| We are living in a complex and changing environment.When facing these changes,the human beings need to constantly learn and adjust their behaviors to reduce the uncertainty about environment as well as future events,maintaining the stability of internal and external world.Actually,the human beings have an instinctive aversion about the uncertainty,and how to eliminate these uncertainties and deal with the future events in a more adaptive way will directly affect the individual’s survival and development.For human beings,in addition to the same external environmental factors those have to adapt to as most animals do,such as when weather or season changes,people need to timely make corresponding adjustments,but also including adapting to a variety of complex social factors.This adaptive ability is also called as social ability or social adaptability,which refers to a person’s ability to deal with daily life and survive in the social environment.Adaptive behavior also reflects the human’s instinct to avoid conflict,that is,the conflict between the actual outcome(or event)and the individual’s expectation.No matter in works,study or life,the human beings,as organisms with particular adaptability,are usually able to make an advance expectation for the upcoming events and form a psychological preparation,rather than passively and unprepared to accept any changes.When the actual outcome conflicts with people’s expectation,they are able to make corresponding adjustment to reduce or avoid the same conflict.In the laboratory,the learning paradigms,such as probabilistic learning task and associative learning task,are usually used to study individual’s adaptability.This adaptive learning process includes two stages: prediction stage and feedback stage.In the face of the upcoming event,people tend to make an expectation(prediction stage)before that event occurs to form an appropriate psychological preparation;and when the actual outcome appears(feedback stage),they will evaluate the current outcome,and then based on the difference between the actual outcome and expected outcome,namely the prediction error(PE),to adjust the expectation and behavior when face the same event in the next time,so as to reduce such prediction error and adapt to the changes better.Hence,the advance prediction can form an appropriate psychological preparation for the upcoming events,while the actual feedback will promote the adjustment of prediction and behavior in the next time based on the difference between actual outcome and expected outcome.The two stages are mutually promoting and closely related,and both are crucial for adaptive learning.Researching on the prediction and feedback evaluation can provide insight into how human beings adapt to the changes.However,till now,only a few studies have focused on the formation of prediction,making us know little about the formation and change of the prediction,and it’s also unclear whether there is a difference in the prediction under different contexts,as well as the neural mechanism behinds that difference.At the same time,although a large number of studies have discussed the evaluation of feedback,especially the prediction error,there is still a great controversy about whether the evaluation of feedback reflects a discrimination about outcome from good to bad dimension(reward expected error,RPE),or just reflects the deviation of expectations(unsigned prediction error,UPE).We speculate that the feedback’s evaluation should include both the outcome’s valence and the expectation’s deviation.The reason why there always exists a controversy in previous studies is also one of the questions that we want to explore in current study.Furthermore,if the assessment of the feedback involves processing these different aspects of information,are they processed simultaneously or in different stages? If there exists different stages,what is the order of those stages? Therefore,in this paper,we have conducted three studies,in which from the expectation’s formation to the feedback’s evaluation,especially how the prediction error information in the feedback stage will adjust the behavior and the next prediction,as well as the neural mechanism behind them,are systematically investigated,setting up a comprehensive study to deeply understand and promote human’s adaptive behavior.In study 1,we explored the formation of prediction with two experiments.As we know,the prediction is primarily set up by integrating internal and external circumstances based on the past experience and current cue.However,in previous researches focusing on the outcome’s prediction,they usually provide a probabilistic cue,and then classify the prediction just according to the provided cue.But the prediction should not be invariable and is not completely dependent on the external cue.In real life,people can also adjust their prediction according to the actual outcomes.Therefore,it is more accurate to explicitly measure people’s prediction at every time.In addition,we suggest that the prediction may be diverse in different contexts.For example,one person may expect to win as much money as possible when under the reward context,but he will expect to loss as little money as possible when under the punishment context.However,few studies have directly compared the outcome’s prediction in different contexts to explore their difference.In this study,the Experiment1 was a behavioral experiment,which adopted an adapted probabilistic learning paradigm.At each time,the participants were informed of the probability(25% or 75%)about the occurrence of an event in that trial via a cue;and when that cue appeared,they were instructed to press the corresponding key button to indicate their prediction.Meanwhile,the prediction was directly compared between the gain context and loss context.As a result,we found that the participants’ prediction did closely track with the given cue;at the same time,their real prediction was apparently different from the cued probability,suggesting that their prediction was also affected by the past experience.Most importantly,under the gain and loss contexts,despite that all the aspects were the same except the outcome’s context,the participant’s prediction of winning money was significantly higher than that of losing money.To exclude the potential influence of stimulus’ s type,in Experiment 2,the monetary reward and electric pain shock were used as stimuli to further explore whether the difference observed in Experiment 1 was stable and widespread.Simultaneously,the functional magnetic resonance imaging(f MRI)approach was used to capture the participant’s brain activities when they performed the task.The same behavioral results confirmed our assumption.Our f MRI results found that the outcome’s prediction triggered a brain activity similar to that induced by the stimulus itself.Unlike that the prediction of reward had activated substantial activities in reward circuitry regions,the prediction of pain shock had also activated significant activities in control-related frontal areas.In addition,when predicted pain shock,the anterior cingulate cortex(ACC)showed a significant functional connectivity with striatum.Through the study 1,we demonstrated that the formation of prediction was not constant and should be influenced by the current cue,historical experience and outcome’s context.The prediction of an outcome would elicit a neural activity similar to that of the stimulus itself,but the neural mechanisms underlying the reward’s prediction and punishment’s prediction were different.Expecting reward was mainly involved in the reward network,while expecting punishment not only recuited part of the reward network,but also would trigger the activity of cognitive control system.In study 2,we mainly focused on the outcome’s evaluation in the feedback stage,especially the difference between actual outcome and the prediction,such as how it would regulate the behavior and prediction in the next time,as well as the neural mechanism behind it.The evaluation of feedback usually will elicit a change in an electrical component,namely the feedback-related negativity(FRN).Therefore,we had conducted three electroencephalogram(EEG)experiments in Study 2,simultaneously collecting the participants’ event-related potentials(ERP)activities when they performed task.The Experiment 3 adopted a commonly used associative learning paradigm,and the difference between our design and previous studies is that in order to directly explore how the prediction error will regulate behavior and prediction,as well as what the relationship between the prediction error and FRN is,we had varied the size of prediction(0-20)and feedback(0-20),and then asked the participant to directly report their predicted amount in the prediction phase;correspondingly,the size of prediction error(actual outcome minus expected outcome)would also be varied.At the same time,we had adopted two kinds of incentives,namely the gain and loss,to directly compare the feedback evaluation process under these two conditions.The results showed that the prediction error could effectively regulate participants’ behavior.The ERP results found that no matter in gain or loss condition,the lower than expected gain(worse than expected outcome)and loss(better than expected outcome)both had induced more negative amplitudes,which was obviously inconsistent with the view that "the feedback’s evaluation reflects a discrimination about it from good to bad dimension".The further single-trial correlation analysis on the prediction error and feedback-related potentials found that in this experiment the feedback evaluation was more sensitive to its deviation from prediction.In order to rule out the possibility that in difficult task(Experiment 3),the participants’ high attention to prediction’s deviation might weaken or obscure their attention about the outcome’s good or bad information,the Experiment 4 used the pure numbers as stimuli.Hence,in Experiment 4,there did not exist any evaluation about outcome’s valence,and participants just needed to pay attention to whether the actual number was deviated from their expected number.The similar results to the Experiment 3 confirmed our hypothesis,that is,the high concern about the predictive accuracy had indeed occupied the participants’ most attention,which thus weaken their evaluation of an outcome’s valence.Furthermore,in Experiment 5,we used the same task as that in Experiment 3,except that in half task,we explicitly told the participants that the outcome was not related to their predictive accuracy,and they only needed to choose the option that was most beneficial to maximize their final earnings;while in the other half task,the participants were told that the more accurate their prediction was,the more money they could win(in the gain condition),or the less money they would lose(in the loss condition).The results showed that when participants paid more attention to the good or bad dimension of the outcomes,the worse than expected outcomes(lower than expected gain,higher than expected loss)elicited more negative amplitudes.However,when more attention was paid to the predictive accuracy,the same result as experiments 3 and 4 was found.Through these three experiments in Study 2,we proved that the prediction error was significantly correlated with the adjustment of prediction in next trial.Meanwhile,the evaluation of feedback should include both the good or bad dimension of that outcome and its deviation from the prediction,and when the emphasized information was different,the response would be different.Our results also provide a direct evidence for understanding the controversial issue in previous studies,facilitating an in-depth knowledge of the feedback’s evaluation.In study 3,we further explored whether there were different stages in the processing of different aspects of the feedback and what was the order of each stage via two experiments.We adopted a reverse learning task paradigm that is developed in recent years.The task is slightly more difficult and is closer to the real life.In reverse learning task,the participants need to pay attention to both the outcome’s good or bad information and its deviation from the prediction to learn the rule and maximize their benefit.The Experiment 6 investigated how the participants found the rule and adjusted their prediction based on the prediction error signal when multiple objects were needed to learn at the behavioral level.The Experiment 7 used the EEG technique to further investigate the neural mechanism underlying these behaviors.Consequently,we found that when the number of learning objects was increased to three,the learning speed was significantly slower than the preceding studies,and the participants could effectively adjust their behavior and prediction based on the prediction error.The ERP results had found a significant valence evaluation on feedback,as well as the effect of prediction’s deviation and deviation’s size.The multivariate pattern analysis(MVPA)of the ERP results further proved that the evaluation of feedback outcome contained diverse aspects X of information,such as good or bad evaluation of outcome,the deviation from the prediction,and the magnitude of that deviation.The processing of different aspects of information should follow the corresponding order: the first stage was an early evaluation about the outcome’s valence(120-390 ms,peaking at 200ms);this stage was followed by an assessment of whether that outcome was deviated from the prediction(200-400 ms,peaking at 270ms);at the same time,the evaluation of the positive and negative dimension still continued,which would then integrate the prediction’s deviation information into a deeper processing(160-570 ms,peahing at 350ms);finally,the magnitude of the prediction’s deviation was integrated with the previous valence and deviation information(260-600 ms,peaking at 350ms)to form a comprehensive evaluation of the feedback outcome.Therefore,through two experiments,the study 3proved that feedback evaluation involved the processing of various information,including outcome value,the deviation from the prediction,and the magnitude of that deviation in both behavior and ERP results,and the processing of these information had different stages and followed the corresponding order: the first stage was the early value evaluation,which followed by the evaluation of whether that result was deviated from the expectation,and the value evaluation was continued to integrate the deviation information;the last stage was the processing of the magnitude of expectation deviation,which integrated with the previous information to form a complete evaulation of feedback.In summary,we had examined the entire adaptive learning process of anticipation-feedback from multimodal perspective through seven experiments across three studies,combining behavioral,f MRI,and EEG techniques.Our study had directly measured the prediction for the first time.The prediction’s formation as well as feedback’s evaluation was directly compared between reward and punishment contexts,and the magnitude of prediction error was manipulated,too.We prove that the formation of prediction requires comprehensive consideration of multiple factors,such as explicit cue,past experience and outcome’s context.When evaluating the feedback,the outcome’s good or bad information,its deviation from prediction and the magnitude of that deviation should all be considered.All these signals are orderly processed,forming a comprehensive evaluation of the feedback,which is also the most important enlightening contribution of our study.The current study also has important practical meaning,as it has deepened our knowledge about the adaptive learning process and provides a reference for the future understanding and treatment of the individuals with abnormal adaptation,such as people with post-traumatic stress disorder,pathological gambler,drug addict and schizophrenics. |
PDF Full Text Request