Emergence in Artificial Societies Evolving Communication and Cooperation in a Sugarscape world by Pieter Buzing.

Emergence in Artificial Societies

Evolving Communication and Cooperation in a Sugarscape world

by Pieter Buzing

Plan

• What are ‘artificial societies’?

• Sugarscape

• Our goal: communication and cooperation

• Our model: VUScape

• Setup

• Results

• Conclusions

Artificial Society?

• Multi agent system– 2 levels: autonomous parts, behaviour of whole

– AS: more control over agents and world

• Artificial life– emergent behaviour

– AS: important role for individual

• Agent based simulation– AS: no “problem to solve”, like optimization

• Social modeling– interactions of agents; effects individual goals on population

Sugarscape

• Epstein & Axtell, 1996

• Torus shaped world: 50 x 50 cells

• Sugar resources [0 - 4]• Agents looking for

food• Evolution

Sugarscape

Agent “Internals”• Age [0 - 100]• Vision [1 - 6]• Sugar Amount [0 - inf]• Metabolism [1 - 4]• Gender [m/f]

Agent Actions• Die• Move• Harvest• Metabolise• Reproduce

Sugarscape

Reproduction rule

• Agent has gender: male or female

• Metabolism and vision are genetic!

• Parents: <m, v> X <M, V>

• Child: <m,V>, <M,v>, <m,v> or <M,V>

• Child inherits half of parents’ sugar

Sugarscape

• Agents will tend to move towards sugarhills

• Agents with high vision are better off

• Agents with low metabolism are better off

Our Goal

• Individual:– limited harvesting capabilities (maxSugarHarvest)

– ability to talk and listen

• Emergent behaviour:– cooperation

– communication

• “If cooperation is needed then talking is beneficial and communication will emerge”

VUScape

• Had to implement own testbed: VUScape

• Model is highly based on SugarScape

• The major changes:– Sugar randomly distributed; multi-agent cells– talkPref [0 - 1] and listenPref [0 - 1] genes– Talk actions and Listen actions– MaxSugarHarvest value: cooperation threshold

VUScape: random landscape

• Instead of 2 sugar hills a random distribution• 2,500 sugar units are spread across 2,500 cells• 30% population drop; but still viable world• (because it is harder to find food?)

VUScape: limited vision range

• Vision range set to 1 instead of gene range [1-6]• Evolution of vision is not the aim of our project• local info from vision, global from communication• Short-sighted agents face a tough environment

VUScape: multiple agents

• Cooperation scheme requires multi-agent cells• Higher population size is now possible

VUScape: re-seed sugar

• Agents find food, wait there until it regenerates• We need agents that are constantly searching• Explorativeness is increased by reseeding sugar

after consumption

VUScape: sex recovery period

• To avoid possible birth explosions we implement a sex recovery period

• Recovery period of 5 yields pop decrease of 11%• Flattens the age distribution

Step 1: in need of help

IF localAmount > maxSugarHarvest

THEN inNeedOfHelp

Step 2: talking

IF inNeedOfHelp AND rand < talkPref

THEN communicate to others on x and y axis:– cell coordinates and sugar value

Step 3: listening

IF rand < listenPref

THEN listen to others on x and y axis

Step 4: cooperating

• Use obtained information in movement decision.• Two agents can conquer any pile!• Cooperation is beneficial for both parties.• Communicative agents have an advantage?

Setup

• Stepwise increase cooperative pressure and monitor the communicative gene values.

• Experiment A: no communication– Talk and listen genes disabled

• Experiment B: with communication– Talk and listen genes initiated between 0 and 1

• If communication is beneficial then an increase of talk and listen values is expected.

Results: no communication

Results: with communication

Results: listenPref

Results: talkPref

Conclusion

• Communication makes society more viable

• High talking and listening preferences give agents a selective advantage

• Higher cooperative pressure induces communication

• Future work: other topologies, communication protocols