An Adaptive Affective Social Decision Making Model Alexei Sharpanskykh Jan Treur vrije Universiteit amsterdam.

An Adaptive Affective Social Decision Making Model

Alexei SharpanskykhJan Treur

vrije Universiteit amsterdam

2

Motivation

Traditionally, human decision making has been modelled as the problem of rational choice from a number of options using economic utility-based theories

Research Aim: To create a more biologically plausible model of human decision making based on theoretical principles from Neuroscience and Social Science

3

Decision Making Aspects

Predicted effects of the options

Valuing of these effects

Emotions felt in relation to this valuing

Social influence

4

Predicted effects of the options

Simulated behavior and perception chains by Hesslow

s1 r1

s2

r3s3

r2

...

s1, s2, s3 ... are sensory statesr1, r2, r3 ... are preparation statesw1, w2, ... are link strengthsV1, V2, ... are state values

w1

w2

V1 V2

5

Predicted effects of the options

Simulated behavior and perception chains by Hesslow

s(evacuation_required) r(goto(staircaseA))

s(at(staircaseA)) r(goto(crossingA))

s(at(crossingA)) r(goto(exit))

s(at(exit))staircaseA

crossingA

exit

6

Emotion generation

“As if” body loop (by A.Damasio)

sensory state

preparation for the induced bodily response

sensory representation of the bodily response

induced feeling

s1 r1

s2

r(bem)s(bem)

feeling(bem)

7

Emotions and Valuing

Decision making involves emotional valuing of predicted consequences of decision options

A notion of value, involving emotions is represented in the amygdala

s1 r1

s2

r(bem)s(bem)

feeling(bem)

Damasio’s Somatic Marker HypothesisEach represented decision option induces (via an emotional response) a feeling which is used to mark the option

8

Social contagion

A’s emotion state for option O

A’s intention state

for option O

emotion states of

other group members

for option O

intention states of

other group members

for option O

A’s somatic marking

for option O

A’s mirroring of emotion

for option O

A’s mirroring of intention

for option O

9

Learning

Hebbian learning principle: connections between neurons that are activated simultaneously are strengthened

s1 r1

s2

r(bem)s(bem)

feeling(bem)

s(G(r1))

s1 r1

s2

r(bem)s(bem)

feeling(bem)

s(G(r1))

s1 r1

s2

r(bem)s(bem)

feeling(bem)

s(G(r1))

action-effect prediction links emotion-related valuation links social influence links

d (r1, s2)/dt = r1s2 (1 – (r1, s2)) – (r1, s2)

10

Learning

s1 r1

s2

r(bem)s(bem)

feeling(bem)

s(G(r1))

s1 r1

s2

r(bem)s(bem)

feeling(bem)

s(G(r1))

s1 r1

s2

r(bem)s(bem)

feeling(bem)

s(G(r1))

(I) action-effect prediction links (II) emotion-related valuation links (III) social influence links

Learning of links (II) has the greatest impact on decision making Learning of links (III) has a negligible effect on decision making, when agents are similar A combination of learning of all links (I), (II) and (III) results in the strongest discrimination

between the options

11

Combined model

s(evacuation_required)

r(move_to(E))

s(goal)

s(bfear)

s(is_at(E))

s(eval_for(is_at(E),bfear))

r(bfear)

s(eval_for(is_at(E),bhope))

hoper(bhope)

s(G(move_to(E)))

s(G(bfear))

s(G(bhope))

fear

s(bhope)

12

Simulation Results

0

50

100

150

1

2

3

0

0.2

0.4

0.6

0.8

1

0

50

100

150

1

2

3

0

0.2

0.4

0.6

0.8

1

0

50

100

150

12

34

56

78

910

0

0.2

0.4

0.6

0.8

1

The path is short, but becomes dangerous

The path is average, but dangerous

The path is long, but safe

timetimetime

# action # action# action

Preparation for the execution of the options

The developed model could be used to evaluate and predict emotional decisions of individuals in groups under stressful conditions

Learning of the emotion-related links has the strongest effect on discrimination of decision making options (cf. the role of the Amygdala in valuing)

In the future emotion regulation mechanisms (e.g., to cope with fear and stress) will be investigated

Conclusions