Transcript Socio-environmental Agents

Socio-environmental Agents
Thomas E Downing
SEI Oxford
•
•
•
•
Choosing methodology
Attributes of multi-agent modelling
A water example
Challenges
Modelling approaches
•
Frameworks and observation:
– Descriptive: Good as sources & validation, but difficult to generalise from
– Sociological Theory: rich, difficult to unambiguously relate to any specific case
– Statistical and experimental: Valid but impossible to extend to future
•
Disciplinary(?):
–
–
–
–
–
•
Micro-economic: Puts techniques above problem
Game theory: Only solvable with a small number of discrete choices
Population dynamics: Does not (really) relate to micro behaviour
Physics-derived models: Can be useful for post hoc encapsulation
Descriptive computational simulation: difficult to get enough observations
Multi-agent
–
–
–
–
Robotic experiments: costly and unreliable, experiments take a lot of time and effort
Artificial life computational models: Good on process, can be disconnected
Artificial Intelligence/Machine Learning: Useful techniques but strongly a priori
Agent-based social simulation: emerging integration?
From Bruce Edmunds: www.cpm.mmu.ac.uk
Integrating nature & society
Qualitative
Quantitative
Choosing meta-methodology
Individuals
Groups
Representing society
Societies
Attributes of multi-agent systems
Software agents…
• Correspond to real-world actors
– sample diversity of populations
• Embed behaviour
– beliefs, norms, goals, plans
• Interact
Environment
– environment
– each other
Perception
Action
Internal process
Demand for water in southern England
WATER
EA
WC
EA1
NEG1
NEG2
NEG3
WC1
WC2
WC3
Water demand ABSS
Historical climate
MH climate
Individual
(30% N)
Group
(55% N)
CCDeW Project: Edmunds, Moss et al.
Challenges
•
•
•
•
•
Validation: what is modelling for anyway?
Scale: what grain of analysis is best?
Complexity: simple may not be better?
Computational speed: slow!
Stakeholder interface: distributed games?