ArticlePublisher preview available

Nonreactive Testing: Evaluating the Effect of Withholding Feedback in Predictive Learning

American Psychological Association
Journal of Experimental Psychology: Animal Learning and Cognition
Authors:
Jessica C Lee at The University of Sydney
Jessica C Lee
Mike E. Le Pelley
Mike E. Le Pelley
Mike E. Le Pelley
  • This person is not on ResearchGate, or hasn't claimed this research yet.
Peter F. Lovibond
Peter F. Lovibond
Peter F. Lovibond
  • This person is not on ResearchGate, or hasn't claimed this research yet.
Read publisher preview
Download citation
To read the full-text of this research, you can request a copy directly from the authors.

Abstract

Learning of cue-outcome relationships in associative learning experiments is often assessed by presenting cues without feedback about the outcome and informing participants to expect no outcomes to occur. The rationale is that this "no-feedback" testing procedure prevents new learning during testing that might contaminate the later test trials. We tested this assumption in 4 predictive learning experiments where participants were tasked with learning which foods (cues) were causing allergic reactions (the outcome) in a fictitious patient. We found that withholding feedback in a block of trials had no effect on causal ratings (Experiments 1 and 2), but it led to regression toward intermediate ratings when the missing feedback was embedded in the causal scenario and information about the outcome replaced by a "?" (Experiment 3). A factorial experiment manipulating cover story and feedback revealed that the regression-to-baseline effect was primarily driven by presentation of the "?" feedback (Experiment 4). We conclude that the procedure of testing without feedback, used widely in studies of human cognition, is an appropriate way of assessing learning, as long as the missing data are attributed to the experimenter and the absence of feedback is not highlighted in a way that induces uncertainty. (PsycInfo Database Record (c) 2021 APA, all rights reserved).
A preview of this full-text is provided by American Psychological Association.
Learn more
Content available from Journal of Experimental Psychology: Animal Learning and Cognition 
This content is subject to copyright. Terms and conditions apply.
Nonreactive Testing: Evaluating the Effect of Withholding Feedback in
Predictive Learning
Jessica C. Lee, Mike E. Le Pelley, and Peter F. Lovibond
School of Psychology, University of New South Wales
Learning of cue-outcome relationships in associative learning experiments is often assessed by present-
ing cues without feedback about the outcome and informing participants to expect no outcomes to
occur. The rationale is that this “no-feedback”testing procedure prevents new learning during testing
that might contaminate the later test trials. We tested this assumption in 4 predictive learning experi-
ments where participants were tasked with learning which foods (cues) were causing allergic reactions
(the outcome) in a fictitious patient. We found that withholding feedback in a block of trials had no
effect on causal ratings (Experiments 1 and 2), but it led to regression toward intermediate ratings when
the missing feedback was embedded in the causal scenario and information about the outcome replaced
by a “?”(Experiment 3). A factorial experiment manipulating cover story and feedback revealed that
the regression-to-baseline effect was primarily driven by presentation of the “?”feedback (Experiment
4). We conclude that the procedure of testing without feedback, used widely in studies of human cogni-
tion, is an appropriate way of assessing learning, as long as the missing data are attributed to the experi-
menter and the absence of feedback is not highlighted in a way that induces uncertainty.
Keywords: associative learning, predictive learning, feedback, extinction, methodology
Supplemental materials: https://doi.org/10.1037/xan0000311.supp
Studies of human associative learning phenomena often require
separation of training and testing. For example, in a predictive
learning task such as the classic “allergist”task (Wasserman,
1990), participants must learn what foods (cues) cause allergic
reactions (outcomes) in a fictitious patient. During training, partic-
ipants are presented with various foods and make predictions
about the expected outcome (e.g., allergic reaction or no allergic
reaction). Feedback about the actual outcome is then provided to
help participants learn. In the case of conditioning procedures,
feedback consists of presentations of biologically salient uncondi-
tioned stimuli (US; e.g., electric shocks, also termed “reinforcers”)
which are paired with conditioned stimuli (CS; e.g., tones). A core
principle in associative models is that learning is a function of pre-
diction error, the discrepancy between what is expected and what
occurs (e.g., Rescorla & Wagner, 1972). Pairings of cues and out-
comes (or CSs and USs) increase the strength of associations
between them (acquisition), while presentation of cues in the ab-
sence of outcomes weakens those associations (extinction). This
strengthening and weakening of associative links is thought to
underlie how humans learn about causal/predictive relationships
between cues and outcomes (Gluck & Bower, 1988;Le Pelley et
al., 2017;Shanks & Dickinson, 1987).
In order to prevent further learning (strengthening of associa-
tions) from occurring during testing, it is common in conditioning
studies to test in the absence of any outcomes (i.e., testing under
extinction). This would mean presenting the to-be-tested CSs with
no USs and observing responding. One issue with this procedure is
that extinction may have differential effects on the test stimuli.
Associative models (e.g., Rescorla & Wagner, 1972) predict that
the amount of extinction will increase over test trials, and that cues
that start with higher associative strength will have more negative
prediction error and therefore undergo more extinction. Further,
there is evidence showing that the rate of extinction depends on the
training history of the cue (e.g., partially reinforced cues extinguish
slower than continuously reinforced cues, e.g., Humphreys, 1939).
Some studies have attempted to remedy this problem of testing
under extinction by administering intermittent reinforcement to pre-
viously reinforced cues (CSþ)attest(e.g.,Dunsmoor & LaBar,
2013). However, this procedure arguably trains a new discrimina-
tion: that the CSþis reinforced and other stimuli are not (Honig &
Urcuioli, 1981). This is particularly problematic when assessing
generalization as rather than intermittent reinforcement slowing the
rate of extinction, rapid decreases in responding might instead occur
to all stimuli except the CSþas participants learn the new discrimina-
tion. In sum, both options (presenting or not presenting the outcome)
This article was published Online First November 29, 2021.
Jessica C. Lee https://orcid.org/0000-0003-4253-2008
Mike E. Le Pelley https://orcid.org/0000-0002-5145-5502
Peter F. Lovibond https://orcid.org/0000-0003-2146-9054
This study was funded by an Australian Research Council Discovery
Grant (DP190103738) awarded to Peter F. Lovibond.
The data and materials for all experiments are available at https://osf.io/
dvn9g/. None of the experiments were preregistered.
Correspondence concerning this article should be addressed to Jessica C.
Lee, School of Psychology, University of New South Wales, Mathews
Building, Kensington, NSW 2052, Australia. Email: jessica.lee@unsw.edu
.au
17
Journal of Experimental Psychology:
Animal Learning and Cognition
©2021 American Psychological Association 2022, Vol. 48, No. 1, 17–28
ISSN: 2329-8456 https://doi.org/10.1037/xan0000311
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
... While this strategy is commonly employed in fear conditioning studies to prevent extinction of a conditioned response (e.g., skin conductance in fear conditioning), it is not commonly required in predictive learning tasks. J. C. Lee et al. (2022) have shown that testing under "no-feedback" conditions is effective in preventing extinction in predictive learning tasks where there is no motivationally significant outcome such as an electric shock (J. C. Lee et al., 2022). ...
... J. C. Lee et al. (2022) have shown that testing under "no-feedback" conditions is effective in preventing extinction in predictive learning tasks where there is no motivationally significant outcome such as an electric shock (J. C. Lee et al., 2022). Further, J. C. Lee et al. (2022) have argued that testing generalization under continuous or even intermittent reinforcement of the CS+ has the drawback of training a new discrimination: that the CS+ is reinforced and all other stimuli are not. ...
... C. Lee et al., 2022). Further, J. C. Lee et al. (2022) have argued that testing generalization under continuous or even intermittent reinforcement of the CS+ has the drawback of training a new discrimination: that the CS+ is reinforced and all other stimuli are not. This will sharpen the generalization gradient around the CS+, since this procedure is essentially conditioning of an identification judgment. ...
The Role of Perception in Generalization: Commentary on
Article
Full-text available
  • Apr 2025
Stimulus generalization, or the transfer of learned responses between stimuli, is a critical ability for adaptation to everyday life. In a typical experiment, generalization is assessed by measuring responses to stimuli varying along a physical dimension. Variations in the gradient of learned responses are usually interpreted as differences in the underlying cognitive process of generalization. A recent study by Zaman, Yu, and Verheyen (2023) seeks to challenge this view, arguing that generalization is best modeled by perceptual factors and that individual differences in perception or ability to identify the stimuli are primary drivers of generalization. In this commentary, we outline issues in the methodology and analysis of Zaman et al.’s study and show that their key result is not robust to the addition of theoretically informed alternative models. We conclude that the evidence is not strong enough to support their conclusions regarding the primacy of perceptual processes in generalization. We propose some ways forward for researchers in this field attempting to understand the psychological mechanisms underlying individual differences in stimulus generalization.
View
... Therefore, we elected to leave the associative status of stimulus D indeterminate, by presenting it without any feedback as to the presence or absence of the outcome. We have previously demonstrated that this "no feedback" procedure has little or no effect on the associative status of a stimulus (Lee, Le Pelley & Lovibond, 2022). Finally, we included a range of additional stimuli in order to ensure that participants experienced both single and compound stimuli with and without the outcome, and also to match the design of subsequent planned experiments. ...
... Participants were instructed prior to the experiment that on some trials they would not receive feedback about whether an allergic reaction had occurred or not. We have previously shown that such a procedure leaves the associative status of a cue unchanged in an extinction design (Lee et al., 2022). Instructions. ...
... We also included a Retardation of acquisition 8 brief instruction check which participants were required to pass before proceeding. For further details, please see Lee and Lovibond (2021) and Lee et al. (2022). ...
Retardation of Acquisition After Conditioned Inhibition and Latent Inhibition Training in Human Causal Learning
Article
Full-text available
  • Apr 2023
Inhibitory stimuli are slow to acquire excitatory properties when paired with the outcome in a retardation test. However, this pattern is also seen after simple non-reinforced exposure: latent inhibition. It is commonly assumed that retardation would be stronger for a conditioned inhibitor than for a latent inhibitor, but there is surprisingly little empirical evidence comparing the two in either animals or humans. Thus, retardation after inhibitory training could in principle be attributable entirely to latent inhibition. We directly compared the speed of excitatory acquisition after conditioned inhibition and matched latent inhibition training in human causal learning. Conditioned inhibition training produced stronger transfer in a summation test, but the two conditions did not differ substantially in a retardation test. We offer two explanations for this dissociation. One is that learned predictiveness attenuated the latent inhibition that otherwise would have occurred during conditioned inhibition training, so that retardation in that condition was primarily due to inhibition. The second explanation is that inhibitory learning in these experiments was hierarchical in nature, similar to negative occasion-setting. By this account, the conditioned inhibitor was able to negatively modulate the test excitor in a summation test, but was no more retarded than a latent inhibitor in its ability to form a direct association with the outcome.
View
... Test. Upon the completion of the two stages of training, additional on-screen instructions informed participants that they would continue to see meals and that they should keep making predictions about the severity of Mr. X's allergic reaction, but that they would no longer receive any feedback (for the efficacy of testing in the absence of feedback, see Lee et al., 2022). Instead, after participants made a rating by clicking on any point along the outcome prediction scale, a second rating scale to measure uncertainty appeared immediately below. ...
... A probe test was interpolated between training Stages 1 and 2. It was conducted in the same way as the test in the previous experiments: participants were asked to predict the severity of the outcome when presented with a cue, and to rate how confident they were in their outcome prediction. The transition between Stage 1 and the probe test was not signaled and no feedback was presented on the probe test trials (see Lee et al., 2022). To minimize any disruption created by the probe test, the experimental instructions (and the check of these instructions) were modified to inform participants that: (a) there may be trials on which no feedback would occur; and (b) on these trials, they should continue to make outcome prediction (severity) and confidence ratings as usual. ...
The Role of Uncertainty in Regulating Associative Change
Article
Full-text available
  • Apr 2024
Rescorla (2000, 2001) interpreted his compound test results to show that both common and individual error terms regulate associative change such that the element of a conditioned compound with the greater prediction error undergoes greater associative change than the one with the smaller prediction error. However, it has recently been suggested that uncertainty, not prediction error, is the primary determinant of associative change in people (Spicer et al., 2020, 2022). The current experiments use the compound test in a continuous outcome allergist task to assess the role of uncertainty in associative change, using two different manipulations of uncertainty: outcome uncertainty (where participants are uncertain of the level of the outcome on a particular trial) and causal uncertainty (where participants are uncertain of the contribution of the cue to the level of the outcome). We replicate Rescorla’s compound test results in the case of both associative gains (Experiment 1) and associative losses (Experiment 3) and then provide evidence for greater change to more uncertain cues in the case of associative gains (Experiments 2 and 4), but not associative losses (Experiments 3 and 5). We discuss the findings in terms of the notion of theory protection advanced by Spicer et al., and other ways of thinking about the compound test procedure, such as that proposed by Holmes et al. (2019).
View
... A third piece of evidence is that a preventative cue (i.e., a conditioned inhibitor) fails to show greater retardation of subsequent excitatory conditioning compared with a latently inhibited cue (Lovibond et al., 2023). In a series of experiments, we compared the properties of a negative feature X trained in a feature negative discrimination (A+ AX−) with an equivalent cue E presented in compound with a cue (D), which was separately presented with no feedback about the outcome (D DE−; see Lee et al., 2022, for experiments validating this no-feedback procedure). According to the Rescorla-Wagner model (Rescorla & Wagner, 1972), X should become a conditioned inhibitor, whereas E should undergo latent inhibition. ...
View
... appearing over the static in a blue font. This screen was used to help avoid carryover effects in testing (Lee et al., 2022). ...
Extinction Induced Representational Change
Article
Full-text available
  • Jan 2025
Extinction may alter the representation of a cue (e.g., it becomes less salient). To assess that idea, three groups learned to suppress mouse clicking in a video game in negative-patterning (X+/Y+/XY−) and positive-patterning (Z+/W+/ZW++) discriminations followed by extinction of X and Z. The negative-patterning discrimination should depend on a configural cue that is dependent on the representation of X and Y. Removal of the excitatory influence of X should further reduce responding to XY. In contrast, if extinction alters the representation of X, the original XY configural cue supporting the discrimination should also be changed, affecting inhibitory control, increasing responding to XY. Following patterning, groups received extinction in the same context as training (Ext A), a different context (Ext B), or received no extinction (no extinction). All stimuli were tested in Context A. Group no extinction showed negative patterning; suppression to X and Y was greater than to XY while suppression to Z, W, and ZW was equally strong. In group Ext A extinction reduced suppression to X, increased suppression to XY, reversed the X/XY discrimination, and weakened the Y/XY discrimination. Extinction of Z reduced suppression to Z with no effect on W or ZW. Group Ext B showed renewal of X and a renewal of the X/XY and Y/XY discriminations. Results suggest some form of representational change in X occurred during extinction disrupting the original XY configural cue that was dependent on that representation. Findings are discussed with respect to theories of associative learning.
View
... Feedback was not given during the test phase which should minimize extinction occurring on test (J. Lee et al., 2022). Instead, participants passed to the next trial each time they made a prediction during this phase. ...
Perceived stress and renewal: The effects of long-term stress on the renewal effect
Article
Full-text available
  • Apr 2024
  • NEUROBIOL LEARN MEM
Two online experiments evaluated the relationship between long-term stress, as measured with the Perceived Stress Scale-10, and the Renewal Effect. In the first experiment renewal was assessed with a behavioral suppression task in a science-fiction based video game. Participants learned to suppress mouse clicking during a signal for an upcoming attack to avoid losing points. The signal was first paired with an attack in Context A and extinguished in Context B and tested back in Context A. The contexts were different space galaxies where the gameplay took place. Experiment 2 used a food/illness predictive-learning paradigm. Two food items were paired with stomachache in one restaurant (A) and extinguished in Context B prior to testing in both contexts without feedback. Positive correlations were obtained between renewal and stress in each experiment. Unlike acute stress (Drexler et al., 2017), long term stress was associated with greater renewal. The effects of stress, both chronic and punctual, on renewal are discussed.
View
... Participants experienced seven stimuli in a random order without feedback about the outcomes. Note that this procedure does not result in extinction from testing several stimuli in the absence of outcome (see Lee et al., 2022). On each test trial, a stimulus appeared with the question: "The [symbol/ word] above appears on the machine. ...
The impact of emotional valence on generalization gradients
Article
Full-text available
  • Jan 2024
Generalization enables individuals to respond to novel stimuli based on previous experiences. The degree to which organisms respond is determined by their physical resemblance to the original conditioned stimulus (CS+), with a stronger response elicited by more similar stimuli, resulting in similarity-based generalization gradients. Recent research showed that cognitive or conceptual dimensions also result in gradients similar to those observed with manipulations of physical dimensions. Such findings suggest that attributes beyond physical similarity play a role in shaping generalization gradients. However, despite its adaptive relevance for survival, there is no study exploring the effectiveness of affective dimensions in shaping generalization gradients. In two experiments (135 Spanish and 150 English participants, respectively), we used an online predictive learning task, in which different stimuli (words and Gabor patches) were paired with the presence – or absence – of a fictitious shock. After training, we assessed whether valence (i.e., hedonic experience) conveyed by words shape generalization gradients. In Experiment 1, the outcome expectancy decreased monotonically with variations in valence of Spanish words, mirroring the gradient obtained with the physical dimension (line orientation). In Experiment 2, conducted with English words, a similar gradient was observed when non-trained (i.e., generalization) words varied along the valence dimension, but not when words were of neutral valence. The consistency of these findings across two different languages strengthens the reliability and validity of the affective dimension as a determinant of generalization gradients. Furthermore, our data highlight the importance of considering the role of affective features in generalization responses, advancing the interplay between emotion, language, and learning.
View
... It should be noted that the method of rule assessment did not allow Lovibond et al. (2020) to assess when participants formed rules (i.e., during training or test). It is possible that despite the no-feedback procedure of testing being adequate in preventing extinction (see Lee et al., 2022), participants may still learn about the relational features of the stimuli and continue to learn or update rules at test (see Livesey & McLaren, 2009). ...
Individual Differences in Stimulus Identification, Rule Induction, and Generalization of Learning
Article
Full-text available
  • Aug 2022
In the field of stimulus generalization, an old yet unresolved discussion pertains to what extent stimulus misidentifications contribute to the pattern of conditioned responding. In this article, we perform cluster analysis on six datasets (four published datasets and two unpublished datasets, included N = 950) to examine the relationship between interindividual differences in (a) stimulus identification, (b) patterns of generalized responding, and (c) verbalized generalization rules. The datasets were obtained from online predictive learning tasks where participants learned associations between colored cues and the presence or absence of a hypothetical outcome. In these datasets, stimulus identification and expectancy ratings were assessed in separate phases to a range of colors varying between blue-green. Using cluster analyses on performance during stimulus identification, we identified different subgroups of participants (good vs. bad identifiers). In all six datasets, we found a close relationship between the pattern of stimulus identification and the shape of the expectancy gradient across the test dimension between the identified subgroups. Furthermore, participants classified as good identifiers were more likely to report a similarity generalization rule than a relational or linear rule, suggesting that individual differences in stimulus identification are related to individual differences in generalization rules. These findings suggest that greater consideration should be given to interindividual variability in stimulus identification, inductive rules, and their relationship in explaining patterns of generalized responses. (PsycInfo Database Record (c) 2022 APA, all rights reserved).
View
Using Neuroeconomics to study Emotional Valence at Icon Metal Marketing Pvt. Ltd
Article
Full-text available
  • Jan 2025
This study delves into the complexities of emotional advertising effectiveness in industrial markets through the innovative use of advanced neuroeconomic models, machine learning algorithms, and network analysis techniques. The primary objective was to explore how emotional responses to advertisements influence brand perceptions, purchase intentions, and long-term business relationships among industrial decision-makers. Utilizing functional Magnetic Resonance Imaging (fMRI), Electroencephalography (EEG), and biometric sensors, we captured and analyzed real-time emotional and cognitive reactions to advertisements from Icon Metal Marketing Pvt. Ltd. Our neuroeconomic analysis revealed significant neural activations in the amygdala and prefrontal cortex, elucidating the neural underpinnings of emotional engagement. Concurrently, machine learning algorithms were instrumental in processing and interpreting the extensive neuroimaging and biometric data, enabling the prediction of emotional engagement levels and the optimization of advertising content. Furthermore, our network analysis provided a detailed visualization of emotional contagion effects across industrial buyer networks. By identifying key influencers and mapping emotional pathways, we gained insights into how emotional engagement amplifies and shapes collective brand perceptions. This approach allowed for the strategic deployment of emotionally resonant content, enhancing the impact of advertising campaigns across diverse industrial sectors. The findings suggest that precision targeting and personalization of advertising content, informed by neural and biometric data, can significantly enhance emotional resonance and brand perception. Additionally, understanding network dynamics and emotional contagion effects offers a powerful framework for optimizing advertising strategies and fostering long-term business relationships.
View
Action Slips in Food Choices: A Measure of Habits and Goal-Directed Control
Article
Full-text available
  • Jan 2023
  • LEARN BEHAV
We report a new, simple instrumental action-slip task, which sets goal-directed action against putative S-R associations. On each training trial, participants were presented with one of two stimuli (blue or green coloured screen). One stimulus (S1) signalled that one joystick response (R1 – left or right push) would earn one of two rewards (O1 – jellybeans or Pringles points). A second stimulus (S2) signalled a different instrumental relationship (S2:R2-O2). On each test trial, participants were told which outcome could be earnt (O1/O2) on that trial. They were required to withhold responding until the screen changed colour to S1 or S2. On congruent test trials, the stimulus presented (e.g., S1) was associated with the same response (R1) as the outcome available on that trial (O1). On incongruent test trials, in contrast, the outcome (e.g., O1) preceded a stimulus that was associated with a different response (e.g., S2). Hence, in order to obtain the outcome (O1) on incongruent trials, participants were required to supress any tendency they might have to make the response associated with the stimulus (R2 in response to S2). In two experiments, participants made more errors on incongruent than congruent trials. This result suggests that, on incongruent trials, the stimulus drove responding (e.g., S2 increased R2 responding) in a manner that was inconsistent with goal-directed action (e.g., R1 responding to obtain O1) – an action slip. The results are discussed in terms of popular dual-process theories of instrumental action and a single-process alternative.
View
Individual differences in causal structures inferred during feature negative learning
Article
Full-text available
  • Jan 2021
Traditional associative learning theories predict that training with feature negative (A+/AB-) contingencies leads to the feature B acquiring negative associative strength and becoming a conditioned inhibitor (i.e., prevention learning). However, feature negative training can sometimes result in negative occasion setting, where B modulates the effect of A. Other studies suggest that participants learn about configurations of cues rather than their individual elements. In this study, we administered simultaneous feature negative training to participants in an allergist causal learning task and tested whether evidence for these three types of learning (prevention, modulation, configural) could be captured via self-report in the absence of any procedural manipulation. Across two experiments, we show that only a small subset of participants endorse the prevention option, suggesting that traditional associative models that predict conditioned inhibition do not completely capture how humans learn about negative contingencies. We also show that the degree of transfer in a summation test corresponds to the implied causal structure underlying conditioned inhibition, occasion-setting, and configural learning, and that participants are only partially sensitive to explicit hints about causal structure. We conclude that feature negative training is an ambiguous causal scenario that reveals individual differences in the representation of inhibitory associations, potentially explaining the modest group-level inhibitory effects often found in humans.
View
The Role of Common Elements in the Redundancy Effect
Article
Full-text available
  • Jul 2020
In 2 experiments, participants received a predictive learning task in which the presence of 1 or 2 food items signaled the onset or absence of stomachache in a hypothetical patient. Their task was to identify the cues that signaled the occurrence, or nonoccurrence of this ailment. The 2 groups in Experiment 1 and the single group in Experiment 2 received a blocking treatment, where Cue A and a combination of Cues A and X both signaled stomachache, A+ AX+. These groups also received a simple discrimination where the outcome was signaled by one compound but not another, BY+ CY-. Subsequent test trials revealed the so-called redundancy effect, where X was regarded as a more reliable predictor of the outcome than Y. This result occurred when the trials with A+ preceded those with AX+ (Group E, Experiment 1 and Experiment 2), and when the trials with A+ and AX+ were intermixed (Group C, Experiment 1). The results challenge theories based on the assumption that cues presented together must compete for a limited pool of associative strength. Rather, they are said to support theories that assume changes in attention determine what is learned when two or more cues are presented together. (PsycInfo Database Record (c) 2020 APA, all rights reserved).
View
The Role of Prediction in Learned Predictiveness
Preprint
Full-text available
  • May 2020
Learning permits even relatively uninteresting stimuli to capture attention if they are established as predictors of important outcomes. Associative theories explain this “learned predictiveness” effect by positing that attention is a function of the relative strength of the association between stimuli and outcomes. In three experiments we show that this explanation is incomplete: learned overt visual-attention is not a function of the relative strength of the association between stimuli and an outcome. In three experiments, human participants were exposed to triplets of stimuli that comprised (i) a target (which defined correct responding), (ii) a stimulus which was perfectly correlated with the presentation of the target and (iii) a stimulus which was uncorrelated with the presentation of the target. Participants’ knowledge of the associative relationship between the correlated/uncorrelated stimuli and the target was always good. However, eye-tracking revealed that an attentional bias towards the correlated stimulus only developed when it AND target-relevant responding preceded the target stimulus. We propose a framework in which attentional changes are modulated during learning as a function the relative strength of the association between stimuli and the task-relevant response, rather than an association between stimuli and the task-relevant outcome.
View
Transfer of Associability and Relational Structure in Human Associative Learning
Article
Full-text available
  • Feb 2019
A wealth of recent studies have demonstrated that predictive cues involved in a linearly solvable component discrimination gain associability in subsequent learning relative to non-predictive cues. In contrast, contradictory findings have been reported about the fate of cues involved in learning biconditional discriminations in which the cues are relevant but none are individually predictive of a specific outcome. In three experiments we examined the transfer of learning from component and biconditional discriminations in a within-subjects design. The results show a greater benefit in associability for cues that had previously served as predictive cues in a component discrimination than cues previously used in a biconditional discrimination. Further, new biconditional discriminations were learned faster when they were composed of cues that were previously trained in separate biconditional discriminations. Similarly, new component discriminations were learned faster when they were composed of cues that were previously trained in a separate component discriminations irrespective of whether they were previously predictive or previously non-predictive. These results provide novel evidence that cue-specific learning of relational structure affects subsequent learning, suggesting changes in cue processing that go beyond simple changes in cue associability based on learned predictiveness.
View
Negative Evidence and Inductive Reasoning in Generalization of Associative Learning
Article
Full-text available
  • Nov 2018
When generalizing properties from known to novel instances, both positive evidence (instances known to possess a property) and negative evidence (instances known not to possess a property) must be integrated. The current study compared generalization based on positive evidence alone against a mixture of positive evidence and perceptually dissimilar negative evidence in an interdimensional discrimination procedure. In two experiments, we compared generalization following training with a single positive stimulus (that predicted shock) against groups where an additional negative stimulus (that did not predict shock) was presented in a causal judgement (Experiment 1) and a fear conditioning (Experiment 2) procedure. In contrast to animal conditioning studies, we found that adding a “distant” negative stimulus resulted in an overall increase in generalization to stimuli varying on the dimension of the positive stimulus, consistent with the inductive reasoning literature. We show that this key qualitative result can be simulated by a Bayesian model that incorporates helpful sampling assumptions. Our results suggest that similar processes underlie generalization in inductive reasoning and associative learning tasks.
View
Peak Shift and Rules in Human Generalization
Article
Full-text available
  • Mar 2018
Two experiments tested whether a peak-shifted generalization gradient could be explained by the averaging of distinct gradients displayed in subgroups reporting different generalization rules. Across experiments using a causal judgment task (Experiment 1) and a fear conditioning paradigm (Experiment 2), we found a close concordance between self-reported rules and generalization gradients using a continuous stimulus dimension (hue). Both experiments also showed an overall peak-shifted gradient after differential conditioning, but not after single cue conditioning. Importantly, the peak shift could be decomposed into linear and peaked gradients when participants were divided into rule subgroups. Our results highlight the need to consider individual differences in the rules that participants derive in human generalization studies, and suggest that in some situations, peak shift may be a consequence of averaging across diverse rule subgroups.
View
Goal-Directed and Habit-Like Modulations of Stimulus Processing during Reinforcement Learning
Article
Full-text available
  • Feb 2017
Recent research has shown that perceptual processing of stimuli previously associated with high-value rewards is automatically prioritized even when rewards are no longer available. It has been hypothesized that such reward-related modulation of stimulus salience is conceptually similar to an “attentional habit.” Recording event-related potentials in humans during a reinforcement learning task, we show strong evidence in favor of this hypothesis. Resistance to outcome devaluation (the defining feature of a habit) was shown by the stimulus-locked P1 component, reflecting activity in the extrastriate visual cortex. Analysis at longer latencies revealed a positive component (corresponding to the P3b, from 550–700 ms) sensitive to outcome devaluation. Therefore, distinct spatiotemporal patterns of brain activity were observed corresponding to habitual and goal-directed processes. These results demonstrate that reinforcement learning engages both attentional habits and goal-directed processes in parallel. Consequences for brain and computational models of reinforcement learning are discussed. SIGNIFICANCE STATEMENT The human attentional network adapts to detect stimuli that predict important rewards. A recent hypothesis suggests that the visual cortex automatically prioritizes reward-related stimuli, driven by cached representations of reward value; that is, stimulus–response habits. Alternatively, the neural system may track the current value of the predicted outcome. Our results demonstrate for the first time that visual cortex activity is increased for reward-related stimuli even when the rewarding event is temporarily devalued. In contrast, longer-latency brain activity was specifically sensitive to transient changes in reward value. Therefore, we show that both habit-like attention and goal-directed processes occur in the same learning episode at different latencies. This result has important consequences for computational models of reinforcement learning.
View
Rule-based generalisation in single-cue and differential fear conditioning in humans
Article
  • Sep 2017
  • BIOL PSYCHOL
Fear generalisation refers to the spread of conditioned fear to stimuli similar but distinct from the original conditioned stimulus. In this study, participants were presented with repeated pairings of a conditioned stimulus with a shock, in either a single-cue or differential conditioning paradigm. Generalisation of fear was then tested by presenting stimuli that were novel, but similar to the conditioned stimulus along a spatial stimulus dimension. Dependent measures were online shock expectancy ratings and skin conductance level. A diverse range of generalisation gradients was observed, and the shape of the gradients for both expectancy ratings and skin conductance responses corresponded with participants' verbally reported rules. The findings point to an important role for cognitively controlled processes in human fear generalisation, and provide support for a single-system learning model. They also highlight the potential importance of cognitive reappraisal in clinical treatments for over-generalised fear.
View
View
PSYCHOLOGICAL SCIENCE Research Report ATTRIBUTION OF CAUSALITY TO COMMON AND DISTINCTIVE ELEMENTS OE COMPOUND STIMULI
Article
  • Sep 1990
Ciiven the task of di the source of a patient's aUer^'ic reav-tion. college students jiuigcii the causal efficacy of common (A') and distinctive (A and Bj elements of compound stimuli: AX and BX. As the differential correlation of AX and BX with the occurrence and nonoccurrence ofthe allergic reaction rose from .00 to 1.00. ratings of ihe distinctive A and B elements diverged; most importantly, ratings ofthe common X element fell. These causal judgments of humans closely parallel the conditioned responses of animals in associa-tive learning studies, and clearly disclose that stimuli compete with one another for control over behavior.
View