simulating geopolitical decision making processes with leadersim barry silverman mike johns gnana...
TRANSCRIPT
Simulating Geopolitical Decision Making Processes With
LeaderSIMBarry Silverman
Mike JohnsGnana Barthy
Evan SandhausJanuary 2006
PMFServ ArchitectureStimuli
Biology/Stress
Personality, Culture, Emotion
PMFserv
Cognitive
Response
Social Module
Expression
Simulated World
Physiolo-gy Tanks
•sleep•nutrition
•injury
Janis-Mann
Gillis-Hursh
Inte-grated Stress
Energy Tank
Status
CopingStyle
NegativeEmotions
TimePressure
EF ES
TP
PhysiologyUpdates Coping Style
GibsonAffordance
LEGEND: Implements Interprets NewLiterature Literature PMF
Perception
ValueTrees(GSP, Bayes)
Subj.Utility
(Damasio)
Cog.App.(OCC)
SEU
GSPNodeFail/Succeed
Emotions(11 pairs)
Action Choices Afforded
Current World State:GSP Leaf NodeAffordance Updates
AugmentedDecisionTheory
IntentionManagement
NestedIntent-ionalityProc’g
Relationship Tanks•Alignment
•Credibility/Trust•Cognitive Unit
•Valence
iStress
Identity Properties
•Demography•SocialGroup
•Role
Observations of Other Agents
Updates
Relationship ParameterLevels
Action Choice (physical, speech)
Emotions(11 pairs)
SEU
Candidate Action
Relationship ParameterLevels
BR
EV
Memory
PMFServ ArchitectureStimuli
Biology Module/Stress
Personality,Culture,Emotion
Memory
Cognitive
Response
be free
help others
support terrorist
hide terrorist distract guards
crowd together block guards vision
be independent
sacrifice life
protect terrorist
survive
run for cover
protect children
TBR = E [ P U(st, at) ]
t=1+
-
Perception Module Expression
Social Module,Relations,Trust
Agents in Role Playing Game Simulations
Intentions/GSP Trees•Needs & Wants
•Culture & Personality•Relationships & Trust
Model of Others’Intentions/GSPs
Stressors/Personality& Coping Style
PMFserv•Perception•Subj.Utility•Hot Reaction
PMFserv•Obj.Utility•Cold Delibertn•Game Theory
Discourse•Bluffing•Deception•Threats/Pacts
GameWorld•Territory•Resources•Groups•Threats•Tributes•Bluffs•Battles•etc.
Intentions/GSP Trees•Needs & Wants
•Culture & Personality•Relationships & Trust
Model of Others’Intentions/GSPs
Stressors/Stimulants& Coping Style
PMFserv•Perception•Subj.Utility•Hot Reaction
PMFserv•Obj.Utility•Cold Delibtn•Game Theory
Discourse•Bluffing•Deception•Threats/Pacts
Agent Agent
Human
Actions
Utterances
Actions
Utterances
ActionsUtterances
LeaderSim Architecture
GameServer
Game World State
XMLRPC
Athena’sPrism Client
Human Users
Athena’sPrism Client
Athena’sPrism Client
LeaderSim AI
LeaderSim AI
PMFServ
PMFServ
LeaderSim Game Elements
Leaders
Territories Resources
Acknowledge Legitimacy
Open Embassy
Acquire Black Market WMD
Activate Biological WMD
Activate Chemical WMD
Activate Nuclear WMD
Air Strike
Appease Fanatics
Bust
Corporate Lobbying
Covert Weapons Search
Create Aid Project
Create Biological WMD
Create Chemical WMD
Create Nuclear WMD
Debt Forgiveness
Destroy WMD
…
Actions
Con
trol
Have
Aff
ect
Do
Action Selection Algorithm
Preferences Tree
Standards Tree
Goals Tree
RankedAttacking/Target
Resource Pairs
Offensive / Defensive
Actions
RankedActions
PaymentStrategy
Power Vulnerability Scale
AlignmentWMD
ReservoirDifferences
PMFServ
Action
Attacking/TargetResource Pairs
Power And Vulnerability
P a M ax Enem y sourcesD estoryed a( ) ( R e ( ))
V a M ax Own sourcesD estoryed a( ) ( R e ( ))
• Power– Measurement of an agent’s ability to act upon
the resources of others
• Vulnerability– Measurement of the ability of other agents to
act upon one’s own resources
Goals, Standards And Preferences
• Goals– Best envisioned as steps in a plan to accomplish a
task– Typically start with “I will…”
• Standards– Typically socially-imposed guidelines for behavior– Typically start with “People should…” or “People
should not…”
• Preferences– Ideals about the desired state of the world– Typically start with “I like…” or “I dislike…”
Goal Tree
Goals
Grow Protect
Eco
no
my
Blac
k Market
Fo
reign
Aid
Military
Dip
lom
acy
WM
D P
rog
ram
s
Med
ia
Zealo
ts
Peo
ple
Au
tho
rity
Eco
no
my
Blac
k Market
Fo
reign
Aid
Military
Dip
lom
acy
WM
D P
rog
ram
s
Med
ia
Zealo
ts
Peo
ple
Au
tho
rity
Standards Tree
MilitaryDoctrine
Treatment of Out Groups
SensitivityTo
Life
Scope OfDoing Good
Task Relationship
Balance
Exercise of
Power
Standards
Neu
tral is O
ut G
rou
p
No
t Sen
sitive
Sen
sitive
Lo
ok A
fter N
arrow
er In
teres
ts
Brin
g A
bo
ut G
reater
Go
od
Be T
ask F
oc
used
Reso
lve
Issues b
y N
ego
tiation
Pro
vide H
elp
Be
Rela
tion
ship
F
ocu
sed
Be O
pen
Be C
on
trollin
g
Use A
sym
metric
Attacks
Use C
on
ven
tion
al A
ttacks
On
ly Co
nven
tion
al
Also
Co
nv
entio
nal
Ou
t Gro
up
s are L
egitim
ate T
argets
Trea
t with
Fairn
ess
and
Justice
Frien
d is O
ut G
rou
p
En
emy is O
ut G
rou
p
Help
N
eutrals
Help
E
nem
ies to
o
Help
F
riend
s
Preferences Tree+ Preferences
- Friendly Territory
- Neutral Territory
- Contested Territory
+ Enemy Territory
- Home Territory
- Own Resources
- Ally’s Resources
- Neutral’s Resources
+ Enemy’s Resources
Grow
Maintain
Contain
Reduce
- Authority
- People
- Zealots
- Media
- WMD Programs
- Diplomacy
- Foreign Aid
- Black Market
+ Economy
AI PerceptionsLeader 1
Leader 2
Leader 3
Leader 4
Leader 5
Territory 1
Territory 2
Territory 3
Territory 4
Territory 5
Ac
tion
1
Ac
tion
2
Ac
tion
3
Ac
tion
4
Self
Ally
Neutral
Enemy
Home
Ally
Neutral
Contested
Enemy
LeaderSim AI
WM
D
Ge
no
cide
Milita
ry Stick
Lig
ht M
il Stick
Asym
me
tric S
tick
Lig
ht
Asym
me
tric S
tick
Po
litical
Sticks
Ca
rrot
Se
lf O
rga
niza
tion
Scope of LeaderSim
ScaleUpTerritories (10)Resources (10)Actions (70)Total (10x10x70xpayment levels)
x no. of pliesx N leaders (10)
PrototypeTerritories (3)Resources (3)Actions (5)Total (3x3x5xpayment levels)
x no. of pliesx N leaders (3)
References www.seas.upenn.edu/~barryg/HBMR.html
•.Silverman, BG, Rees, R., Toth, J, et al., (2005, Jan).“Athena’s Prism – A Diplomatic Strategy Role Playing Game for Generating Ideas and Exploring Alternatives”, 1st Internat’l Conf on Intel Anal
•Silverman, B.G., Johns, M., Bharathy, G. (2004, August). “Agent-Based Simulation of Leaders.” ACASA/UPenn, Tech Report.
•Silverman, B.G., Johns, M., et al. (2002, May). “Constructing Virtual Asymmetric Opponents from Data and Models in the Literature." 11th BRIMS, SISO.
Prototype LeaderSim Results
0
10
20
30
40
50
60
1 21 41 61 81 101 121 141 161
Turn Blue Yellow Red
Nash Equilib: 2 winners in conflictual world
0
10
20
30
40
50
60
1 21 41 61 81 101 121
Turn
Res
ourc
es
Blue Yellow Red
Rare 3 in endgame. Yellow specialized away from Red and Blue.
0
10
20
30
40
50
60
1 21 41 61 81 101 121 141 161
Turn
Re
so
urc
es
Blue Yellow Red
Using threats, Yellow turns Red and Blue against one another
0
10
20
30
40
50
60
1 21 41 61 81 101 121 141 161
Turn
Re
so
urc
es
Blue Yellow Red
Y’s power is curtailed early. Y then uses treaties to negotiate peace.