[Paper Review] Models we Can Trust: Toward a Systematic Discipline of (Agent-Based) Model Interpretation and Validation

Gabriel Istrate|arXiv (Cornell University)|Feb 23, 2021

Opinion Dynamics and Social Influence4 citations

TL;DR

This paper advocates for a systematic discipline of model interpretation and validation in agent-based modeling (ABM) by integrating logic and formal methods. It proposes using logical frameworks to formalize stylized facts and social mechanisms, applying bisimulation for behavioral equivalence, and developing adversarial perturbation theories to test robustness—offering a path toward trustworthy, engineering-grade social simulations.

ABSTRACT

We advocate the development of a discipline of interacting with and extracting information from models, both mathematical (e.g. game-theoretic ones) and computational (e.g. agent-based models). We outline some directions for the development of a such a discipline: - the development of logical frameworks for the systematic formal specification of stylized facts and social mechanisms in (mathematical and computational) social science. Such frameworks would bring to attention new issues, such as phase transitions, i.e. dramatical changes in the validity of the stylized facts beyond some critical values in parameter space. We argue that such statements are useful for those logical frameworks describing properties of ABM. - the adaptation of tools from the theory of reactive systems (such as bisimulation) to obtain practically relevant notions of two systems "having the same behavior". - the systematic development of an adversarial theory of model perturbations, that investigates the robustness of conclusions derived from models of social behavior to variations in several features of the social dynamics. These may include: activation order, the underlying social network, individual agent behavior.

Motivation & Objective

• Address the lack of consensus and rigor in verifying and validating agent-based models (ABM) in social sciences.
• Overcome the ad-hoc nature of current ABM validation practices by establishing a formal, systematic discipline.
• Develop logical frameworks that formalize stylized facts and social mechanisms, including phase transitions in parameter space.
• Introduce adversarial perturbation theories to test robustness of model conclusions under variations in scheduling, networks, and initial conditions.
• Reconcile formal equivalence notions like bisimulation with the coarser, qualitative validation practices common in ABM.

Proposed method

• Formalize stylized facts and social mechanisms using logical frameworks that capture dynamic behavior and phase transitions in parameter space.
• Adapt bisimulation from reactive systems theory to define behavioral equivalence between ABMs, focusing on macro-level patterns rather than individual agent actions.
• Apply backward chaining in logical reasoning to identify necessary conditions (e.g., scheduler fairness) for the validity of stylized facts.
• Develop adversarial perturbation models to test robustness against changes in scheduling order, social network structure, and initial conditions.
• Use modal logic to express and verify stylized facts, ensuring that bisimilar systems satisfy the same formalizable properties.
• Integrate formal methods as middleware to bridge sociological theory and ABM simulations, enhancing credibility and robustness of conclusions.

Experimental results

Research questions

• RQ1 How can logical frameworks be systematically used to formalize stylized facts and social mechanisms in ABM, including phase transitions?
• RQ2 To what extent can bisimulation or its variants serve as a formal criterion for behavioral equivalence in ABM, given the coarser validation practices in social simulation?
• RQ3 What adversarial perturbation strategies (e.g., in scheduling, network structure, initial conditions) can robustly test the validity of ABM conclusions?
• RQ4 How can formal logic identify necessary conditions (e.g., fairness in scheduling) for the emergence of baseline stylized facts in ABM?
• RQ5 In what ways can formal methods serve as middleware to enhance the credibility and robustness of ABM results in policy-relevant contexts?

Key findings

• Logical frameworks can formally identify necessary conditions—such as scheduler fairness—for the validity of stylized facts, using backward chaining to trace dependencies.
• Bisimulation provides a mathematically rigorous notion of behavioral equivalence but is too microscopic for typical ABM validation, which focuses on macro-level patterns.
• Adversarial perturbation of scheduling, social networks, and initial conditions reveals robustness or fragility of model outcomes, offering a systematic validation strategy.
• Formal logic can derive that the statement (∀i)◇[State(i)=A] (eventual global adoption of state A) is not derivable from action axioms alone without the assumption (∀i)◇Scheduled(i), highlighting the need for fairness.
• Phase transitions in parameter space can be formally captured in logical frameworks, revealing critical thresholds where stylized facts lose validity.
• The integration of logic and formal methods into ABM can transform modeling from an art into a systematic, engineering-grade discipline with verifiable and trustworthy results.