PosterPDF Available

Trading off the costs of conflict and expected rewards

Authors:

Abstract

Adaptive behaviour requires the continued monitoring of action and its outcomes, to detect conflicts and correct behaviour accordingly. Conflict is considered aversive, and is typically avoided: people may choose easier tasks, or be biased by external suggestions. Conflict has also been shown to reduce the perceived control over action outcomes, and alter the perceived valence of subsequent events. Yet, it remains unclear whether external information could influence goal-directed decision-making, and whether response conflict could impact instrumental learning. The present study investigated these questions by embedding irrelevant flankers within a reversal learning task, with intermixed free and instructed trials. Results showed that participants learned to adapt their choices to maximize rewards. Nevertheless, participants were still biased by flanker stimuli, and were more likely to choose to follow, than to go against, the flankers’ suggestions. The perceived cost of being in conflict with an irrelevant suggestion can sometimes trump the evaluation of internal value representations. Adapting computational models of reinforcement learning allowed us to assess the influence of response conflict on the decision and on learning. Modelling results showed that the cost of conflict was traded-off against expected rewards, such that conflict was avoided when evidence for the conflicting option was weak. Turning to the learning phase, we found that instructed choices were associated with lower learning rates than free choices. Learning rates were further reduced when there was conflict between instructions and subjective beliefs. However, there was no robust evidence that conflict between actions and external distractors influenced learning. Our results show that external information can interfere with value-based decision-making, but that only some types of conflict affect instrumental learning.
!"
Results'
#$%&'()$*+,'%-./0,"
Background'
120%/3$4536733#4%((3-38'#64'-63(#'6%83$4/'#4534
5%'$3&45943:63(#'-4%#;0(,'6%0#<=>."
?:63(#'-4$6%,)-%4/'#46(%++3(4(3$@0#$34/0#A%/64594
'/6%8'6%#+4/0,@36%#+4(3$@0#$3$."
10#A%/64'&'@6'6%0#4@(0/3$$3$4'-63(4$)5$3B)3#64
'/6%0#$C4(3&)/%#+4/0#A%/643D3/6$C40(4'80%&%#+4
/0#A%/6$4723#4@0$$%5-3<E>."
10#A%/64%$4/0#$%&3(3&4'83($%83C4'#&4/'#4'-63(46234
@(0/3$$%#+40;4'/6%0#40)6/0,3$C4-3'&%#+4604'4
(3&)/3&4$3#$340;4'+3#/9<=>C40(4,0(34#3+'6%834
38'-)'6%0#$40;4#3)6('-4$6%,)-%<F>."
Conclusions'
G0'-H&%(3/63&4/20%/3$4/'#45345%'$3&45943:63(#'-4%#;0(,'6%0#."
I3$@0#$34/0#A%/6463#&$4604534'80%&3&C4)#-3$$4%#63(#'-4
38%&3#/34%$4$6(0#+C4(383'-%#+462'646234/0$640;4/0#A%/64%$4
6('&3&H0D4'+'%#$643:@3/63&4(37'(&$."
I3/3#64/0#A%/643:@3(%3#/34(3&)/3$4$)5$3B)3#64/0#A%/64
3D3/6$C4'#&4%#/(3'$3$4/0#A%/64'80%&'#/3."
J3'(#%#+4%$4(3&)/3&4723#4;0--07%#+4%#$6()/6%0#$C462'#4723#4
;(33-94/200$%#+472'64604&0."
J3'(#%#+4,'940#-94534'D3/63&4594$0,3469@3$40;4/0#A%/6K"
-/0#A%/64536733#4%#$6()/6%0#$48$.453-%3;$4(3&)/3$4-3'(#%#+L"
-5)64#064/0#A%/64536733#4'/6%0#48$.43:63(#'-4&%$6('/60($."
!
!"!!!!=
1
1+ ! (! !!"#$ !!!!"#!!! ! !!(!))
!
Computational Model'
Models' AIC M4NO4' Model
Frequency'
Exceedance
Probability'
,=K4N6'#&'(&4IJ4<βC4α>" PEF.F4M4=Q=.R" S.=T" S.SE"
,EK4<βC ϕC4α>44" P=R.F4M4=UF.T" S.SU" S.SS"
,FK4<βC ϕC4αC4V4αI>" P=T.E4M4=UR.S" S.SR" S.SS"
,RK4<βC ϕC4αFree4V4αInstructed>" PSR.P4M4=PW.U" S.SU" S.SS"
,TK4<βC ϕC4αFree_C4V4αFree_I4V4αInstructed>" PSR.Q4M4=PU.R" S.SW" S.SS"
,WK4<βC ϕC4αFree4V4αInstructed_C4V4αInstructed_I>" PST.E4M=PP.R" S.ST" S.SS"
,PK4<β, ϕ, αFree_C4V4αFree_I4V4αInstructed_C4V4
αInstructed_I>4" PST.R4M4=PU.U" S.=S" S.SS"
m8:4<β, ϕ, αFree4V4αInstructed_Best4V4
αInstructed_Worst>4" PS=.U44M4=PP.U" S.RR" S.QP"
0.00
0.25
0.50
0.75
1.00
21 0 1 2
Q(Left) Q(Right)
P(Right)
Flanker Action Congruent Incongruent Phi = 0
Phi = 0.5 vs Phi = 0
O3/%$%0#$4'(345%'$3&4607'(&$4A'#X3(H/0#+()3#640@6%0#$."
10#A%/6463#&$4604534'80%&3&C4%;4'/6%0#48'-)34&%D3(3#/3$4'(34$,'--."
J3'(#%#+4('63$4'(34(3&)/3&459K"
Y'8%#+4604;0--074%#$6()/6%0#$L"
10#A%/64536733#4%#$6()/6%0#$4'#&4$)5Z3/6%83453-%3;$."
J3'(#%#+4%$4#064(05)$6-94'D3/63&4594/0#A%/64536733#4'/6%0#$4'#&4
3:63(#'-4&%$6('/60($."
"
Research Questions'
1'#4+0'-H&%(3/63&4/20%/3$4534%#A)3#/3&45943:63(#'-4&%$6('/60($["
O03$4/0#A%/64%#A)3#/34-3'(#%#+4'50)64'/6%0#4/0#$3B)3#/3$["
-10#A%/64536733#4'/6%0#4'#&43:63(#'-4&%$6('/60($["
-10#A%/64536733#4%#$6()/6%0#$4'#&4$)5Z3/6%83453-%3;$["
Task'
10,5%#3&4A'#X3(4'#&4(383($'-4-3'(#%#+46'$X$4\N = 20]."
^'(6%/%@'#6$4-3'(#4723623(4-3;6_(%+264'/6%0#4%$4,0$64(37'(&%#+."
`/6%0#H(37'(&4,'@@%#+4/2'#+3$4';63(4)#@(3&%/6'5-34&3-'9."
a(334'#&4%#$6()/63&46(%'-$4'(34('#&0,-94%#63(,%:3&C4'#&4
@'(6%/%@'#6$4'(3460-&4604-3'(#43B)'--94;(0,4506246(%'-469@3$."
!!+1= !!+ !(!! !!) !
Decision:)
Learning:)
*n.s.
*
n.s.
5
0
5
10
15
20
Congruent Incongruent
Previous Congruency
Flanker Bias on Choices (CI)
Previous Outcome Positive Negative
b0(3"
10#A%/6"
`80%&'#/3"
Conflict Avoidance)
10#A%/64536733#4'/6%0#4'#&4
3:63(#'-4&%$6('/60($4%,@'%($4
'/6%0#4$3-3/6%0#."
a0--07%#+4/0#A%/6C4\'c3#6%0#'-]4
'&Z)$6,3#6$4'--074'4(3&)/6%0#4
%#4$)5$3B)3#64/0#A%/643D3/6$."
a0--07%#+4/0#A%/6C40(4-0$$3$C4
/0#A%/64'80%&'#/34%#/(3'$3$4
\%.3.42%+23(4A'#X3(45%'$]."
^'(6%/%@'#6$4-3'(#3&4604,'X34
623453$64/20%/3C45)6473(345%'$3&4
5946234A'#X3($d4$)++3$6%0#."
<=>4N%&'()$4e4Y'++'(&4\ES=W].4O%f/)-64'/6%0#4&3/%$%0#$4(3&)/346234$3#$340;4'+3#/9K4`4$6)&94)$%#+46234?(%X$3#4A'#X3(46'$X.4Acta PsychologicaC4166C4=g==."
<E>4O%+#'62C4h%3$3-4e4?&3(4\ES=T].4a-3:%5-34/0#A%/64,'#'+3,3#6K410#A%/64'80%&'#/34'#&4/0#A%/64'&Z)$6,3#64%#4(3'/6%834/0+#%6%834/0#6(0-.4JEP: LMCC441\R]C4QPTgQUU."
<F>4a(%i4e4O(3%$5'/24\ES=F].410#A%/6$4'$4'83($%834$%+#'-$K410#A%/64@(%,%#+4%#/(3'$3$4#3+'6%834Z)&+,3#6$4;0(4#3)6('-4$6%,)-%.4CABNC413\E]C4F==gF=P."
Learning)
20
30
40
54321 0 1 2 3 4 5
Free Trial Relative to Reversal
High Reward Choices (%)
Flankers Action Congruent Incongruent
_
Data Model
0
1
2
3
4
β
Arbitrary Units
0.00
0.25
0.50
0.75
ϕ
0.00
0.25
0.50
0.75
1.00
αFree αInstructed
Best
αInstructed
Worst
Decision) Learning)
b0(3"
10#A%/6"
Conflict Adjustments)
0
25
50
75
Free Instructed
Choice
RT Congruency Effect (IC)
Previous Congruency Congruent Incongruent
* *
n.s. *
300
400
500
600
700
Free Instructed
Choice
Reaction Times (ms)
Current Congruency Congruent Incongruent
Action Selection)
Target &
Flankers Action Outcome
400 ms 100 ms RT
300 ms
+ 1
700 ms
(R)
(R)
(L)
(L)
R
R
L
L
Free
Instructed
(C)
(I)
(C)
(I)
C
I
C
I
+1
+1
+1
+1
-1
-1
-1
-1
Choice
Flanker-Action
Congruency
-1
-1
-1
-1
+1
+1
+1
+1
75% 25%
ResearchGate has not been able to resolve any citations for this publication.
ResearchGate has not been able to resolve any references for this publication.