[Paper Review] Generative AI-assisted Participatory Modeling in Socio-Environmental Planning under Deep Uncertainty

Zhihao Pei, Nir Lipovetzky+3 more|arXiv (Cornell University)|Mar 17, 2026

Sustainability and Climate Change Governance0 citations

TL;DR

The paper proposes a templated, LLM-assisted workflow to bootstrap problem conceptualization in socio-environmental planning under deep uncertainty, demonstrated on lake and electricity market cases, to produce a multi-perspective model and Python implementation for subsequent DMDU analysis.

ABSTRACT

Socio-environmental planning under deep uncertainty requires researchers to identify and conceptualize problems before exploring policies and deploying plans. In practice and model-based planning approaches, this problem conceptualization process often relies on participatory modeling to translate stakeholders' natural-language descriptions into a quantitative model, making this process complex and time-consuming. To facilitate this process, we propose a templated workflow that uses large language models for an initial conceptualization process. During the workflow, researchers can use large language models to identify the essential model components from stakeholders' intuitive problem descriptions, explore their diverse perspectives approaching the problem, assemble these components into a unified model, and eventually implement the model in Python through iterative communication. These results will facilitate the subsequent socio-environmental planning under deep uncertainty steps. Using ChatGPT 5.2 Instant, we demonstrated this workflow on the lake problem and an electricity market problem, both of which demonstrate socio-environmental planning problems. In both cases, acceptable outputs were obtained after a few iterations with human verification and refinement. These experiments indicated that large language models can serve as an effective tool for facilitating participatory modeling in the problem conceptualization process in socio-environmental planning.

Motivation & Objective

• Introduce a templated workflow that uses large language models to aid initial problem conceptualization in DMDU-based socio-environmental planning.
• Enable identification of essential model components from stakeholders’ descriptions and assemble them into a unified model.
• Provide a Python implementation of the composed model suitable for integration with EMA Workbench or Rhodium.
• Demonstrate the workflow through two case studies and discuss strengths and limitations.
• Encourage participation and rapid prototyping by translating intuitive problem descriptions into structured model components.

Proposed method

• Propose an LLM-assisted workflow with four iterative steps guiding problem conceptualization from descriptive problem narratives to a formal multi-perspective model.
• Use chain-of-thought prompting and prompt chaining to decompose tasks and create a structured, memory-consistent outputs.
• In Step 1, formalize stakeholders’ problem descriptions into model components (state variables, decisions, transitions, objectives, uncertainties).
• In Step 2, infer multiple stakeholder perspectives and derive perspective-specific models.
• In Step 3, compose perspectives into a unified model with a shared environment and exogenous inputs for other perspectives.
• In Step 4, implement a modular Python program representing the unified model, enabling simulation and objective evaluation.
• Incorporate validation, verification, and refinement via a checklist inspired by the ODD protocol and multiple testing strategies (unit, property-based, scenario tests).

Experimental results

Research questions

• RQ1 How can LLMs facilitate the initial problem conceptualization for DMDU-based socio-environmental planning?
• RQ2 Can an LLM-assisted workflow produce a multi-perspective, unified model suitable for Python-based simulation and subsequent decision analysis?
• RQ3 What are the benefits and limitations of using LLMs for translating intuitive problem descriptions into formal model components?
• RQ4 How do the lake problem and electricity market problem illustrate the workflow's effectiveness and generalizability?

Key findings

• LLMs can identify essential model components and perspectives from stakeholder descriptions with only a few human validations.
• A unified, multi-perspective model can be produced that preserves common environment concepts while embedding perspective-specific decisions and objectives.
• The workflow yields a Python implementation that is compatible with EMA Workbench and can be adapted to other DMDU tooling with minimal changes.
• Experiments show acceptable outputs after a few iterations, supporting LLMs as effective aids for problem conceptualization in socio-environmental planning under deep uncertainty.
• The approach highlights strengths and limitations, including reliance on human validation and the need for careful prompt design.