Agent-based models under uncertainty.

Area of Science:

• Computational Science
• Complex Systems Modeling

Background:

• Agent-based models (ABMs) often use Monte Carlo (MC) simulations to assess uncertainty.
• MC is appropriate for aleatory uncertainty (variation) but not epistemic uncertainty (lack of knowledge).
• This study models epistemic uncertainty in an agent-based battleship simulation.

Purpose of the Study:

• To contrast Monte Carlo (MC) and interval implementations for epistemic uncertainty in agent-based models (ABMs).
• To evaluate the suitability of different uncertainty quantification methods in complex simulations.
• To analyze the impact of imperfect information (e.g., radar) on simulation outcomes.

Main Methods:

• Developed a battleship simulation with agents representing ships.
• Implemented both Monte Carlo (MC) and interval-based approaches to model epistemic uncertainty.
• Introduced an imperfect radar system to simulate a lack of knowledge about agent status.

Main Results:

• Interval implementation provides broad system bounds but lacks quantitative insights into expected outcomes.
• MC simulations tend to conclude with fewer remaining agents compared to interval methods.
• Identities of surviving agents in interval methods showed partial overlap with MC results, but with fewer total identities.

Conclusions:

• Interval methods can be implemented in ABMs, yielding results useful for defining system bounds.
• Interval approaches do not provide clear insights into expected outcomes or trends in uncertain environments.
• The choice of uncertainty quantification method significantly impacts simulation interpretation and decision-making.