Generation and Simulation of Artificial Human Societies using Anthropologically modelled
Learning Agents
Autor: Radu Galan
Mentor: Dr. Czibula Gabriela
TABLE OF CONTENTS
ABSTRACT 01
02
04 05
MOTIVATIONS
THE DESIGN MODELLING & DATA
In tro du ct io n
06
EXPERIMENTS
07
VALIDATION
Me th odo lo gy
09
CONSIDERATIONS 08
RELATED WORK
FUTURE WORK
Co nc lusi on
03
INTRODUCTION
ABSTRACT
01
What?
The goal was building general anthropological social simulations using an
agent-based system . Essential components are:
- High quality modelling
- A scalable and flexible architecture - Means of visualizing and analysing
Abstract
Computational Model
Anthropological
model
Interaction
Agent-based system
Experiments in an Artificial Society
Social Simulation
MOTIVATION
02
Why?
Motivation
Sociological
Motivation
Ecological
Drennan - JASSS vol. 8
(…) artificial societies offer "insight into the relationship between micro-level
cognition and macro-level social behavior"
Joshua M. Epstein
Only by ‘growing’ a society in simulations we can declare that
it is thoroughly understood
RELATED WORK
03
RELATED WORK
Specific simulations
aircraft evacuation
fire evacuation
crowd control
traffic flow optimization
Electronic market analysis
RELATED WORK
General simulations
Conway & Schelling
Sugarscape
Complex social networks
RELATED WORK
Spatially-explicit simulations
Wild-life simulations of jaguars
Wild-life simulations of capuchins
RELATED WORK
Social simulations (anthropological)
Hominids: ABS (Australopithecus boisei )
Virtual Neanderthals
METHODOLOGY
THE DESIGN
04
How?
Who?
Where?
What?
When?
Agent-Based Social Simulation
Homo S Sapiens EUROPE
Hunter-Gatherer uncontested 500’000 – 0
100’000 – 0 100’000 – 11’000
40’000 – 11’000
(Late Pleistocene)
Flow
Experiment steps
Repeatability
Scalability
Compromises
The trade-offs
Causality vs Large scale
Generalization vs Performance
Low-level modelling vs Time
Architecture
UML
MODELLING & DATA
05
Modelling
Statistical
Iterative
Context
• Homo S Sapiens
• EUROPE
• Hunter-Gatherer uncontested
• 40’000 – 11’000 (Late Pleistocene)
Environment
• -20° → 60° longitude; 30° → 70° latitude
• Cylindrical map projection
Environment
SPATIALLY EXPLICIT TERRAIN
• resolution 80x160: 1951 km2 in one cell
• resolution 160x320: 487.8 km2 in one cell
• resolution 320x640: 121.9 km2 in one cell
• resolution 480x960: 54.2 km2 in one cell
• resolution 800x1600: 19.5 km2 in one cell
Environment
Environment
ELEVATION
Climate
Climate
TEMPERATURE 20 years modern dataset 8 days time step
DAY
NIGHT
Resources
Food source:
• Plant population:
EnvironmentAgent->ResourceAgent
3 types of plants:
• high cal. Fruits
• mid cal. Roots/Vegetables
• herbs and spices (health benefits)
Resources
Plant attributes:
• color {1,2,3}
• energy (0,700] – calories per 100g
• heal [0,3] – health points
• quantity [1,100] – multiple of 100g
• population [100,1000] * avg. area of cell
• probability for each terrain [0,100], …, [0,100]
• probability preferred temperature for ripping[-30,70]
• probability referred month for ripping [1,12]
• spread ability [0,100]
• Sensibility [0,100]
• current food
Plant dynamics:
• Population fluctuation
• Food production
Humans
Attributes:
• age (0,100)
• sex {0,1}
• health [1,100] – health points
• fitness [1,100]
• Hunger need [0,100]
• Thirst need [0,100]
• Reproductive need [0,100]
• Safety need [0,100]
• Emotional State [0,100]
• Plant find ability [0,100]
• Social interaction ability [0,100]
• Food Inventory
• Character
• Offspring
Dynamics:
• Aging
• Needs changing
Actions:
• Move
• Eat
• Scavenge for food
• Social request
• Social response (automated)
Memory:
• Human social status
• Places fondness
Humans
Actions:
• Move
• Scavenge for food
• Eat
• Social request
• Social response (automated)
Food inventory Health
Environment food
EXPERIMENTS
06
Architecture Testing
An environment
One environment agent in each square One attribute: temperature
One dynamic: convection
Environment
Surface night temperature Surface day temperature
Surface terrain
Surface elevation
Environment with resources
Environment with resources and non-intelligent humans
Random decision-making
Label Dataset
Action
Environment with resources and intelligent humans
Label Dataset
Circ.
Fitness label: health, emotional-state, hunger, thirst, reproductive
Circumstances:
Relation to another human attr. (4) Target environment attr.(8)
Resource attributes(Plants Color)
Personal State attr. (8)
Environment with resources and intelligent humans
Label Dataset
Action
Move Scavenge Eat Request
Environment with resources and intelligent humans
Eat
Environment with resources and intelligent humans
Eat Scavenge
Move Social request
Depth:
5-20
Environment with resources and intelligent humans
Global Attribute Histogram Imagine – Think – Decide
Color-Mapping of Attributes
(personal or by tribe average)
VALIDATION
07
Mortality
Metrics
AVG:
MAPE 40.25 min: 7.67 and max: 132.46 RMSPE 0.47 min: 0.095 and max: 1.92 SPRING AVG:
MAPE 37.50 min: 6.93 and max: 123.86
RMSPE 0.44 min: 0.082 and max: 1.91
Metrics
Mean absolute percentage error Root Mean Square Error
Root Mean Square Percentual Error
Spring
Complete
CONCLUSION
CONSIDERATIONS
08
Considerations
Personal contribution
Philosophical Limitations
• One of the few (if not only) Spatially Explicit Social Simulation of early Homo Sapiens
• Unique Data Gathering Results
• SOTA validation process
• Accurately modelled sapiens
• Social implication of implementation
• Great knowledge, Great responsibility
• Subjective interpretations
• Simulation of society from inside a society
• Computational power
• Current social research might have low accuracy
FUTURE WORK
09
Future work
Optimization
Extended modelling
Better visualization
More Experiments
More parallelization, Better memory allocation, Cloud computing
Animals, More actions, Settlement study, More specific plant species, Technologies, Cultural habits, Changing geography, Other hominins
3D animation, More statistical analysis, correlation analysis
Longer experiments with offspring enabled, validation on population growth
or range areas or migration trends
REFERENCES
[1] W. G. Kennedy, “Modelling human behaviour in agent-based models,” Agent-Based Model. Geogr. Syst., pp. 167–179, Jan. 2012, doi: 10.1007/978-90-481-8927-4_9.
[2] A. Drogoul and J. Ferber, “Multi-agent simulation as a tool for modeling societies: Application to social differentiation in ant colonies,” in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 1994, vol. 830 LNAI, pp. 3–23, doi: 10.1007/3-540-58266-5_1.
[3] M. Drennan, “The Human Science of Simulation: a Robust Hermeneutics for Artificial Societies,” Jan. 2005.
[4] M. Gardner, “MATHEMATICAL GAMES The fantastic combinations of John Conway’s new solitaire game ‘life,’” Sci. Am., vol. 223, pp. 120–123, 1970, Accessed: Oct. 22, 2021. [Online]. Available: http://ddi.cs.uni- potsdam.de/HyFISCH/Produzieren/lis_projekt/proj_gamelife/ConwayScientificAmerican.htm.
[5] S. Sharma, H. Singh, and A. Prakash, “Multi-agent modeling and simulation of human behavior in aircraft evacuations,” IEEE Trans. Aerosp. Electron. Syst., vol. 44, no. 4, pp. 1477–1488, 2008, doi: 10.1109/TAES.2008.4667723.
[6] L. Tan, M. Hu, and H. Lin, “Agent-based simulation of building evacuation: Combining human behavior with predictable spatial accessibility in a fire emergency,” Inf. Sci. (Ny)., vol. 295, pp. 53–66, Feb. 2015, doi:
10.1016/J.INS.2014.09.029.
[7] Y. Murakami, K. Minami, T. Kawasoe, and T. Ishida, “Multi-agent simulation for crisis management,” in Proceedings - IEEE Workshop on Knowledge Media Networking, KMN 2002, 2002, pp. 135–139, doi: 10.1109/KMN.2002.1115175.
[8] S. Sharma, K. Ogunlana, D. Scribner, and J. Grynovicki, “Modeling human behavior during emergency evacuation using intelligent agents: A multi-agent simulation approach,” Inf. Syst. Front. 2017 204, vol. 20, no. 4, pp. 741–757, Aug.
2017, doi: 10.1007/S10796-017-9791-X.
[9] P. Paruchuri, A. R. Pullalarevu, and K. Karlapalem, “Multi agent simulation of unorganized traffic,” p. 176, 2002, doi: 10.1145/544741.544786.
[10] S. R. Wolfe, P. A. Jarvis, F. Y. Enomoto, M. Sierhuis, and B.-J. van Putten, “A Multi-Agent Simulation of Collaborative Air Traffic Flow Management,” https://services.igi-global.com/resolvedoi/resolve.aspx?doi=10.4018/978-1-60566-226- 8.ch018, pp. 357–381, Jan. 1AD, doi: 10.4018/978-1-60566-226-8.CH018.
[11] “GENESIS PROPOSAL.” https://web.archive.org/web/20081016131324/http://www.genesis.ucl.ac.uk/proposal.html (accessed Oct. 22, 2021).
[12] “New and Emergent World Models Through Individual, Evolutionary, and Social Learning | NEW TIES Project | Fact Sheet | FP6 | CORDIS | European Commission.” https://cordis.europa.eu/project/id/003752 (accessed Oct. 22, 2021).