Decision Making 1

Different paths from stimulus to response

reflex

habit

rule-driven

pavlovian

deliberative

different types of input into a decision

perceptual

value based

perceptual input into a decision

which response best fits the sensory information

e.g., are there more dots moving to the left or right?

perceptual information is generated in

sensory-specific regions (e.g., MT), with greater activity for stronger info

this info feeds into a decision

value based input to a decision

which response best fits my preferences?

process of learning (acquiring) in value based decision making

anticipate value (can be automatic/implicit)

experience outcome

update future expectation based on how much better or worse outcome was than expected (prediction error)

FutureValuePrediction =

CurrentPrediction + PredictionError*

rewardExpected(t+1) =

rewardExpected(t) + (RewardRecieved(t) - RewardExpected(t))*

Midbrain (VTA) dopamine neurons signal

predicted reward and violations of predictions based on likelihood and delay

think monkey with juice

low likelihood of reward =>

large prediction error when reward occurs

high likelihood of reward

small prediction error when reward occurs

reward occurs around when expected =>

no prediction error

reward occurs long after expected =>

large prediction error

striatum receives dopamine inputs, and shows similar patterns of

expected reward to cues and prediction error to outcomes

expected rewards to cues

cues:

greater activity when reward expected to be more likely

prediction error to outcomes

outcomes:

greater activity for unexpected wins (positive prediction errors)

lower activity for unexpected no-wins (negative prediction errors)

we can learn to associate rewards with stimuli and with actions

pavlovian learning, and instrumental learning

pavlovian learning

learning to associate rewards with stimuli (e.g., tone)

instrumental learning

learning to associate actions that caused rewards (e.g., turn left)

basis for habits

learning of stimulus value have been associated with

ventral regions of striatum

learning of action values have been associated with

dorsal regions of striatum

ventral mPFC* and or/ v. striatum* activity increases with

greater subjective pleasantness of stimulus (e.g., odor, taste, attractiveness)

Assesing a value: ventral striatum (vStr)

includes nucleus accumbens (NAcc)

Assessing a value: ventral medial PFC (vmPFC)

includes medial OFC (mOFC)

Overlapping regions track

different kinds of rewards

primary (e.g., food), secondary (e.g., social, $$)

may form “common currency”

auction-like procedures encourage participants to indicate how much

they would actually be willing to pay for an item (WTP)

lateral orbitofrontal cortex (OFC) may hold

more specific information about stimulus features, vmPFC may integrate these

vStr/vmPFC signal value even when

unrelated to the task, and predict future choices

think: car priming

vmPFC/vStr activity can predict

how the success of a pitch for funding (e.g., crowdfund, microloan), better than ratings alone

The ___ of $100 depends on how much we have alreadt

utility

utility

how much pleasure or satisfaction we get from something

people weigh ___ 2x more than ___

anticipated losses

anticipated gains

some evidence for overlapping gain and loss regions

vmPFC, vStr