Year: 2026
Type: article
Abstract: Many complex systems of scientific and societal importance, including epidemics, energy grids, transportation networks, and supply chains, are governed by nonlinear dynamics, are only partially observed through noisy and heterogeneous data, and are continuously reshaped by interventions. These properties make decision making fundamentally challenging, particularly when the mechanisms driving system behavior are only partially understood. While modern machine learning has achieved remarkable predictive performance, real‐world applications require AI systems that combine forecasting with uncertainty quantification, counterfactual reasoning, intervention analysis, and mechanism‐aware reasoning. This article synthesizes my line of work on learning, simulation, and reasoning in complex dynamical systems. Using epidemic prediction and decision making as a running example, I describe recent advances that illustrate the growing integration of data‐driven learning, scientific simulation, and mechanistic reasoning, and discuss how these complementary paradigms can support forecasting, intervention analysis, and high‐stakes decision making.
Source: AI Magazine
DOI: 10.1002/aaai.70125
Author: Alexander Rodríguez
Institution: University of Michigan
Language: English
Field: Decision Sciences
Domain: Social Sciences
Keywords: complex dynamical systems, epidemic forecasting, decision-making, uncertainty quantification, counterfactual reasoning, intervention analysis, mechanistic reasoning, scientific simulation, nonlinear dynamics, partially observed systems, forecasting, power grid, transportation networks, supply chain
Open Access status: hybrid
Estimated APC paid: $2,400
OpenAlex ID: W7236248078
