Good introduction about the DMC model of working-memory.
"What is reactive control? As the name suggests, reactive control is engaged after, rather than before the occurrence of some imperative event. Prior to this event, the system remains relatively unbiased, and so is therefore more influenced by bottom-up inputs. Furthermore, reactive control mechanisms are engaged only as needed, on a "just-in- time" basis rather than consistently and in advance of critical events. Finally, when control depends upon the use of context information, the activation of such information by reactive mechanisms occurs transiently rather than in a sustained fashion, and thus decays away quickly. As a consequence, in situations when the same context must be repeatedly accessed, this must occur through full re-activation of the information each time it is needed."
"The distinction between proactive and reactive control can be thought of as a distinction between early selection and late correction (Jacoby, Kelley, & McElree, 1999). Concrete examples can help illustrate the proactive-reactive distinction. A real- world example might be the typical prospective memory situation in which an intention is formed about a behavioral goal to be completed at some later point, such as stopping at the dry cleaners after work. A proactive control strategy would require the goal information to be actively sustained from the time the intention is formed until the goal is satisfied (e.g., the end of the day). The usefulness of such a proactive strategy is that plans and behaviors can be continually adjusted to facilitate optimal completion of the goal (e.g., not scheduling a late meeting). In contrast, with a reactive control strategy the goal would only be transiently activated at the time of intention, and then need to be re- activated again by an appropriate trigger event (e.g., opening the car door). Because of this need for repeated re-activation, there is greater dependence on the trigger events themselves, since if these are insufficiently salient or discriminative they will not drive re-activation (e.g., the dry cleaning errand might only be remembered because of cleaning ticket left on the car seat)."
ErikThor likes this post.
Published: 02-04-2015 04:32 am
Updated: 02-04-2015 04:33 am
From Page 15 - 20:
Below, we list some of the limitations and disadvantages of proactive control, followed by the negative consequences of reactive control:
* Proactive control requires the presence of predictive contextual cues. Many times predictive contextual information is not present in the environment , and as such, control can not be prepared in advance. In these circumstances, the only possible control strategy is a reactive one.
* Proactive control requires predictive contextual cues to be highly reliable. In situations where predictive cues turn out to be invalid, there can be a strong cost if the cue-based contextual information is used as a basis for proactive control. Such cue invalidity costs can be seen in a range of cognitive situations (e.g., the Posner spatial cued reaction time task, Posner, Snyder, & Davidson, 1980). Thus, adoption of a proactive control strategy is only likely in situations where contextual cues serve as highly reliable predictors of upcoming events or required actions.
* Proactive control is metabolically costly. According to our theoretical model, the active maintenance of goal-relevant information requires a high and sustained level of neuronal activity in lateral PFC during the entire retention interval. Such extended periods of high firing are likely to require additional metabolic resources (e.g., for glucose consumption, waste removal, neurotransmitter recycling, etc.) that may not always be available, or at the minimum, reduce the amount available for other purposes. Even without considering metabolic requirements directly, it seems clear that proactive control is capacity demanding, since only a small number of goals can be actively maintained in the focus of attention (Cowan, 2001). Thus, proactive control draws away resources from other active maintenance demands. Consequently , it is likely to only be used if sufficient capacity is available (i.e., other WM demands are low, general cognitive resources are high, and cortical arousal is optimal).
* Proactive control is prohibitive with very long retention intervals. Because the sustained, active context maintenance associated with proactive control is so resource-demanding, the longer the interval between the maintenance initiation and context utilization, the less feasible this strategy becomes. Thus, proactive control is unlikely with retention intervals longer than a few minutes. Certain prospective memory tasks are the best example of situations in which the interval between goal formation and goal realization can be hours or days. In such situations, actions are accomplished through reactive control – that is, by transiently re-activating the goal (via episodic retrieval) at the appearance of an appropriate trigger stimulus such that it can bias the goal-relevant behavior. This idea is consistent with the bulk of the prospective memory literature, which has suggested that retrospective processes are the primary mechanisms guiding delayed-intention behavior (Einstein & McDaniel, 1996). Nevertheless, recent studies have begun to suggest that in certain experimental prospective memory paradigms, preparatory control may be occurring, even across longer time-scales of retention (R. E. Smith, 2003) . It will be important to determine more conclusively whether, and under what constraints, active maintenance processes are being used in such paradigms.
* Proactive control is less sensitive to changes in reward/punishment contingencies. Because environments are typically non-stationary, contingencies can often change without warning. The active representation of goals during proactive control modes causes the system to be biased to attend primarily to goal-relevant features of the environment, and to be predisposed to interact with these features in a goal-driven manner. This leads to a reduction in incidental encoding of goal-irrelevant or goal-incongruent features, which may, in fact, serve as cues that the environment is changing. Theorists have suggested that continuous monitoring of environmental (or internal) background information is a critical function of motivationally-oriented neural systems. For example, such mechanisms can lead to optimal detection of low-probability but potential threats (Goschke, 2003) . Thus, high-demands or a pre- existing bias for background monitoring (such as when vigilance towards potential threats is required) will make the utilization of proactive control less likely.
* Proactive control impedes the natural progression towards automatization. There is a fundamental tension between the exertion of cognitive control and the development of automaticity, which has been termed the “control dilemma” (Goschke, 2003) . Because automatic processes are robust, fast and efficient, it is likely that there is an inherent computational pressure or bias on the cognitive system to automatize processing wherever possible, via strengthening of internal associations and stimulus-response bindings. Proactive control processes oppose such mechanisms by providing a sustained top-down flow of information that enables contextual goals to override default processing. Indeed , it is reactive rather proactive control that allows for the best optimization of the control dilemma, by introducing a highly transient and minimalist (i.e., only-as-needed) form of intervention , that allows habits, skills, and procedures to be learned while still enabling the system to override these forces if necessary.
* Reactive control is more susceptible to proactive interference (PI) Control mechanisms are necessary because many times the effects of past experience conflict with current goals. However, such sources of proactive interference (PI) cannot be completely counteracted by a reactive control strategy. This is because reactive control is initiated by post-stimulus processing, such that potentially interfering stimulus-based associations will already be activated by the time control mechanisms are engaged. In contrast, proactive control may lead to complete suppression of PI, via optimal attentional configuration. Thus, in conditions where PI effects are very strong and the costs of interference are high, the disadvantages of reactive control will be most apparent.
* Reactive control is sub-optimal when stimulus-driven processing is insufficient. Because reactive control is stimulus-driven rather than preparatory, it is always a sub-optimal control strategy. However, the limitations of reactive control are most prominent in conditions where perceptual information is weak, response selection parameters are underdetermined, and/or when there is a premium on optimal performance (i.e., high speed and accuracy constraints).
* Reactive control does not maximize rewards Maximizing reward often depends upon the ability to predict its occurrence and magnitude. Proactive control aids in maximizing rewards through the utilization of predictive contextual cues that can bias action selection. The strong link between proactive control and reward prediction can be seen in the phasic DA signals that are postulated to engage proactive control processes in PFC, according to the DMC account, and also appear to signal reward-related salience of predictive cues, according to influential reinforcement learning models (Schultz et al., 1997). In contrast , reactive control is not geared toward maximization of rewards but rather towards resolving interference and facilitating the transition to automaticity. Thus, in conditions where processing is oriented towards reward maximization, and where reward attainment depends on precise focusing of attention or optimal response preparation, a reactive control strategy will be highly disadvantageous.
ErikThor likes this post.
Published: 02-05-2015 06:12 am
Super interesting, especially the part about the energy requirements when proactive control, and the problems of Execution vs Automatization and how Execution responds to Automatized behavior vs controlled proactive behavior. Speculation here: would you say the control level is done through Cognitive responses, and that the automatic, non-controlled reactive/proactive behavior is a result of Affective responses?
Christian likes this post.
Published: 02-05-2015 08:06 am
Great point, didn't think about that. Yes I think so, Affective Executors are probably more content with automatic proactive behavior than Cognitive Executors.