Let's first consider an example to describe and to integrate how different existing theoretical conceptions may be combined to understand the mechanisms that lead to error commissioning. After this description, we will examine the evidence and conceptual grounds for a neuro-cognitive perspective in detail (refer section A Neuro-cognitive Perspective of Error Commissioning). Along the lines of this example, we will elucidate different conceptions and the links between these conceptions to understand the mechanisms of error commissioning and how these may influence future formalized computational models of errors commission. Goal of this integrative review is to provide an overview of potential neurobiological and functional neuroanatomical factors that are possibly essential to consider when interested in the neural mechanisms leading to an error. Therefore, we do not intend to put forward a formalized computational model of error commissioning.
Consider a typical situation that provokes response errors and is characterized by discrepancies between a desired goal state and the response that is actually executed. Figure 2 depicts a perspective about current cognitive and neurophysiological theories about how action and behavioral adaptation are implemented.
It can be assumed that task-goal representations are stored in the working memory in the prefrontal cortex (PFC) and as a copy that is likely transferred to the basal ganglia (BG) via functional prefrontal BG loops (Chudasama and Robbins, 2006). Due to this transfer, a representation of the task goal is also set up in the BG, and sensory input is simultaneously provided to the BG. These different inputs form a “map” of different neural activities that may vary in strength (Figure 1, Alternative B). However, the selected action does not only depend on the neural representations but also on those parameters of stimulus processing (Lawrence et al., 2003) that are related to differences in the saliency of stimuli.
Suppose there are two stimuli (A and B) that differ in saliency and are related to opposing actions (A = correct; B = erroneous) that are attempting gain control over behavior. Due to their different saliencies, these actions evoke correspondingly different degrees of activation in their competition with each other (Desimone and Duncan, 1995; Reynolds and Chelazzi, 2004; Knudsen, 2007). Whether stimulus A wins this perceptual competition depends on (i) the relative saliency of stimulus B and (ii) the intentional biases (i.e., top-down influences) that favor stimulus B and disfavor the processing of stimulus A (Knudsen, 2007). The net result of these perceptual competitive influences determines whether feature A or B is detected and controls behavior. If the net result favors the erroneous stimulus B, stimulus A loses the competition, is not detected and will not govern behavior (Desimone and Duncan, 1995). In addition to these factors, other factors also influencing this net-result may be related to top–down attentional biases, like previously learned stimulus-response associations or bottom–up influences on these association strength like spontaneous fluctuations and lapses in the attentional system (which is of particular importance when considering altered error processing/commission in diseases; e.g., Weissman et al., 2006; Sonuga-Barke and Castellanos, 2007; Sarter and Paolone, 2011). Thus, this process may ultimately lead to an error. We assume that the net result of these processes is also fed into fronto-striatal loops. Within the basal ganglia, an action selection mechanism operates using the principles of a “winner-takes-all” network that converges to a single winner (Bar-Gad et al., 2003). The most salient of the representations provided by the PFC, sensory signals from the primary sensory areas and efference copies of motor activity is selected. In the next step, the winning representation is fed to the motor cortex, and the response is executed. There are at least three possible constellations by which a correct task-goal representation can compete with other, error-favoring sensory inputs (Figure 1, Alternative B). If the neural activation at the BG level caused by the correct task representation is stronger than the other neural activations, the correct task representation wins, and the correct response is likely to be executed. In the opposite case of a dominant irrelevant sensory input, an erroneous response is the most likely to be provoked. However, if the activations of task representation and the sensory input are equivalent, the motor response may be selected almost randomly via a stochastic process (e.g., consider a situation in which one is uncertain about the correct response) or even activate conflicting motor responses. However, task processing does not end with the initiation of response execution because inadequate task handling (irrespective of being erroneous or unexpected) requires controlled adaptation. This control and the corresponding structures (i.e., the ACC and PFC) are the next important functional stations in this framework that should be considered. Once the response is executed, a novel action can, in principal, be executed immediately. However, if the previous response was erroneous, this processing sequence is not adequate. To adapt behavior, the previously described process can begin again but has to wait for the outcome of an evaluation that provides information about the adequacy/correctness of the executed response. At this point, the above-mentioned neuro-cognitive perspective on error commissioning connects with the established error processing models that focus on the consequences of an error and how those consequences are used to adapt behavior.
An important aspect of the above neuro-cognitive conception of error commissioning is that the likelihood of committing an error is not stable over time. Rather, the fluctuations in error likelihood are grounded in the processes discussed above. We assume that each of the above-described factors reveals a specific processing pattern over time. For prefrontal processes, this pattern is related to fluctuations in the stability of task-goal representations that might emerge because in prefrontal networks stability of information is not stable, but fluctuates. Moreover, fluctuations in the saliency of error-favoring sensory input emerge due to the ability to suppress this erroneous input and the relative saliencies of relevant and irrelevant (i.e., error-favoring) inputs. The interplay of each of these factors then determines the resulting error likelihood. This interplay likely depends on different neurobiological factors and their temporal properties. Fluctuations in error likelihood are therefore not only determined by the strengths of representations in the prefrontal networks. Rather, fluctuations in error likelihood are also influenced by mechanisms related to attentional selection and vigilance that are associated with the perceptual processing of stimuli. However, the precise contribution of each of these processes might vary considerably. To provide a simple illustration of error likelihood, only the influences of the strength of the task representation in the PFC and the saliency of the error-favoring stimulus input are depicted in Figure 1, Alternative B. The left image in Figure 1, Alternative B represents a putative case of low error likelihood at the time point of action selection. In this case, the strength of the prefrontal representation is considered to be high, and the saliency of error-favoring stimuli is considered to be low. The resulting error likelihood emerges by combining both factors. However, error likelihood can increase (Figure 1, Alternative B, middle and right); for example, if low-stability task representations coincide with a high-saliency error-favoring sensory input at the time of action selection, a high likelihood of error will result. Indeed, the error likelihood at the time of action selection can remain at a fairly low level for some period of time and can also remain at a fairly high level for some period of time. Obviously, the durations of the time periods in which the fluctuation of the error likelihood is low depend on the strength of the task-goal representation and the saliency of the error-favoring sensory input.
Details on the neuro-cognitive basis of error-commissioning
Having described factors that may play a role in error commissioning, several established conceptions appear to play roles in a comprehensive account leading to an understanding of the principles resulting in response errors. The processes described above can be understood in terms of the core ideas of reinforcement learning theory (Holroyd and Coles, 2002) and conceptions proposing that the basal ganglia play an important role in action selection (e.g., Bar-Gad et al., 2003; Gurney et al., 2004b; Maia and Frank, 2011). Moreover, assumptions of the dual-state theory of dopamine function (Seamans and Yang, 2004; Durstewitz and Seamans, 2008) and the guided activation theory (Miller, 2000; Miller and Cohen, 2001) seem to be important when trying to examine the mechanisms that lead to an error, i.e., the investigation of error commissioning. This is because these theories deal with questions how representations stored in prefrontal networks are maintained and used to guide action selection. Obviously, these are important to consider the processes that lead to an error. In particular, in tasks that require continuous and rapid processing of stimuli, errors emerge due to attentional lapses or even conflicts in stimulus processing. Thus, aspects of attentional selection processes also have to be considered. In this regard the “biased competition model” of attention (Desimone and Duncan, 1995) may provide useful grounds. However, the investigation of the processing of information and action selection (or decision) from a neurophysiological perspective should not ignore the heuristic value of the existing well-established cognitive approaches because these provide a solid theoretical basis regarding how information is processed and how errors might emerge. Thus, it appears at hand to combine different theoretical approaches within the field of neuroscience, with established psychological models such as the “drift diffusion model” (DDM; Ratcliff, 1978, 1979, 1980, 2006; Ratcliff and Rouder, 1998; Vandekerckhove and Tuerlinckx, 2007; Ratcliff and McKoon, 2008) to approach the question how errors are committed. In the forthcoming we detail what how these factors may influence error commissioning and what aspect are necessary to consider in future computational models on error commissioning.
From prefrontal networks to attentional selection
Central to the above-suggested integration of different concepts are assumptions about the stability of information in the prefrontal cortex (PFC). In nearly every daily situation people have a goal that they try to fulfill. Often, this goal is the establishment of proper stimulus-response mappings in terms of the defined task at hand; e.g., “Respond with the left hand when stimulus X is present, and respond with the right hand when stimulus Y is present.” Such task-goal representations are stored in working memory buffers within the prefrontal cortex (PFC) (e.g., Jonides et al., 2008). The most established assumption about the function of the PFC networks is that they hold and manipulate information (task-goal representations) for future use via persistent activity states (e.g., Seamans and Yang, 2004). Within the PFC, dopamine (DA) influx may serve as a gating signal that instructs the network when to maintain a given activity state (Miller, 2000). The neuromodulatory effects of dopamine may strengthen current representations and protect them against interference due to disruption by irrelevant distracting information (Miller, 2000). According to the “dual-state theory of dopamine function,” whether new information can easily access working memory buffers (state 1) or current representations are maintained and stabilized within prefrontal networks (state 2; Seamans and Yang, 2004; Durstewitz and Seamans, 2008) appears to depend on the state of the dopaminergic system.
In the first state (state 1), D2-receptor-mediated neural transmission predominates and allows multiple inputs to access working memory buffers. In contrast, the second state (state 2) is dominated by D1-receptor-related neural transmission. In this state, the working memory buffers are relatively closed, but information held within these buffers is more stable and controls the output of prefrontal networks (Miller, 2000; Seamans and Yang, 2004). It may be speculated that these different states serve different brain functions in relation to performance monitoring. Recent evidence has shown that D2-receptor mediated neural transmission, though less important in the prefrontal cortex, is associated with exploitative learning that adjusts response times as a function of positive and negative outcomes (Frank et al., 2009), which suggests that D2-receptor-mediated neural transmission is of particular importance for processes that are related to error commission and monitoring. However, the D1 and D2 states fluctuate spontaneously and are organized in an antagonistic fashion. During phases in which the D1-state dominates, D2-receptor-related neural transmission is less active, and vice versa. In the mechanism that mediates the transitions between the D1 and D2 states, it is assumed that GABAergic and NMDA processes play important roles (Seamans and Yang, 2004; Durstewitz and Seamans, 2008). It is assumed that increases in D1-receptor activation initially augment the robustness of representations within the PFC (Seamans and Yang, 2004). However, further increases above an optimal level reverse these effects and shift the system away from robustness. This process leads to changes in the mode of the cognitive processes occurring in the PFC. The dual-state theory of the dopamine system is not the only approach to assume the existence of dynamic gating of representations in the PFC (Hazy et al., 2010; O'Reilly et al., 2010). As illustrated in the work of Dayan (2008, 2007), several approaches on PFC functioning involve switching between network states that either process information according to a habit that is inflexible or more goal-directed and rule-governed states that enable the flexible processing of task goals. In this regard, the biophysical properties of the PFC that determine the stability of information and other models of prefrontal cortical functioning reflect the assumption of fluctuating states and the stability of task goals in the PFC and, hence, the essential basis of the assumption of differences in error likelihood over time.
However, the above-described mechanisms related to task goal representations in the PFC have important consequences for information processing in other brain areas. This idea constitutes the core of the “guided activation theory” (Miller, 2000; Miller and Cohen, 2001). This theory proposes that representations of task-specific rules stored in PFC networks form attentional templates and goals (see also: Wood and Grafman, 2003) that are needed to enable goal-directed behavior (Miller, 2000). This task information enables the PFC to control processing in other brain systems and direct that processing toward task-relevant information (Miller, 2000). The aggregate effect of these bias signals is to guide the flow of neural activity along pathways that establish the proper mappings between the inputs, internal states and outputs that are needed to (correctly) perform a given task (Miller and Cohen, 2001). This control is particularly important whenever stimuli are ambiguous, i.e., when stimuli activate more than one input representation (Miller and Cohen, 2001). From this perspective, the constellation of PFC biases can be viewed as the neural implementation of rules or goals depending on the target of their biasing influences. If these “bias signals” affect sensory modalities, this can affect the mechanisms of attentional control and thereby affect another major factor that may influence error commissioning: attentional selection processes.
Frontal regions, including the anterior cingulate (Cabeza and Nyberg, 2000; Lawrence et al., 2003), the right frontal cortex (middle and inferior frontal gyrus) and bilateral parietal regions (i.e., inferior parietal cortex), play a role in maintaining and controlling attention over time (Posner and Petersen, 1990; Coull, 1998). They are the source of the biasing influences on sensory processing (Knudsen, 2007) that can reduce the saliency of the error-favoring stimuli. To the extent that biasing influences on attention depend on representations in the prefrontal networks (Desimone and Duncan, 1995; Knudsen, 2007), these mechanisms may also be affected by the fluctuations in prefrontal networks described above. As an effect that these influences are not always able to bias attentional selection toward desired task goals, which will, in addition to influences related to the relative saliency of error-leading stimuli, determine error likelihood. The “biased competition theory of attention” (Knudsen, 2007; Desimone and Duncan, 1995) assumes that the outcome of perceptual competition is determined by the saliency of the stimuli (Beste et al., 2011b, 2012a; Beste and Dinse, 2013) and intentional biases (top-down influences). The net result of these perceptual competitive influences determines whether feature A or B is detected and controls behavior (Sänger and Wascher, 2011; Labrenz et al., 2012). When the net result favors the erroneous stimulus B, stimulus A loses the competition, is not detected and does not govern behavior (Desimone and Duncan, 1995), which may ultimately lead to an error. Thus, fluctuations in the stability of the task goal representation in the PFC affect attentional selection processes and are therefore another major property that determines the likelihood of committing an error. It is simply the lack of biasing influences on attentional selection processes that further contributes to the likelihood of committing an error. Both aspects of error commissioning, i.e., the stability of the task goal representation and attentional selection processes, are interrelated determinants of error commissioning.
Convergent input to the basal ganglia and the “winner-take-all principle”
However, functioning of the prefrontal cortex cannot be understood without referring to the basal ganglia, as these structures are well known to from closed functional loops (e.g., Chudasama and Robbins, 2006). Moreover, when considering the possible importance of the dopamine system for fluctuations in error likelihood and hence mechanisms of error commission, the basal ganglia need also be taken into account. The basal ganglia are important when considering the gating functions of the prefrontal cortex (O'Reilly et al., 2010) that are related to shifts in the processing modes of task goal representations in prefrontal cortical networks. However, to fulfill this requirement, the basal ganglia must serve as a “hub region” that receives information from the PFC and is modulated by processes occurring in the sensory modalities. It is well established that the PFC is closely connected to the basal ganglia via distinct functional loops (Chudasama and Robbins, 2006). Several theoretical accounts suggest that the basal ganglia are central to response selection mechanisms (Redgrave et al., 1999; Gurney et al., 2001, 2004a; Humphries and Gurney, 2002; Humphries et al., 2006; Maia and Frank, 2011). Conceptions that stress the importance of basal ganglia structures in action selection and control propose that the selection of actions (motor commands) depends on the relative salience of competing actions and that the most salient competitor wins this selection process (Redgrave et al., 1999, 2011). Action selection should be terminated when it has been successful or if it proves to be ineffective or erroneous. The selection of a correct action may also be “interrupted” by a competitor that is relatively more salient than the desired action (Redgrave et al., 1999) because action selection at the level of the basal ganglia is described by a “winner-take-all” (WTA) mechanism (Kropotov and Etlinger, 1999; Redgrave et al., 1999; Plenz, 2003). Following selection, the winning outcome may begin to reduce the salience of its predisposing conditions (as these become partially fulfilled). When the salience of a selected action falls below that of a close competitor, this competitive action will be executed (Redgrave et al., 1999). This WTA mechanism that mediates the selection of actions is implemented using a neural network that converges to a single winner (Bar-Gad et al., 2003). In biologically constrained computational models (Gurney et al., 2004b), these mechanisms are modeled as the action of striatal medium spiny neurons (MSNs). The MSNs are strongly modulated by the dopaminergic system via dopamine D1 and D2 receptors (overview: Surmeier et al., 2010). The dense network of inhibitory connections between MSNs is assumed to inhibit neighboring neurons and thereby maintain the activity of only a single neuron. The selection process takes place within the striatum, and the chosen action is then conveyed to the output layer of the basal ganglia and subsequently fed back to the cortex where the selected response is executed (Bar-Gad et al., 2003). Because bias signals affect response execution processes (Miller and Cohen, 2001), it is reasonable to assume that representations or contextual inputs may influence basal ganglia processes and, hence, WTA (for a review see: Samejima et al., 2005; Redgrave and Gurney, 2006). This assumption is even more likely because the functioning of prefrontal cortical areas cannot be understood in isolation from the modulatory influences of the basal ganglia (Parvizi, 2009). The ultimate effect of these bias signals is the re-weighting of the relative influences of goal-directed and error-favoring representations in the basal ganglia. The basal ganglia WTA network “decides” in favor of the activation with the strongest representation. In turn, the basal ganglia WTA mechanism most likely selects another (possibly correct) action simply because the relative weights favor that action. This re-weighting of task representations at the level of the basal ganglia may be attributable to one or both of the following processes:
the correct task representation is strengthened in the prefrontal cortical networks;
the sensitivity of the neural structures of the error-favoring information is reduced.
The above-mentioned (“re”)-strengthening of task representations in the PFC by dopaminergic influx (Miller, 2000) is necessary because the error may partly have occurred due to weaker task-goal representations, which might ultimately lead to a disadvantageous pattern of activations in the basal ganglia that may foster the occurrence of an error, particularly when a distracting, error-favoring sensory input is concomitantly present.
Similar to prefrontal representations, basic visual inputs enter the basal ganglia networks (Hikosaka, 1989; Hikosaka and Wurtz, 1989; Silkis, 2000; Coizet et al., 2007; Redgrave et al., 2011). A growing body of evidence suggests that a sub-cortical structure in the dorsal midbrain (i.e., the superior colliculus) is the most likely source of the early visual input to dopaminergic neurons (Redgrave and Gurney, 2006; Silkis, 2000). This subcortical pathway has been substantiated through electrophysiological studies in rats, cats and monkeys (Comoli et al., 2003; McHaffie et al., 2006). Thus, several cortical structures provide convergent input to the basal ganglia, and this input consists of (i) a set of stimuli, (ii) a task set that has been established in the PFC, and (iii) a corresponding efference copy. This information basically provides the basis for action selection in striatal networks (Bar-Gad et al., 2003; Redgrave and Gurney, 2006).
As originally conceptualized by Redgrave et al. (1999), the idea of a WTA mechanism is also central to other computational models of response selection in the basal ganglia. The model of Bar-Gad et al. (2003) conceives of MSNs as central to the comparison of different response options and, hence, the selection between those options. Similar assumptions have been put forward in the den model of Plenz (2003) and the model of Plenz and Kitai (1998). Moreover, other recent models (e.g., Humphries et al., 2006) assume that the basal ganglia perform response selection processes via a restrictive mechanism; however, these models assume that an important property of the regulation of action selection in the basal ganglia is the dopaminergic system. Humphries et al. (2006) assumed that response selection and control are mediated via different dopaminergic receptor systems. This assumption is also implemented in other models of the basal ganglia (e.g., Maia and Frank, 2011; Wiecki and Frank, 2013) in which the distinctions between dopaminergic subsystems include Go- and Nogo-neuron populations. These latter models developed by Frank et al. therefore primarily address response selection in terms of response inhibition mechanisms (Wiecki and Frank, 2013). However, the models can also be applied to two-choice decision-making and response selection processes, which are the focus of the model of Humphries et al. (2006). These other conceptions of action selection in the basal ganglia are not at odds with the WTA network upon which our error-commissioning model is based because the efficiency of the WTA mechanism is modulated by dopaminergic signaling (Gurney et al., 2004a; Humphries et al., 2006).
Supposed roles of the BG and the PFC in error commission may be linked to drift-diffusion models (DDM). The model developed by Frank (2005) suggests that following stimulus presentation, the medial PFC generates possible actions with different probabilities of execution that depend on the specific stimulus (Cavanagh et al., 2011). This notion is in line with the finding that a correlate of response monitoring, i.e., the CRN, varies as a function of S–R mapping (refer Box 2): the CRN and thus PFC functions appear to be attenuated if the probability of a certain response is reduced because, prior to response selection, a defined visual spatial stimulus position is also inhibited (Hoffmann and Wascher, 2012). However, it can be assumed that if there is a response conflict, the medial PFC-STN network should increase the decision threshold (i.e., the boundary separation) to enable the cortico-striatal network to assess the reward values of the response alternatives (Cavanagh et al., 2011). Cavanagh et al. investigated this assumption via integration of the core ideas of the BG model of Frank (2005, 2006) and the DDM. More specifically, they investigated whether the subthalamic nucleus (STN), as important basal ganglia structure receiving input from the frontal cortex, has an inhibitory influence during decision conflict. They found that trial-to-trial medial PFC activity, as measured based on EEG theta power, is correlated with the threshold for evidence accumulation. This relationship is modulated by conflict. Besides this correlative finding they found that deep-brain stimulation of the STN in Parkinson's patients strongly modulates this relationship. Further corroborating the relevance of basal ganglia processes, are studies by Forstmann et al. (2008, 2010). These studies show that decision processes, modeled by DDM-like processes are strongly related to basal ganglia processes (Forstmann et al., 2008). Importantly, activation in cortico-striatal networks and especially the subthalamic nucleus, considered to be important for processes of error commission (refer sections above), have been shown to directly predict the modulation of decision processes at the behavioral level in choice-response tasks (Forstmann et al., 2010). Due to the strong interrelation of the DDM with neurobiological processes of the basal ganglia, the DDM provides opportunities to test a key assumption of error commission on a behavioral level, i.e., the modulation of error likelihood. The DDM allows a quantification of the parameters that reflect inter-trial fluctuations in decision processes. These fluctuations are captured by the variability in the drift rate (e.g., variability due to fluctuations in attention), starting point (e.g., due to online reward or punishment) and the non-decisional parameter. Apparently, there is a close link between the DDM and the error commissioning framework proposed herein; i.e. according to the framework presented, fluctuations in error likelihood are determined by the strengths of the representations in prefrontal networks, which are akin to the boundary separation and starting point variabilities in the DDM. Moreover, the mechanisms associated with attentional selection are conceptually similar to the drift rate parameter proposed in the DDM. Aspects of vigilance are closely related to the perceptual processing of stimuli, which is akin to the non-decisional parameter and its variability. Accordingly, if these processes/structures are manipulated, one would expect variations in the corresponding DDM parameters and, hence, changes in the fluctuations in error likelihood, as outlined at the end of this article. It should be noted that the neurobiological mechanisms modulating fronto-striatal networks described above are not exhaustive. There are many other modulators affecting processing in prefrontal networks.
Timing constraints in error-commissioning
It can be assumed that the basal ganglia and the action selection processes mediated via these structures play an important role when considering factors possibly important for the commissioning of errors. Basically, the basal ganglia can be conceptualized as a “hub region” that receives different types of information. These inputs influence action selection, and hence, the mechanisms of error commissioning. It is therefore critical that the basal ganglia, as a “hub region,” receive all of this information in a timely manner.
Another aspect is the stability of task-goal representations in prefrontal structures, which is determined by dopaminergic signaling via dopamine D1 and D2 receptors. A major problem in cognitive control models is generally that the dopamine system is considered to be too slow to support the rapid processes of action selection that ultimately lead to erroneous responses. Thus, Jocham and Ullsperger (2009) proposed that error monitoring and the behavioral adaptation processes that follow an error, such as those related to the ERN, are mediated via slower dopamine responses after the error monitoring system has been activated through other (faster) neurotransmitter systems.
However, it can be assumed that the stabilities of task-goal representations are dependent on the properties of the dopaminergic system, which have been shown to exert long latency responses in prefrontal structures that last several seconds (e.g., Robinson et al., 2003; Seamans and Yang, 2004; Heien et al., 2005). However, it is not clear, whether the effects of prefrontal processes on striatal processes are mediated by the dopaminergic system per se: cortico-striatal synapses are glutamatergic in nature (e.g., Bolam et al., 2000), and it has recently been shown that striatal states can reliably be changed by these glutamatergic inputs and influence action selection processes (e.g., Tomkins et al., 2014). Indeed, it is likely that that changes in the stabilities of task-goal representations in prefrontal networks affect the striatal mechanisms of action control via glutamatergic projections. In addition to these “contextual” influences, “sensory” influences also modulate striatal structures. As outlined above, these sensory influences likely enter striatal structures (for a review, see Redgrave et al., 2011) and use short-latency dopamine signals. This short-latency component of the visual input to DA neurons derives from subcortical visual processing regions in the superior colliculus in the midbrain (e.g., McHaffie et al., 2006; May et al., 2009), and it has been shown that sensory processing in the cerebral cortex can drive phasic responses in DA neurons (for a review, see Redgrave et al., 2011). These findings show that dopaminergic neurons are active prior to behavioral responses. The dopaminergic system may thus play a role in error commissioning either at the level of short latency dopaminergic signaling or at the level of task-goal representations in the prefrontal cortex. However, the latter, slow dopaminergic influences at the neocortical level, might not be the only dominant mechanism.
Error processing
The review above describes what neuronal components are necessary to consider when being interested in error commissioning. Mechanisms of error processing or more specifically adaptation are not addressed, because these processes have been formalized in well-established models and it has been shown that the ACC plays an important role (e.g., Holroyd and Coles, 2002). An overview of theories on error processing can be found in Box 1.
Bush et al. (2000) distinguished the ACC regions that are involved in “cognitive” and “emotional processing”. These authors stated that “cognitive” processing is related to the 32′, 24c′, 24b′m, and 24a′ subregions and, thus, to the dorsal part of the ACC (dACC). In contrast, emotional processing is due to regions 32, 24c, 24b, 24a, and 25 and therefore to the ventral part of the ACC (vACC). The functions of the ACC range from very basic (homeostatic) to more complex social-cognitive functions. In general, one can assume that the core function of the ACC is to establish the mobilization required to cope with cognitive and, as recently described, emotional and social demands (Mayberg et al., 1999; Elliott et al., 2000; Paus, 2001; Phillips et al., 2003) to achieve cognitive control (Koban and Pourtois, 2014). Thus, the ACC plays a role in the processing of tasks that require increasing cognitive effort and control and the management of responses when faced with conflicting demands (Luu et al., 2003; Fan et al., 2008). Furthermore, the ACC is a central structure involved in determining how to act in a goal-directed manner and is central to inhibitory control (Braver et al., 2001; Bari and Robbins, 2013), reward responses and their modulation (Amiez et al., 2005) as well as the top-down influence on primary sensory processes (Crottaz-Herbette and Menon, 2006). Together, these results suggest that the ACC is well suited for comparing the results of action selection to the task-goal representation (Bush et al., 2000). This assumption is further supported by the connections of the ACC with even distant regions in the dorsolateral prefrontal cortex (DLPFC), which are important for working memory functions (Paus, 2001). In combination with the connections with the motor areas, this extensive connectivity of the ACC with the lateral regions of the PFC may provide the basis for an interchange between the cognitive and motor systems (Paus, 2001) that is necessary to perform evaluator functions and to provide information that can be used for behavioral adaptation (Paus, 2001). This role is stressed in recent views regarding the functions of the ACC indicating that the diversity of the functions mediated by the ACC can be understood in terms of a single function, i.e., the allocation of control based on an evaluation of the expected value of control (Shenhav et al., 2013). In this sense, this error signal may function as a biasing signal (Miller and Cohen, 2001) that instructs the prefrontal networks when to maintain a given activity state (representation) (Miller, 2000; Shenhav et al., 2013). This error biasing signal has been demonstrated using fMRI and EEG data (Debener et al., 2005; Hoffmann et al., 2014). The error monitoring signal can therefore be employed to strengthen representations in the prefrontal networks. Thus, the error signal closes the circle and influences the processes described above that lead to the commissioning of an error. A brief description of the properties of the error processing signal is given in Box 2.
In this sense, the ACC may play a central role not only in the processing of errors, but also in processes preceding erroneous or successful response selection. A recent integrative account, the “expected value of control (EVC)” model (Shenhav et al., 2013) proposes that the (dorsal) ACC uses information from the current state (i.e., information of task demands, processing capacity etc.) and expected value of the outcomes to determine how much cognitive control is invested in a given task (Shenhav et al., 2013). Critically for its potential role in processes leading to errors are findings showing that above-mentioned processes supposed to be important elements in the processes that lead to an error (i.e., strength of task goal representation, attentional biases, conflict monitoring and cognitive control) are all related to the functioning of the (dorsal) ACC: the ACC differentiates task representations and response rules (e.g., Dixon and Christoff, 2012), specific actions and task sets (Hampton and O'Doherty, 2007; Haynes et al., 2007). For an excellent overview of the ACC's functions refer to Shenhav et al. (2013). In sum, there is hence ample evidence to that the ACC is involved in processes that we are proposing to be central for the understanding of how errors emerge. The ACC may thus not only be seen as an element in error processing, but also as an important entity in mechanisms that may ultimately contribute to the occurrence of errors.