Model-based analyses: Promises, pitfalls, and example applications to the study of cognitive control

Computational mechanisms underlying the dynamics of physical and cognitive fatigue

The willingness to exert effort for reward is essential but comes at the cost of fatigue. Theories suggest fatigue increases after both physical and cognitive exertion, subsequently reducing the motivation to exert effort. Yet a mechanistic understanding of how this happens on a moment-to-moment basis, and whether mechanisms are common to both mental and physical effort, is lacking. In two studies, participants reported momentary (trial-by-trial) ratings of fatigue during an effort-based decision-making task requiring either physical (grip-force) or cognitive (mental arithmetic) effort. Using a novel computational model, we show that fatigue fluctuates from trial-to-trial as a function of exerted effort and predicts subsequent choices. This mechanism was shared across the domains. Selective to the cognitive domain, committing errors also induced momentary increases in feelings of fatigue. These findings provide insight into the computations underlying the influence of effortful exertion on fatigue and motivation, in both physical and cognitive domains.

The computational psychiatry of antisocial behaviour and psychopathy

Antisocial behaviours such as disobedience, lying, stealing, destruction of property, and aggression towards others are common to multiple disorders of childhood and adulthood, including conduct disorder, oppositional defiant disorder, psychopathy, and antisocial personality disorder. These disorders have a significant negative impact for individuals and for society, but whether they represent clinically different phenomena, or simply different approaches to diagnosing the same underlying psychopathology is highly debated. Computational psychiatry, with its dual focus on identifying different classes of disorder and health (data-driven) and latent cognitive and neurobiological mechanisms (theory-driven), is well placed to address these questions. The elucidation of mechanisms that might characterise latent processes across different disorders of antisocial behaviour can also provide important advances. In this review, we critically discuss the contribution of computational research to our understanding of various antisocial behaviour disorders, and highlight suggestions for how computational psychiatry can address important clinical and scientific questions about these disorders in the future.

Task Learnability Modulates Surprise but Not Valence Processing for Reinforcement Learning in Probabilistic Choice Tasks

The goal of temporal difference (TD) reinforcement learning is to maximize outcomes and improve future decision-making. It does so by utilizing a prediction error (PE), which quantifies the difference between the expected and the obtained outcome. In gambling tasks, however, decision-making cannot be improved because of the lack of learnability. On the basis of the idea that TD utilizes two independent bits of information from the PE (valence and surprise), we asked which of these aspects is affected when a task is not learnable. We contrasted behavioral data and ERPs in a learning variant and a gambling variant of a simple two-armed bandit task, in which outcome sequences were matched across tasks. Participants were explicitly informed that feedback could be used to improve performance in the learning task but not in the gambling task, and we predicted a corresponding modulation of the aspects of the PE. We used a model-based analysis of ERP data to extract the neural footprints of the valence and surprise information in the two tasks. Our results revealed that task learnability modulates reinforcement learning via the suppression of surprise processing but leaves the processing of valence unaffected. On the basis of our model and the data, we propose that task learnability can selectively suppress TD learning as well as alter behavioral adaptation based on a flexible cost-benefit arbitration.

Robust and distributed neural representation of action values

Studies in rats, monkeys, and humans have found action-value signals in multiple regions of the brain. These findings suggest that action-value signals encoded in these brain structures bias choices toward higher expected rewards. However, previous estimates of action-value signals might have been inflated by serial correlations in neural activity and also by activity related to other decision variables. Here, we applied several statistical tests based on permutation and surrogate data to analyze neural activity recorded from the striatum, frontal cortex, and hippocampus. The results show that previously identified action-value signals in these brain areas cannot be entirely accounted for by concurrent serial correlations in neural activity and action value. We also found that neural activity related to action value is intermixed with signals related to other decision variables. Our findings provide strong evidence for broadly distributed neural signals related to action value throughout the brain.

Neurostimulation Reveals Context-Dependent Arbitration Between Model-Based and Model-Free Reinforcement Learning

While it is established that humans use model-based (MB) and model-free (MF) reinforcement learning in a complementary fashion, much less is known about how the brain determines which of these systems should control behavior at any given moment. Here we provide causal evidence for a neural mechanism that acts as a context-dependent arbitrator between both systems. We applied excitatory and inhibitory transcranial direct current stimulation over a region of the left ventrolateral prefrontal cortex previously found to encode the reliability of both learning systems. The opposing neural interventions resulted in a bidirectional shift of control between MB and MF learning. Stimulation also affected the sensitivity of the arbitration mechanism itself, as it changed how often subjects switched between the dominant system over time. Both of these effects depended on varying task contexts that either favored MB or MF control, indicating that this arbitration mechanism is not context-invariant but flexibly incorporates information about current environmental demands.

Designing and Interpreting Psychophysical Investigations of Cognition

Scientific experimentation depends on the artificial control of natural phenomena. The inaccessibility of cognitive processes to direct manipulation can make such control difficult to realize. Here, we discuss approaches for overcoming this challenge. We advocate the incorporation of experimental techniques from sensory psychophysics into the study of cognitive processes such as decision making and executive control. These techniques include the use of simple parameterized stimuli to precisely manipulate available information and computational models to jointly quantify behavior and neural responses. We illustrate the potential for such techniques to drive theoretical development, and we examine important practical details of how to conduct controlled experiments when using them. Finally, we highlight principles guiding the use of computational models in studying the neural basis of cognition.

The computational and neural substrates of moral strategies in social decision-making

Individuals employ different moral principles to guide their social decision-making, thus expressing a specific ‘moral strategy’. Which computations characterize different moral strategies, and how might they be instantiated in the brain? Here, we tackle these questions in the context of decisions about reciprocity using a modified Trust Game. We show that different participants spontaneously and consistently employ different moral strategies. By mapping an integrative computational model of reciprocity decisions onto brain activity using inter-subject representational similarity analysis of fMRI data, we find markedly different neural substrates for the strategies of ‘guilt aversion’ and ‘inequity aversion’, even under conditions where the two strategies produce the same choices. We also identify a new strategy, ‘moral opportunism’, in which participants adaptively switch between guilt and inequity aversion, with a corresponding switch observed in their neural activation patterns. These findings provide a valuable view into understanding how different individuals may utilize different moral principles.

The authors show that individuals apply different ‘moral strategies’ in interpersonal decision-making. These strategies are linked to distinct patterns of neural activity, even when they produce the same choice outcomes, illuminating how distinct moral principles can guide social behavior.

Modeling Morality in 3-D: Decision-Making, Judgment, and Inference

Humans face a fundamental challenge of how to balance selfish interests against moral considerations. Such trade-offs are implicit in moral decisions about what to do; judgments of whether an action is morally right or wrong; and inferences about the moral character of others. To date, these three dimensions of moral cognition-decision-making, judgment, and inference-have been studied largely independently, using very different experimental paradigms. However, important aspects of moral cognition occur at the intersection of multiple dimensions; for instance, moral hypocrisy can be conceived as a disconnect between moral decisions and moral judgments. Here we describe the advantages of investigating these three dimensions of moral cognition within a single computational framework. A core component of this framework is harm aversion, a moral sentiment defined as a distaste for harming others. The framework integrates economic utility models of harm aversion with Bayesian reinforcement learning models describing beliefs about others' harm aversion. We show how this framework can provide novel insights into the mechanisms of moral decision-making, judgment, and inference.

Interpreting BOLD: towards a dialogue between cognitive and cellular neuroscience

Cognitive neuroscience depends on the use of blood oxygenation level-dependent (BOLD) functional magnetic resonance imaging (fMRI) to probe brain function. Although commonly used as a surrogate measure of neuronal activity, BOLD signals actually reflect changes in brain blood oxygenation. Understanding the mechanisms linking neuronal activity to vascular perfusion is, therefore, critical in interpreting BOLD. Advances in cellular neuroscience demonstrating differences in this neurovascular relationship in different brain regions, conditions or pathologies are often not accounted for when interpreting BOLD. Meanwhile, within cognitive neuroscience, the increasing use of high magnetic field strengths and the development of model-based tasks and analyses have broadened the capability of BOLD signals to inform us about the underlying neuronal activity, but these methods are less well understood by cellular neuroscientists. In 2016, a Royal Society Theo Murphy Meeting brought scientists from the two communities together to discuss these issues. Here, we consolidate the main conclusions arising from that meeting. We discuss areas of consensus about what BOLD fMRI can tell us about underlying neuronal activity, and how advanced modelling techniques have improved our ability to use and interpret BOLD. We also highlight areas of controversy in understanding BOLD and suggest research directions required to resolve these issues.This article is part of the themed issue 'Interpreting BOLD: a dialogue between cognitive and cellular neuroscience'.

Probabilistic language models in cognitive neuroscience: Promises and pitfalls

Cognitive neuroscientists of language comprehension study how neural computations relate to cognitive computations during comprehension. On the cognitive part of the equation, it is important that the computations and processing complexity are explicitly defined. Probabilistic language models can be used to give a computationally explicit account of language complexity during comprehension. Whereas such models have so far predominantly been evaluated against behavioral data, only recently have the models been used to explain neurobiological signals. Measures obtained from these models emphasize the probabilistic, information-processing view of language understanding and provide a set of tools that can be used for testing neural hypotheses about language comprehension. Here, we provide a cursory review of the theoretical foundations and example neuroimaging studies employing probabilistic language models. We highlight the advantages and potential pitfalls of this approach and indicate avenues for future research.

A causal account of the brain network computations underlying strategic social behavior

During competitive interactions, humans have to estimate the impact of their own actions on their opponent's strategy. Here we provide evidence that neural computations in the right temporoparietal junction (rTPJ) and interconnected structures are causally involved in this process. By combining inhibitory continuous theta-burst transcranial magnetic stimulation with model-based functional MRI, we show that disrupting neural excitability in the rTPJ reduces behavioral and neural indices of mentalizing-related computations, as well as functional connectivity of the rTPJ with ventral and dorsal parts of the medial prefrontal cortex. These results provide a causal demonstration that neural computations instantiated in the rTPJ are neurobiological prerequisites for the ability to integrate opponent beliefs into strategic choice, through system-level interaction within the valuation and mentalizing networks.

Classification and treatment of antisocial individuals: From behavior to biocognition

Antisocial behavior is a heterogeneous construct that can be divided into subtypes, such as antisocial personality and psychopathy. The adverse consequences of antisocial behavior produce great burden for the perpetrators, victims, family members, and for society at-large. The pervasiveness of antisocial behavior highlights the importance of precisely characterizing subtypes of antisocial individuals and identifying specific factors that are etiologically related to such behaviors to inform the development of targeted treatments. The goals of the current review are (1) to briefly summarize research on the operationalization and assessment of antisocial personality and psychopathy; (2) to provide an overview of several existing treatments with the potential to influence antisocial personality and psychopathy; and (3) to present an approach that integrates and uses biological and cognitive measures as starting points to more precisely characterize and treat these individuals. A focus on integrating factors at multiple levels of analysis can uncover person-specific characteristics and highlight potential targets for treatment to alleviate the burden caused by antisocial behavior.

Taming the beast: extracting generalizable knowledge from computational models of cognition

Generalizing knowledge from experimental data requires constructing theories capable of explaining observations and extending beyond them. Computational modeling offers formal quantitative methods for generating and testing theories of cognition and neural processing. These techniques can be used to extract general principles from specific experimental measurements, but introduce dangers inherent to theory: model-based analyses are conditioned on a set of fixed assumptions that impact the interpretations of experimental data. When these conditions are not met, model-based results can be misleading or biased. Recent work in computational modeling has highlighted the implications of this problem and developed new methods for minimizing its negative impact. Here we discuss the issues that arise when data is interpreted through models and strategies for avoiding misinterpretation of data through model fitting.

Chasing probabilities - Signaling negative and positive prediction errors across domains

Adaptive actions build on internal probabilistic models of possible outcomes that are tuned according to the errors of their predictions when experiencing an actual outcome. Prediction errors (PEs) inform choice behavior across a diversity of outcome domains and dimensions, yet neuroimaging studies have so far only investigated such signals in singular experimental contexts. It is thus unclear whether the neuroanatomical distribution of PE encoding reported previously pertains to computational features that are invariant with respect to outcome valence, sensory domain, or some combination of the two. We acquired functional MRI data while volunteers performed four probabilistic reversal learning tasks which differed in terms of outcome valence (reward-seeking versus punishment-avoidance) and domain (abstract symbols versus facial expressions) of outcomes. We found that ventral striatum and frontopolar cortex coded increasingly positive PEs, whereas dorsal anterior cingulate cortex (dACC) traced increasingly negative PEs, irrespectively of the outcome dimension. Individual reversal behavior was unaffected by context manipulations and was predicted by activity in dACC and right inferior frontal gyrus (IFG). The stronger the response to negative PEs in these areas, the lower was the tendency to reverse choice behavior in response to negative events, suggesting that these regions enforce a rule-based strategy across outcome dimensions. Outcome valence influenced PE-related activity in left amygdala, IFG, and dorsomedial prefrontal cortex, where activity selectively scaled with increasingly positive PEs in the reward-seeking but not punishment-avoidance context, irrespective of sensory domain. Left amygdala displayed an additional influence of sensory domain. In the context of avoiding punishment, amygdala activity increased with increasingly negative PEs, but only for facial stimuli, indicating an integration of outcome valence and sensory domain during probabilistic choices.

Different Ways of Linking Behavioral and Neural Data via Computational Cognitive Models

Cognitive neuroscientists sometimes apply formal models to investigate how the brain implements cognitive processes. These models describe behavioral data in terms of underlying, latent variables linked to hypothesized cognitive processes. A goal of model-based cognitive neuroscience is to link these variables to brain measurements, which can advance progress in both cognitive and neuroscientific research. However, the details and the philosophical approach for this linking problem can vary greatly. We propose a continuum of approaches that differ in the degree of tight, quantitative, and explicit hypothesizing. We describe this continuum using four points along it, which we dub qualitative structural, qualitative predictive, quantitative predictive, and single model linking approaches. We further illustrate by providing examples from three research fields (decision making, reinforcement learning, and symbolic reasoning) for the different linking approaches.

Differential effects of reward and punishment in decision making under uncertainty: a computational study

Computational models of learning have proved largely successful in characterizing potential mechanisms which allow humans to make decisions in uncertain and volatile contexts. We report here findings that extend existing knowledge and show that a modified reinforcement learning model, which has separate parameters according to whether the previous trial gave a reward or a punishment, can provide the best fit to human behavior in decision making under uncertainty. More specifically, we examined the fit of our modified reinforcement learning model to human behavioral data in a probabilistic two-alternative decision making task with rule reversals. Our results demonstrate that this model predicted human behavior better than a series of other models based on reinforcement learning or Bayesian reasoning. Unlike the Bayesian models, our modified reinforcement learning model does not include any representation of rule switches. When our task is considered purely as a machine learning task, to gain as many rewards as possible without trying to describe human behavior, the performance of modified reinforcement learning and Bayesian methods is similar. Others have used various computational models to describe human behavior in similar tasks, however, we are not aware of any who have compared Bayesian reasoning with reinforcement learning modified to differentiate rewards and punishments.

Dysfunctions of decision-making and cognitive control as transdiagnostic mechanisms of mental disorders: advances, gaps, and needs in current research

Disadvantageous decision-making and impaired volitional control over actions, thoughts, and emotions are characteristics of a wide range of mental disorders such as addiction, eating disorders, depression, and anxiety disorders and may reflect transdiagnostic core mechanisms and possibly vulnerability factors. Elucidating the underlying neurocognitive mechanisms is a precondition for moving from symptom-based to mechanism-based disorder classifications and ultimately mechanism-targeted interventions. However, despite substantial advances in basic research on decision-making and cognitive control, there are still profound gaps in our current understanding of dysfunctions of these processes in mental disorders. Central unresolved questions are: (i) to which degree such dysfunctions reflect transdiagnostic mechanisms or disorder-specific patterns of impairment; (ii) how phenotypical features of mental disorders relate to dysfunctional control parameter settings and aberrant interactions between large-scale brain systems involved in habit and reward-based learning, performance monitoring, emotion regulation, and cognitive control; (iii) whether cognitive control impairments are consequences or antecedent vulnerability factors of mental disorders; (iv) whether they reflect generalized competence impairments or context-specific performance failures; (v) whether not only impaired but also chronic over-control contributes to mental disorders. In the light of these gaps, needs for future research are: (i) an increased focus on basic cognitive-affective mechanisms underlying decision and control dysfunctions across disorders; (ii) longitudinal-prospective studies systematically incorporating theory-driven behavioural tasks and neuroimaging protocols to assess decision-making and control dysfunctions and aberrant interactions between underlying large-scale brain systems; (iii) use of latent-variable models of cognitive control rather than single tasks; (iv) increased focus on the interplay of implicit and explicit cognitive-affective processes; (v) stronger focus on computational models specifying neurocognitive mechanisms underlying phenotypical expressions of mental disorders.

Psychopathy-related traits and the use of reward and social information: a computational approach

Psychopathy is often linked to disturbed reinforcement-guided adaptation of behavior in both clinical and non-clinical populations. Recent work suggests that these disturbances might be due to a deficit in actively using information to guide changes in behavior. However, how much information is actually used to guide behavior is difficult to observe directly. Therefore, we used a computational model to estimate the use of information during learning. Thirty-six female subjects were recruited based on their total scores on the Psychopathic Personality Inventory (PPI), a self-report psychopathy list, and performed a task involving simultaneous learning of reward-based and social information. A Bayesian reinforcement-learning model was used to parameterize the use of each source of information during learning. Subsequently, we used the subscales of the PPI to assess psychopathy-related traits, and the traits that were strongly related to the model's parameters were isolated through a formal variable selection procedure. Finally, we assessed how these covaried with model parameters. We succeeded in isolating key personality traits believed to be relevant for psychopathy that can be related to model-based descriptions of subject behavior. Use of reward-history information was negatively related to levels of trait anxiety and fearlessness, whereas use of social advice decreased as the perceived ability to manipulate others and lack of anxiety increased. These results corroborate previous findings suggesting that sub-optimal use of different types of information might be implicated in psychopathy. They also further highlight the importance of considering the potential of computational modeling to understand the role of latent variables, such as the weight people give to various sources of information during goal-directed behavior, when conducting research on psychopathy-related traits and in the field of forensic psychiatry.

A simple solution for model comparison in bold imaging: the special case of reward prediction error and reward outcomes

Conventional neuroimaging techniques provide information about condition-related changes of the BOLD (blood-oxygen-level dependent) signal, indicating only where and when the underlying cognitive processes occur. Recently, with the help of a new approach called “model-based” functional neuroimaging (fMRI), researchers are able to visualize changes in the internal variables of a time varying learning process, such as the reward prediction error or the predicted reward value of a conditional stimulus. However, despite being extremely beneficial to the imaging community in understanding the neural correlates of decision variables, a model-based approach to brain imaging data is also methodologically challenging due to the multicollinearity problem in statistical analysis. There are multiple sources of multicollinearity in functional neuroimaging including investigations of closely related variables and/or experimental designs that do not account for this. The source of multicollinearity discussed in this paper occurs due to correlation between different subjective variables that are calculated very close in time. Here, we review methodological approaches to analyzing such data by discussing the special case of separating the reward prediction error signal from reward outcomes.

Model-based analysis of context-specific cognitive control

Interference resolution is improved for stimuli presented in contexts (e.g., locations) associated with frequent conflict. This phenomenon, the context-specific proportion congruent (CSPC) effect, has challenged the traditional juxtaposition of “automatic” and “controlled” processing because it suggests that contextual cues can prime top-down control settings in a bottom-up manner. We recently obtained support for this “priming of control” hypothesis with functional magnetic resonance imaging by showing that CSPC effects are mediated by contextually cued adjustments in processing selectivity. However, an equally plausible explanation is that CSPC effects reflect adjustments in response caution triggered by expectancy violations (i.e., prediction errors) when encountering rare events as compared to common ones (e.g., incongruent trials in a task context associated with infrequent conflict). Here, we applied a quantitative model of choice, the linear ballistic accumulator (LBA), to distil the reaction time and accuracy data from four independent samples that performed a modified flanker task into latent variables representing the psychological processes underlying task-related decision making. We contrasted models which differentially accounted for CSPC effects as arising either from contextually cued shifts in the rate of sensory evidence accumulation (“drift” models) or in the amount of evidence required to reach a decision (“threshold” models). For the majority of the participants, the LBA ascribed CSPC effects to increases in response threshold for contextually infrequent trial types (e.g., congruent trials in the frequent conflict context), suggesting that the phenomenon may reflect more a prediction error-triggered shift in decision criterion rather than enhanced sensory evidence accumulation under conditions of frequent conflict.

Learning from experience: event-related potential correlates of reward processing, neural adaptation, and behavioral choice

To behave adaptively, we must learn from the consequences of our actions. Studies using event-related potentials (ERPs) have been informative with respect to the question of how such learning occurs. These studies have revealed a frontocentral negativity termed the feedback-related negativity (FRN) that appears after negative feedback. According to one prominent theory, the FRN tracks the difference between the values of actual and expected outcomes, or reward prediction errors. As such, the FRN provides a tool for studying reward valuation and decision making. We begin this review by examining the neural significance of the FRN. We then examine its functional significance. To understand the cognitive processes that occur when the FRN is generated, we explore variables that influence its appearance and amplitude. Specifically, we evaluate four hypotheses: (1) the FRN encodes a quantitative reward prediction error; (2) the FRN is evoked by outcomes and by stimuli that predict outcomes; (3) the FRN and behavior change with experience; and (4) the system that produces the FRN is maximally engaged by volitional actions.

Modulation of the feedback-related negativity by instruction and experience

A great deal of research focuses on how humans and animals learn from trial-and-error interactions with the environment. This research has established the viability of reinforcement learning as a model of behavioral adaptation and neural reward valuation. Error-driven learning is inefficient and dangerous, however. Fortunately, humans learn from nonexperiential sources of information as well. In the present study, we focused on one such form of information, instruction. We recorded event-related potentials as participants performed a probabilistic learning task. In one experiment condition, participants received feedback only about whether their responses were rewarded. In the other condition, they also received instruction about reward probabilities before performing the task. We found that instruction eliminated participants' reliance on feedback as evidenced by their immediate asymptotic performance in the instruction condition. In striking contrast, the feedback-related negativity, an event-related potential component thought to reflect neural reward prediction error, continued to adapt with experience in both conditions. These results show that, whereas instruction may immediately control behavior, certain neural responses must be learned from experience.

Effective connectivity reveals important roles for both the hyperdirect (fronto-subthalamic) and the indirect (fronto-striatal-pallidal) fronto-basal ganglia pathways during response inhibition

Fronto-basal ganglia pathways play a crucial role in voluntary action control, including the ability to inhibit motor responses. Response inhibition might be mediated via a fast hyperdirect pathway connecting the right inferior frontal gyrus (rIFG) and the presupplementary motor area (preSMA) with the subthalamic nucleus or, alternatively, via the indirect pathway between the cortex and caudate. To test the relative contribution of these two pathways to inhibitory action control, we applied an innovative quantification method for effective brain connectivity. Functional magnetic resonance imaging data were collected from 20 human participants performing a Simon interference task with an occasional stop signal. A single right-lateralized model involving both the hyperdirect and indirect pathways best explained the pattern of brain activation on stop trials. Notably, the overall connection strength of this combined model was highest on successfully inhibited trials. Inspection of the relationship between behavior and connection values revealed that fast inhibitors showed increased connectivity between rIFG and right caudate (rCaudate), whereas slow inhibitors were associated with increased connectivity between preSMA and rCaudate. In compliance, connection strengths from the rIFG and preSMA into the rCaudate were correlated negatively. If participants failed to stop, the magnitude of experienced interference (Simon effect), but not stopping latency, was predictive for the hyperdirect-indirect model connections. Together, the present results suggest that both the hyperdirect and indirect pathways act together to implement response inhibition, whereas the relationship between performance control and the fronto-basal ganglia connections points toward a top-down mechanism that underlies voluntary action control.

Reciprocal relations between cognitive neuroscience and formal cognitive models: opposites attract?

Cognitive neuroscientists study how the brain implements particular cognitive processes such as perception, learning, and decision-making. Traditional approaches in which experiments are designed to target a specific cognitive process have been supplemented by two recent innovations. First, formal cognitive models can decompose observed behavioral data into multiple latent cognitive processes, allowing brain measurements to be associated with a particular cognitive process more precisely and more confidently. Second, cognitive neuroscience can provide additional data to inform the development of formal cognitive models, providing greater constraint than behavioral data alone. We argue that these fields are mutually dependent; not only can models guide neuroscientific endeavors, but understanding neural mechanisms can provide key insights into formal models of cognition.

Time scales of representation in the human brain: weighing past information to predict future events

The estimates that humans make of statistical dependencies in the environment and therefore their representation of uncertainty crucially depend on the integration of data over time. As such, the extent to which past events are used to represent uncertainty has been postulated to vary over the cortex. For example, primary visual cortex responds to rapid perturbations in the environment, while frontal cortices involved in executive control encode the longer term contexts within which these perturbations occur. Here we tested whether primary and executive regions can be distinguished by the number of past observations they represent. This was based on a decay-dependent model that weights past observations from a Markov process and Bayesian Model Selection to test the prediction that neuronal responses are characterized by different decay half-lives depending on location in the brain. We show distributions of brain responses for short and long term decay functions in primary and secondary visual and frontal cortices, respectively. We found that visual and parietal responses are released from the burden of the past, enabling an agile response to fluctuations in events as they unfold. In contrast, frontal regions are more concerned with average trends over longer time scales within which local variations are embedded. Specifically, we provide evidence for a temporal gradient for representing context within the prefrontal cortex and possibly beyond to include primary sensory and association areas.