What do you think?
Rate this book
Cognitive Neuroscience of Attention
This volume presents the latest advances in understanding its anatomy, circuitry, functions, and deficits. Outstanding investigators have written brief yet substantive chapters in which they not only summarize key findings but also illuminate their goals and the directions their research is taking. Coverage includes different cognitive models of attention; knowledge emerging from functional imaging and genetic studies; and neurophysiological, developmental, and neuropsychological approaches. Emerging knowledge is presented on processes that impair or alter attention, and clinical implications are discussed. Linking many levels of analysis, and featuring over 100 illustrations, the book moves us closer to a coherent view of the attentional system and the key role it plays in everyday life.
466 pages, Hardcover
First published July 16, 2004
2 people are currently reading
68 people want to read
Ratings & Reviews
Friends & Following
Create a free account to discover what your friends think of this book!
Community Reviews
Displaying 1 of 1 review
January 13, 2026
My computer crashed so I don’t have all my notes but this was a thrilling read. Some of the papers in there are insanely interesting.
Cavanagh, “Attention routines and the architecture of selection”. Spotlight metaphor misleading: more than one spotlight, possibility of multiple object tracking (up to 4 or 5). We don’t have access to raw information within the selection region, but more to something like texture, vulnerable to crowding effects. For instance, a ‘+’ sign with an N next to it: you can identify the N while focusing on the +, and we can draw a circle around the N, representing the hypothetical width of your focus. But take the sign, and put an entire word, like ‘Posner’, with the ‘n’ at the same distance from the + as before, and it’s not possible to identify the N. N should be available, along with S and E. But this is not what happens (p. 17). The width of the selection field is unaffected by the number of items tracked. What happens with crowding is that several items are fused together, and cannot be identified compositionally: there is a minimal spacing requirement for object individuation to be possible (p. 19).
Klein, “Control of visual orienting”. Removal of attended or unattended peripheral stimulus before target increases RTs (the “gap effect”). Why? “Any offset serves as a warning signal and results in generalized alertness, and fixation offsets produce added facilitation by exogenously disengaging the oculo-motor system from fixation” (p. 31). The hypothesis, borrowed from Fischer & Breitenmeyer (1987), is that attentional shifts involve disengagement, movement, and re-engagement with a new stimulus at the new location, and that this process is made easier (if not triggered?) by the removal of the fixation. Exogenous triggers are more efficient, then, than endogenous decisions? This is what Taylor et al. (1998) indicates: they compared performances of subjects with an anticipatory tone (triggering endogenous attentional shift) with fixation offsets (exogenous), and found that “removal of fixation results in a further 40 to 45-msec decrease in SRT around 100 msec after its disappearance. We take this decrease… to be due to the exogenous disengagement of the oculomotor system.” (id.).
Some facts on exogenous vs endogenous attention:
• Exogenous attention is faster,
• people with parietal damage exhibit disengagement deficit with exogenous attention, not endogenous attention,
• inhibition of return (IOR) is a purely exogenous effect,
• and, most surprising of all to me, stimulus enhancement (≠ noise reduction) only occurs with exogenous attention (Lu & Dosher 2000).
Oculomotor readiness hypothesis: “endogenous covert orienting of attention is accomplished by preparing to move the eyes to the to-be-attended location” (p. 38, quoting Klein 1980). Implied by Rizzolatti’s premotor theory (Rizzolatti et al. 1987): getting ready to move the eyes is the mechanism by which attention is covertly oriented to the location in advance of stimulation. This body of theories treats endogenous overt and endogenous covert orienting as one system, and predicts that eye movement to, and detection of events at the location should be facilitated (= more efficient target processing). Both predictions have been disconfirmed (e.g. Klein & Pontrefact 1994) however, motivating the view that endogenous orienting is split into two subsystems, overt and covert.
Fuentes, “Inhibitory proessing in the attentional networks”. Interesting properties of IOR: no Stroop effect, i.e. semantic processing is not compromised, and so no semantic priming occurs, when primes were at locations subject to IOR, i.e. ‘inhibited’ locations—but the effect is constrained in a 250 msec window: with prime-target intervals > 250 msec, the inhibition effect vanishes. This is called ‘inhibitory tagging’ (IT, Fuentes et al. 1999). IT is associated with a functioning executive network (= the ACC and PFC, cf. DiGirolamo & Posner 1996). Stroop interference is about conflict between the orienting network (who has to orient to the color) and the executive network (that has to shut down response to the irrelevant but prepotent semantic information, i.e. the word).
Localization of IOR in the superior colliculus (SC): IOR deficiency in cerebral palsy patients for displays of cue-targets presented vertically (Posner et al. 1985). Parietal lobes also play a (strange) role: we take patients with parietal lesion, and find preserved IOR for stimuli presented in the contralesional, but not the ipsilesional field (Vivas, Humphreys & Fuentes 2003). For instance, Stroop effect only worked for stimuli in the contralesional field. The interpretation is that IT requires IOR, and that IT is a necessary condition for Stroop interference reduction. Interestingly, however, IT is not sufficient for Stroop interference reduction: patients with schizophrenia show both intact IOR, but no Stroop interference reduction for stimuli presented at locations subject to IOR (in both hemifields) (Fuentes et al. 2000). So IOR and IT are distinct.
Another task (Fuentes et al. 1998) involved semantic inhibition: one word (hand), then either an intermediate string of Xs or another word (of a different lexical category), then a word of the same semantic category as the first word (finger): in controls, you find shorter RTs when the intermediate stimulus is a string of Xs. Interpretation: inhibition of return (after disengagement) to a previously activated lexical category. In schizophrenics, you only find this pattern for targets presented in the left visual field, suggesting that semantic processing is not affected by schizophrenia, but that the inhibitory function of the executive network is, albeit in a specifically lateralized way (p. 50).
Carr, “A multi-level approach to selective attention”. Carr focuses on the blood-oxygen-level-depependent physiological cost of focusing attention on one object vs an unbounded region of space in which events may be expected to occur, and on the additional costs of re-deploying/engagement of attention to a different location after its initial engagement at one location has proven unsuccessful (measured in ‘wrong’ trials) (p. 58). This seems to indicate that there are two, perhaps mutually reinforcing pressures, both not to be too precise in attentional resource allocation, and to get it right.
But when it comes to object perception, there is going to be a need for deep processing beyond the level of unbounded spatial region. The problem with that is: if the selection for deep processing comes early, there is the risk of leaving potentially relevant information unattended, and if it occurs late, then there are at least two problems: first, a need for parallel deep stimulus processing, or ‘crosstalk’, which is known to degrade information in proportion to the similarity of the information being handled (p. 59 and refs), and second, the problem of ‘selection for action’, namely the more stimuli are being processed, the higher the likelihood of distraction (of the decision, motor programming etc. systems) by nontarget stimuli, aka ‘flankers’. Thus, the alternative created by the selection-for-action problem is one between slow or wrong action generation (p. 60).
What happens with flankers is distraction, but for distraction to occur, the distractor must be processed, and the possibility for such out-of-left-field processing suggests that attention is never so focused that it inhibits the processing of unexpected stimuli: the attentional field always remains open enough. Thus, there is no choice made between early or late selection absolutely: there is “a dynamic solution in which the locus of selection (early vs late in the stream of information processing) varies from situation to situation as a function of a theoretical construct called perceptual load (Lavie, 1995).” (p. 60). Perceptual load is roughly the ratio of relevant information by distractors. Now we can vary experiments with flankers in and out of the search array, i.e. we can use flankers outside the search array to learn about how early selection happens, assuming, e.g., that sensorily salient, but semantically irrelevant, ‘flankers’ outside the visual search space will be processed in depth (i.e. semantically) if early selection is relaxed enough for purely sensory stimulus salience to motivate deeper processing—as opposed to a case where sensory salience, because it makes the flanker visually resolvable, works in the opposite direction, and causes the flanker to be ignored. Empirically, the prediction is that a first case scenario signals itself by delayed RTs. The results are that selection occurs earlier as perceptual load increases: when the perceptual load is quite small, everything gets automatically processed, regardless of task relevance, and therefore distraction by flankers is quite likely. As perceptual load increases “and begins to exhaust perceptual capacity”, however, we begin to select more and more abstract features of the items in the search space. “Selection shifts from late to early and comes under the control of processes that respond to sensory and perceptual features rather than identity, meaning, and afforded actions. As early seletion comes to dominate input selection, the flanker receives less processing, reducing its impact on performance.” (p. 61, my emphasis). (Note to myself: This may explain the gorilla experiment in inattentional blindness, where the setup is high perceptual load.)
The shift from late to early selection seems to happen earlier (i.e. earlier selection for lower perceptual load) in older people, i.e. they are more likely to ignore flankers even when under low perceptual load. In other words, older people’s perceptual capacities are more quickly exhausted. Same thing with young children: attentional development appears to be an inverse U shape. The path from young children to adult is later onset of flanker-induced distraction reduction via earlier shift from late to early selection, whereas the path from adulthood to old age seems to be marked by earlier onset…, via later shift… (p. 62). ADHD does not seem to be traceable to these measures, but rather to executive control dysfunction, and deficits in maintaining arousal, i.e. “a generalized readiness to respond to any environment input or task goal, independently of the specifics of information content” (p. 62).
Issue of generalized pattern of activation: no one representation comes to dominate. Attention is needed:
Analogy with ‘attention’ in semantic memory space: just like attention to an unbounded region of space can be sufficient for visually guided action to succeed, ‘attention’ to or activation of a general neighborhood in semantic memory may be enough for the purposes at hand. It is also the case that some tasks will not make either one object, or one item in semantic memory ‘pop out’, resulting in a general pattern of co-activation. “In these cases, an attentional process may be needed to focus on one representation and facilitate its retrieval by enhancing the activation of that representation and shielding it from competitive interactions with other representations—again, much as attention works to facilitate information from one stimulus in space while shielding it from competitive interactions with near neighbors in space (Desimone & Duncan, 1995).” (p. 64).
But this attentional enhancement is at the expense of the availability of the rest of the semantic neighborhood for immediate future processing:
Weakly learned new words (e.g. drupe = a cherry) are used as primes in a semantic priming experiments. After being primed with such new words, subjects are asked to make word/nonword judgements for stimuli semantically associated with the words (e.g. peach vs *deach is semantically associated with drupe, which means ‘a cherry’, because peaches and cherries/drupes are fruits). It was found (Dagenbach, Carr & Barnhart 1990) that the newly learned primes whose meanings were most difficult to retrieve (the more weakly learned ones), far from having priming effects, i.e. accelerating RTs, had negative priming effects, i.e. delayed and inihibited responses (p. 65). Why? Because the individual has been primed with one of the weakly learned new words first, and this forced him to retrieve the meaning of the word, consisting of raising activation of the relevant representation (e.g. fruit) above threshold for retrieval, which can only be done by also inhibiting co-activation of nearby representations in semantic space. For instance, then, drupe is only going to allow the retrieval of cherry at the expense of the very availability of, say, peach. Thus, when, a short time later, peach needs activation in the peach vs *deach word/nonword task, RT is delayed (pp. 65-66). Results were replicated in category judgement tasks: weakly learned new primes inhibited performance about older, better learned exemplars of the same category in categorization tasks (appendix, p. 67). “Thus it appears that a center-surround attentional mechanism operates in the service of selection among activated semantic representations from memory, analogous to a similar mechanism observed to operate during selection of activated object representations from environmental space.” (p. 66, my emphasis).
Cohen, Aston-Jones & Gilzenrat, “A systems-level perspective on attention and cognitive control”. Adaptive gating: the ventral tegmental area (VTA, in the brainstem) implements an adaptive gating mechanism that (i) regulates the extent to which the associative layer (where?) is allowed to feed input to the task control layer in the PFC by comparing the input to predicted reward (signals), (ii) while training connections between the associative layer and the VTA, specifically stengthening the connections between the two for cues that successfully predict reward via the mesolimbic dopaminergic pathway. The ACC plays a role in this model as a conflict monitor: when conflict, say, between color and word meaning (Stroop test) occurs, the ACC increases the level of representation activation in the PFC, triggering a wave of top-down control.
As conflict increases in the task, however (e.g. as color information in the Stroop test degrades), more top-down control is going to be needed. But at some point, withdrawal from the task is going to have to be preferred. This is known as the exploration vs exploitation problem: when the trade-off between the resources involved in and the reward derived from the implementation of a behavioral program becomes disadvantageous, further behaviour sampling should be preferred (p. 82). It is further hypothesized that interactions between the ACC and the locus coeruleus (LC) mediates this switch, by itself switching between two modes of activity (‘tonic’ and ‘phasic’), defined by their respective ‘levels of discharge’ (= activation thresholds?), and thus their susceptibility to distractors. See Aston-Jones’s research (e.g. Aston-Jones et al. 1994).
Roughly speaking, we can account for attentional focusing at different levels, in part, by looking at activation thresholds for neurons in different parts of the brain: “In the phasic mode [i.e. high threshold, engaged behaviour], the LC functions as an attentional filter that selects for the occurrence (i.e., timing) of task-relevant stimuli, much as cortical attentional systems filter the content of a stimulus.” (p. 83, original emphasis). Switch of LC activation mode therefore allows the sampling of other behavioral program (i.e., at the behavioral level, the switch from exploitation to exploration) via an adaptive decrease of activation thresholds, and therefore an increase of susceptibility to distractors (“the system is more effectively driven by task-irrelevant stimuli”, p. 83). “Viewed from the perspective of attentional control, the LC phasic mode supports the current control state (exploitation), while the LC tonic mode provokes a withdrawal of control from the current task, favoring the sampling of other behavioral goals (exploration), which raises one more important question: What information can the system use to determine whether it should exploit (LC phasic mode) or explore (LC tonic mode)?” (p. 83, my emphasis). The answer to the question is, again, the ACC, provided it can integrate data on conflict over both short and longer time frames—since what motivates a withdrawal would be a situation where reward is consistently low while conflict remains high.
Hope I can edit this and add the rest of my notes when I get the data from my laptop—if I do
Cavanagh, “Attention routines and the architecture of selection”. Spotlight metaphor misleading: more than one spotlight, possibility of multiple object tracking (up to 4 or 5). We don’t have access to raw information within the selection region, but more to something like texture, vulnerable to crowding effects. For instance, a ‘+’ sign with an N next to it: you can identify the N while focusing on the +, and we can draw a circle around the N, representing the hypothetical width of your focus. But take the sign, and put an entire word, like ‘Posner’, with the ‘n’ at the same distance from the + as before, and it’s not possible to identify the N. N should be available, along with S and E. But this is not what happens (p. 17). The width of the selection field is unaffected by the number of items tracked. What happens with crowding is that several items are fused together, and cannot be identified compositionally: there is a minimal spacing requirement for object individuation to be possible (p. 19).
Klein, “Control of visual orienting”. Removal of attended or unattended peripheral stimulus before target increases RTs (the “gap effect”). Why? “Any offset serves as a warning signal and results in generalized alertness, and fixation offsets produce added facilitation by exogenously disengaging the oculo-motor system from fixation” (p. 31). The hypothesis, borrowed from Fischer & Breitenmeyer (1987), is that attentional shifts involve disengagement, movement, and re-engagement with a new stimulus at the new location, and that this process is made easier (if not triggered?) by the removal of the fixation. Exogenous triggers are more efficient, then, than endogenous decisions? This is what Taylor et al. (1998) indicates: they compared performances of subjects with an anticipatory tone (triggering endogenous attentional shift) with fixation offsets (exogenous), and found that “removal of fixation results in a further 40 to 45-msec decrease in SRT around 100 msec after its disappearance. We take this decrease… to be due to the exogenous disengagement of the oculomotor system.” (id.).
Some facts on exogenous vs endogenous attention:
• Exogenous attention is faster,
• people with parietal damage exhibit disengagement deficit with exogenous attention, not endogenous attention,
• inhibition of return (IOR) is a purely exogenous effect,
• and, most surprising of all to me, stimulus enhancement (≠ noise reduction) only occurs with exogenous attention (Lu & Dosher 2000).
Oculomotor readiness hypothesis: “endogenous covert orienting of attention is accomplished by preparing to move the eyes to the to-be-attended location” (p. 38, quoting Klein 1980). Implied by Rizzolatti’s premotor theory (Rizzolatti et al. 1987): getting ready to move the eyes is the mechanism by which attention is covertly oriented to the location in advance of stimulation. This body of theories treats endogenous overt and endogenous covert orienting as one system, and predicts that eye movement to, and detection of events at the location should be facilitated (= more efficient target processing). Both predictions have been disconfirmed (e.g. Klein & Pontrefact 1994) however, motivating the view that endogenous orienting is split into two subsystems, overt and covert.
Fuentes, “Inhibitory proessing in the attentional networks”. Interesting properties of IOR: no Stroop effect, i.e. semantic processing is not compromised, and so no semantic priming occurs, when primes were at locations subject to IOR, i.e. ‘inhibited’ locations—but the effect is constrained in a 250 msec window: with prime-target intervals > 250 msec, the inhibition effect vanishes. This is called ‘inhibitory tagging’ (IT, Fuentes et al. 1999). IT is associated with a functioning executive network (= the ACC and PFC, cf. DiGirolamo & Posner 1996). Stroop interference is about conflict between the orienting network (who has to orient to the color) and the executive network (that has to shut down response to the irrelevant but prepotent semantic information, i.e. the word).
Localization of IOR in the superior colliculus (SC): IOR deficiency in cerebral palsy patients for displays of cue-targets presented vertically (Posner et al. 1985). Parietal lobes also play a (strange) role: we take patients with parietal lesion, and find preserved IOR for stimuli presented in the contralesional, but not the ipsilesional field (Vivas, Humphreys & Fuentes 2003). For instance, Stroop effect only worked for stimuli in the contralesional field. The interpretation is that IT requires IOR, and that IT is a necessary condition for Stroop interference reduction. Interestingly, however, IT is not sufficient for Stroop interference reduction: patients with schizophrenia show both intact IOR, but no Stroop interference reduction for stimuli presented at locations subject to IOR (in both hemifields) (Fuentes et al. 2000). So IOR and IT are distinct.
Another task (Fuentes et al. 1998) involved semantic inhibition: one word (hand), then either an intermediate string of Xs or another word (of a different lexical category), then a word of the same semantic category as the first word (finger): in controls, you find shorter RTs when the intermediate stimulus is a string of Xs. Interpretation: inhibition of return (after disengagement) to a previously activated lexical category. In schizophrenics, you only find this pattern for targets presented in the left visual field, suggesting that semantic processing is not affected by schizophrenia, but that the inhibitory function of the executive network is, albeit in a specifically lateralized way (p. 50).
Carr, “A multi-level approach to selective attention”. Carr focuses on the blood-oxygen-level-depependent physiological cost of focusing attention on one object vs an unbounded region of space in which events may be expected to occur, and on the additional costs of re-deploying/engagement of attention to a different location after its initial engagement at one location has proven unsuccessful (measured in ‘wrong’ trials) (p. 58). This seems to indicate that there are two, perhaps mutually reinforcing pressures, both not to be too precise in attentional resource allocation, and to get it right.
But when it comes to object perception, there is going to be a need for deep processing beyond the level of unbounded spatial region. The problem with that is: if the selection for deep processing comes early, there is the risk of leaving potentially relevant information unattended, and if it occurs late, then there are at least two problems: first, a need for parallel deep stimulus processing, or ‘crosstalk’, which is known to degrade information in proportion to the similarity of the information being handled (p. 59 and refs), and second, the problem of ‘selection for action’, namely the more stimuli are being processed, the higher the likelihood of distraction (of the decision, motor programming etc. systems) by nontarget stimuli, aka ‘flankers’. Thus, the alternative created by the selection-for-action problem is one between slow or wrong action generation (p. 60).
What happens with flankers is distraction, but for distraction to occur, the distractor must be processed, and the possibility for such out-of-left-field processing suggests that attention is never so focused that it inhibits the processing of unexpected stimuli: the attentional field always remains open enough. Thus, there is no choice made between early or late selection absolutely: there is “a dynamic solution in which the locus of selection (early vs late in the stream of information processing) varies from situation to situation as a function of a theoretical construct called perceptual load (Lavie, 1995).” (p. 60). Perceptual load is roughly the ratio of relevant information by distractors. Now we can vary experiments with flankers in and out of the search array, i.e. we can use flankers outside the search array to learn about how early selection happens, assuming, e.g., that sensorily salient, but semantically irrelevant, ‘flankers’ outside the visual search space will be processed in depth (i.e. semantically) if early selection is relaxed enough for purely sensory stimulus salience to motivate deeper processing—as opposed to a case where sensory salience, because it makes the flanker visually resolvable, works in the opposite direction, and causes the flanker to be ignored. Empirically, the prediction is that a first case scenario signals itself by delayed RTs. The results are that selection occurs earlier as perceptual load increases: when the perceptual load is quite small, everything gets automatically processed, regardless of task relevance, and therefore distraction by flankers is quite likely. As perceptual load increases “and begins to exhaust perceptual capacity”, however, we begin to select more and more abstract features of the items in the search space. “Selection shifts from late to early and comes under the control of processes that respond to sensory and perceptual features rather than identity, meaning, and afforded actions. As early seletion comes to dominate input selection, the flanker receives less processing, reducing its impact on performance.” (p. 61, my emphasis). (Note to myself: This may explain the gorilla experiment in inattentional blindness, where the setup is high perceptual load.)
The shift from late to early selection seems to happen earlier (i.e. earlier selection for lower perceptual load) in older people, i.e. they are more likely to ignore flankers even when under low perceptual load. In other words, older people’s perceptual capacities are more quickly exhausted. Same thing with young children: attentional development appears to be an inverse U shape. The path from young children to adult is later onset of flanker-induced distraction reduction via earlier shift from late to early selection, whereas the path from adulthood to old age seems to be marked by earlier onset…, via later shift… (p. 62). ADHD does not seem to be traceable to these measures, but rather to executive control dysfunction, and deficits in maintaining arousal, i.e. “a generalized readiness to respond to any environment input or task goal, independently of the specifics of information content” (p. 62).
Issue of generalized pattern of activation: no one representation comes to dominate. Attention is needed:
Analogy with ‘attention’ in semantic memory space: just like attention to an unbounded region of space can be sufficient for visually guided action to succeed, ‘attention’ to or activation of a general neighborhood in semantic memory may be enough for the purposes at hand. It is also the case that some tasks will not make either one object, or one item in semantic memory ‘pop out’, resulting in a general pattern of co-activation. “In these cases, an attentional process may be needed to focus on one representation and facilitate its retrieval by enhancing the activation of that representation and shielding it from competitive interactions with other representations—again, much as attention works to facilitate information from one stimulus in space while shielding it from competitive interactions with near neighbors in space (Desimone & Duncan, 1995).” (p. 64).
But this attentional enhancement is at the expense of the availability of the rest of the semantic neighborhood for immediate future processing:
Weakly learned new words (e.g. drupe = a cherry) are used as primes in a semantic priming experiments. After being primed with such new words, subjects are asked to make word/nonword judgements for stimuli semantically associated with the words (e.g. peach vs *deach is semantically associated with drupe, which means ‘a cherry’, because peaches and cherries/drupes are fruits). It was found (Dagenbach, Carr & Barnhart 1990) that the newly learned primes whose meanings were most difficult to retrieve (the more weakly learned ones), far from having priming effects, i.e. accelerating RTs, had negative priming effects, i.e. delayed and inihibited responses (p. 65). Why? Because the individual has been primed with one of the weakly learned new words first, and this forced him to retrieve the meaning of the word, consisting of raising activation of the relevant representation (e.g. fruit) above threshold for retrieval, which can only be done by also inhibiting co-activation of nearby representations in semantic space. For instance, then, drupe is only going to allow the retrieval of cherry at the expense of the very availability of, say, peach. Thus, when, a short time later, peach needs activation in the peach vs *deach word/nonword task, RT is delayed (pp. 65-66). Results were replicated in category judgement tasks: weakly learned new primes inhibited performance about older, better learned exemplars of the same category in categorization tasks (appendix, p. 67). “Thus it appears that a center-surround attentional mechanism operates in the service of selection among activated semantic representations from memory, analogous to a similar mechanism observed to operate during selection of activated object representations from environmental space.” (p. 66, my emphasis).
Cohen, Aston-Jones & Gilzenrat, “A systems-level perspective on attention and cognitive control”. Adaptive gating: the ventral tegmental area (VTA, in the brainstem) implements an adaptive gating mechanism that (i) regulates the extent to which the associative layer (where?) is allowed to feed input to the task control layer in the PFC by comparing the input to predicted reward (signals), (ii) while training connections between the associative layer and the VTA, specifically stengthening the connections between the two for cues that successfully predict reward via the mesolimbic dopaminergic pathway. The ACC plays a role in this model as a conflict monitor: when conflict, say, between color and word meaning (Stroop test) occurs, the ACC increases the level of representation activation in the PFC, triggering a wave of top-down control.
As conflict increases in the task, however (e.g. as color information in the Stroop test degrades), more top-down control is going to be needed. But at some point, withdrawal from the task is going to have to be preferred. This is known as the exploration vs exploitation problem: when the trade-off between the resources involved in and the reward derived from the implementation of a behavioral program becomes disadvantageous, further behaviour sampling should be preferred (p. 82). It is further hypothesized that interactions between the ACC and the locus coeruleus (LC) mediates this switch, by itself switching between two modes of activity (‘tonic’ and ‘phasic’), defined by their respective ‘levels of discharge’ (= activation thresholds?), and thus their susceptibility to distractors. See Aston-Jones’s research (e.g. Aston-Jones et al. 1994).
Roughly speaking, we can account for attentional focusing at different levels, in part, by looking at activation thresholds for neurons in different parts of the brain: “In the phasic mode [i.e. high threshold, engaged behaviour], the LC functions as an attentional filter that selects for the occurrence (i.e., timing) of task-relevant stimuli, much as cortical attentional systems filter the content of a stimulus.” (p. 83, original emphasis). Switch of LC activation mode therefore allows the sampling of other behavioral program (i.e., at the behavioral level, the switch from exploitation to exploration) via an adaptive decrease of activation thresholds, and therefore an increase of susceptibility to distractors (“the system is more effectively driven by task-irrelevant stimuli”, p. 83). “Viewed from the perspective of attentional control, the LC phasic mode supports the current control state (exploitation), while the LC tonic mode provokes a withdrawal of control from the current task, favoring the sampling of other behavioral goals (exploration), which raises one more important question: What information can the system use to determine whether it should exploit (LC phasic mode) or explore (LC tonic mode)?” (p. 83, my emphasis). The answer to the question is, again, the ACC, provided it can integrate data on conflict over both short and longer time frames—since what motivates a withdrawal would be a situation where reward is consistently low while conflict remains high.
Hope I can edit this and add the rest of my notes when I get the data from my laptop—if I do
Displaying 1 of 1 review