such an associationist mechanism is the "means" to a normative or statistical "end".

Human instrumental learning: a critical review of data and theory.

This article reviews experimental data from human instrumental learning tasks in which people acquire knowledge about the consequences of their actions. The main part of the paper examines the stimulus conditions which appear to control the acquisition of instrumental knowledge. These conditions include contiguity between the action and outcome, the degree of contingency between them, and also the extent to which the action is a good relative predictor of the outcome. Several accounts are examined of the mechanism by which instrumental knowledge might be acquired, including: (i) a variety of rule-based models, in which learning consists of the acquisition of knowledge about statistical relationships between contingent events; (ii) a relative contiguity model, in which learning involves the acquisition of knowledge about temporal relationships; and (iii) an associative model, in which learning involves the formation of mental associations which are updated by a learning rule. The review indicates that at present, the data seem most consistent with the associative learning model. However, there remain empirical phenomena which have resisted theoretical analysis. A variety of questions which future research might profitably explore are considered.

On the existence of independent explicit and implicit learning systems: an examination of some evidence.

Hayes and Broadbent (1988), among others, have proposed that there are two independent systems of learning in humans. In their theory, they postulate that the explicit learning system gives rise to knowledge that is verbally describable, whereas the implicit learning system deals with knowledge that is not amenable to verbal report. Hayes and Broadbent presented data that they claimed supports a distinction between these two systems of learning. Their experiments involved the performance of subjects on two superficially similar tasks, and their claim was that one of the tasks induced learning in the explicit system, and the other induced learning in the implicit system. However, in five experiments here, we failed to find any convincing support for their empirical or theoretical claims. Our results suggest that the two tasks do not differ by inducing different types of learning, but simply differ in their degree of difficulty.

Evidence for a distinction between judged and perceived causality.

Two experiments investigated Michotte's launch event, in which successive motion of two objects appears to evoke an immediate perception that the first motion caused the second, as in a collision. Launching was embedded in event sequences where a third event (a colour change of the second object) was established as a competing predictor of the second motion, in an attempt to see whether subjects' learning of alternative predictive relationships would influence their causal impressions of launch events. In Experiment 1 subjects saw launch events in which temporal contiguity at the point of impact was varied so that an impact itself did not reliably predict when the second object would move. Half of these scenes, however, contained a colour change of the second object which did reliably predict when it would move. In accordance with Michotte's theory, subjects' ratings of the degree of perceived causality were not affected by the colour change. In Experiment 2 subjects saw scenes that contained launch events with or without temporal contiguity and a colour change. These were interspersed with events in which a colour change alone did or did not precede the second motion. Thus, movement of the second object was either contingent on or independent of the impact. Subjects repeatedly (a) rated perceived causality in single launch events and (b) judged the necessity of collisions for movement in the overall set of events. These responses dissociated, in that ratings of type (a) showed only a substantial contiguity effect, whereas judgements of type (b) showed both a contingency and a much smaller contiguity effect. These results appear to support a distinction between judged and perceived causality and are discussed with respect to Michotte's theory of direct causal perception.

Instrumental judgment and performance under variations in action-outcome contingency and contiguity.

The intentional theory of instrumental performance proposes that performance of an action is determined in part by a belief about its causal effectiveness in producing a desired outcome. At variance with this notion, previous implicit learning experiments appear to have yielded dissociations between subjects' performance and beliefs. In two experiments, subjects were given an opportunity to perform an action--pressing a key on a computer keyboard--which was associated with an outcome on the computer screen according to a free-operant contingency. The subjects in one group were asked to judge the effectiveness of the action in causing the outcome, while those in a second group were asked to maximize their points score under a payoff schedule. In the first study, the effect of varying the contingency between the action and outcome was examined by keeping the probability of an outcome contiguous with an action constant and varying the probability of an outcome in the absence of an action. Performance and judgments showed a comparable sensitivity to variations of the instrumental contingency. In the second study, the delay between the action and the resultant outcome was varied. Increasing the action-outcome delay from 0 sec up to 4 sec produced a systematic decline in both causal judgments and performance relative to noncontingent, control conditions. These results are in accord with the intentional theory of performance, but they present difficulties for the notion of implicit learning.

Connectionism and the learning of probabilistic concepts.

Three experiments examine the claim of Gluck and Bower (1986, 1988a, 1988b) that the learning of medical concepts can be simulated by a connectionist network in which the symptoms are the input and the diagnosis is the output. The first experiment replicates the main finding of Gluck and Bower. In this experiment, subjects were required to estimate the probability of each of two diseases, given a particular target symptom. In fact these two probabilities were identical, but because one illness was more common than the other, the target symptom was a better predictor of the rare disease than of the common disease. Contrary to a normative probability judgement account, subjects were biased in that they judged the probability of the rare disease given the target symptom to be greater than the probability of the common disease given the target symptom. Gluck and Bower argued that such a result was predicted by a connectionist network using the Rescorla-Wagner learning rule, but it is argued that Gluck and Bower's network simulation was not appropriate for the experiment they had performed. In fact, it appeared that the connectionist network failed to predict the bias in the subjects' probability estimates. However, this conclusion rests on an assumption that Gluck and Bower implicitly made. They arranged for P(rare disease/target symptom) and P(common disease/target symptom) to be identical across all trials on which the target symptom occurred, both on its own and with other symptoms present. Gluck and Bower assumed that the subjects were estimating these probabilities. But the results of the second experiment showed instead that the subjects were estimating the probability of each disease given only the target symptom. In the final experiment the design was changed so that this problem might be circumvented. In this experiment, again, the subjects were biased in their probability judgements exactly as the connectionist network predicted. Thus, finally, evidence was found which was compatible with the network model but not with a normative account, but this was true only if the network did not include a layer of hidden units.

On the cognitive theory of conditioning .

Furedy (1988) has advanced several arguments which, he considers, produce difficulties for a cognitive theory of Pavlovian conditioning. In this paper I first discuss some of the reasons why a cognitive theory must be couched in the language of intentionality. I then argue that it is extremely difficult to see how a cognitive or intentional theory of Pavlovian conditioning can be constructed. Instead, the evidence points to the view that Pavlovian conditioning consists of the learning of excitatory or inhibitory links between the mental images of the conditioned and unconditioned stimuli. Finally, Furedy introduces considerable confusion into his account by failing to preserve an absolutely crucial distinction: between the mechanism of learning and the nature of the knowledge that is acquired during learning.

Selectional processes in causality judgment.

Two experiments illustrate the way in which competition between potential causes occurs when subjects are asked to judge the extent to which an action is the cause of an outcome. In the first experiment, it was found that introducing occurrences of the outcome in the absence of the action reduced causality judgments, but this effect was attenuated if these outcomes were signaled by another stimulus. In the second experiment, a delay between the action and the outcome reduced judgments, but this could be abolished by inserting a stimulus between the action and the outcome. The results are discussed in terms of a view of causality judgment that assumes that such judgments are based on associations between the mental representations of the action and the outcome.