Dual mechanisms of cognitive control

Todd BraverCognitive Control and Psychopathology Lab,

Washington University

Plus:

Jeremy Reynolds

Josh Brown

Stefan van der Stigchel

2

Cognitive Control

• What is cognitive control?– formation, maintenance and realization of internal goals– structuring thoughts and actions in accordance with these goals

• The problem: Interference– the environment often provides conflicting or distracting information– perceptually salient information or default action tendencies are often incongruent

with intended goals

– Examples: Action capture errors, RED

• A major function of control processes is to successfully manage interference

– What are the underlying neural & cognitive mechanisms?

3

Dual mechanisms of control

• Our hypothesis => Dual mechanisms– Reactive control: Detection and resolution of interference– Proactive control: Anticipation and prevention of interference

• Reactive control– Transient, stimulus-driven activation of lateral prefrontal cortex (PFC) and other areas due to:

Spreading activation processes Detection of conflict (anterior cingulate cortex)

– Late correction mechanism: suppression of goal-irrelevant information when needed or just-in-time– Example: remember to stop at store after work (e.g. prospective memory); event-based triggering

• Proactive control:– Sustained active maintenance of goal-related information in lateral PFC

dependent upon phasic dopamine (DA) activity– Early selection mechanism: top-down attentional biases enhance access to goal-relevant information or

prepare appropriate actions– Example: Remembering your point in a conversation while others are speaking

4

Why dual mechanisms?

• Cost-benefit analysis– Proactive control:

More effective, but More energetically demanding More vulnerable to disruption (precise DA dynamics required)

– Reactive control More susceptible to interference effects (late correction), but Less demanding, more robust

• A mixture model– Both control strategies are used but weightings can differ:

Across task situations: Opportunity & Impact Across individuals: May be dependent upon biases and capabilities

– May even be trial-to-trial variability (e.g,. natural fluctuations) Task-switching

5

Research Strategy

• A cognitive neuroscience approach– Cognitive Experiments: Manipulation of control strategy, look for behavioral

markers– Brain Imaging: Dynamics of activity in lateral PFC, ACC, other regions– Special populations: Effects due to normal individual differences

(personality, intelligence), impaired functions (older adults, schizophrenia)– Computational Modeling: Mechanistic explanations, bridging brain &

behavior

• Domains– Working memory (e.g., Sternberg task)– Controlled attention (e.g., Stroop)– Task-switching

6

Task-switching

• Key Issues– Switch costs imply interference from other task sets– Preparatory interval effects on switch cost implies proactive control is being used (Meiran)

Task-set (goals) are actively maintained and used to bias attention– Residual switch costs with long preparatory intervals suggest this can’t be full account

• Accounts– Task-set inertia (Allport): Emphasis on unresolved interference rather than proactive control– Exogenous cuing (Monsell): Proactive control is incomplete until target presented– Failure-to-engage (DeJong): Probabilistic proactive control; intermittent failures

• Problem– If no proactive control, how come interference effects are not catastrophic?– Ambiguous targets (bivalent, incongruent) can only be performed appropriately if task-set

information is accesssible (e.g., actively maintained)

7

Our hypothesis

• Prepared trials = proactive control– Actively maintained PFC representations lead to preparatory biasing prior to target

stimuli– Updating and maintenance in PFC triggered by phasic dopamine burst occurring

during task cue presentation

• Unprepared trials = reactive control– Transient activation of PFC during cue presentation, but no updating and

maintenance Dopamine system is noisy; no phasic response on some trials

– Target-driven reactivation of cue leads to late biasing of relevant task pathways

• Goal– Develop computational model to demonstrate mechanisms– Demonstrate that model can capture task-switching behavioral phenomena

Switch costs, congruency costs, etc. De Jong mixture analysis: prepared (fast) trials show no switch costs,

unprepared (slow) trials show maximal switch costs– Demonstrate that model can capture task-switching brain activity dynamics

8

Experimental Paradigm

• Unpredictable (trial-by-trial) cueing, letter-digit task– Long preparatory interval (1500 msec)

ConsVowel

TASK A

Odd Even

TASK B

LETTER

X 9

Time TimeNUMBER

X 9

9

Neural network model

10

Simulations

QuickTime™ and a Video decompressor are needed to see this picture. QuickTime™ and a Video decompressor are needed to see this picture.

Proactive control (prepared) Reactive control (unprepared)

11

Simulation Mechanisms

• DA-based updating is probabilistic (50% of trials)– Results in mixture of unprepared and prepared trials– DA gating enables active maintenance

• Co-Activation leads to associative (Hebbian) strengthening (priming)

• Prepared trials (proactive control)– Task set information is actively maintained during preparatory interval (in PFC module)– Biases appropriate task-pathway– Local competition leads to suppression of irrelevant task set and pathway– Reduced interference effects (e.g., switch costs)

• Unprepared trials (reactive control)– Task-set information transiently activated with cue, leads to associative strengthening

(priming) with task pathway– Task-set decays during preparatory interval– Target presentation causes reactivation of appropriate task-set– Task-set interference is greater on switch trials and incongruent trials– Increased interference effects (congruency & switch costs)

12

Error Rate Data

13

RT data

14

RT distribution data

QuickTime™ and a TIFF (Uncompressed) decompressor are needed to see this picture.

15

Mixture Analysis

Nonswitch(plus switch -gating) Switch-no gating

Switch-all trials

16

Preparation & Errors

Unprepared Prepared

17

Simulated PFC Activity

18

Actual PFC ActivityBraver, Reynolds, and Donaldson (2003) Neuron

19

Summary

• Task-switching effects might reflect dual mechanisms of cognitive control– Proactive control: Active maintenance of task-set and preparatory biasing– Reactive control: Reactivation of task-set following target

• This theory provides resolution of key questions– Residual switch costs due to probabilistic preparation– Performance on non-prepared trials aided by reactive control– Task-set interference only present on unprepared trials– PFC activity associated with task-cues may be variable

• Framework may account for other issues as well – Mixing costs: Anterior vs. dorsolateral PFC; proactive biasing of eligible tasks– Sequential effects: Conflict (Incongruency)-based strengthening of active task (e.g. Goschke,

2000)– Asymmetric switch costs: no repetition benefits for weaker task w/o proactive control– Other domains: WM, selective attention, individual differences

20

Future Directions

• Asymmetric Switch Costs (van der Stigchel)

• Mixing Costs (Neuron)

• Conflict Effects (Brown)