Agentic World Analysis (AWA) - an alternative way to explore systems and support decision making
This paper introduces Agentic World Analysis (AWA), a novel framework that combines agentic AI simulation with expert elicitation to navigate complex environmental uncertainties, demonstrated through the WEGA system's successful application in modeling future pathways for the Nitrogen Crisis in the Netherlands.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
For centuries, humans have looked toward the future with a mixture of hope and anxiety, trying to foresee how complex systems like our climate, economies, and societies might change. To make sense of this uncertainty, scientists have long relied on two main tools: computer simulations that follow strict mathematical rules, and panels of human experts who use their experience to guess what might happen. Both methods have their strengths, but they also have significant weaknesses. Computer models can be rigid and difficult to update when new information arrives, while human experts can be biased, inconsistent, or slow to react to rapidly changing events. As the world faces increasingly urgent sustainability challenges, from pollution to resource scarcity, the need for a more flexible and transparent way to explore possible futures has become critical.
A new approach called Agentic World Analysis attempts to bridge this gap by combining the best of both worlds. Instead of relying solely on static equations or a single group of human consultants, this method uses artificial intelligence agents that act like a team of specialized experts. These digital agents do not just crunch numbers; they read documents, gather evidence, debate different viewpoints, and construct detailed stories about how the world might evolve. The goal is not to predict the future with perfect accuracy, which is impossible, but to create a clear, auditable map of different possibilities, showing how specific decisions could lead to different outcomes. This allows decision-makers to see not just what might happen, but why it might happen, and to test their strategies against a wide range of potential scenarios.
In a recent study, researchers Yongchao Zeng and his colleagues introduced this new framework and tested it with a real-world crisis: the nitrogen pollution problem in the Netherlands. Nitrogen emissions from agriculture have exceeded environmental limits for years, threatening biodiversity and causing political conflict between farmers, environmental groups, and government officials. The researchers built a system called the World Engine by Generative Agents, or WEGA, to simulate how this crisis might unfold over the next fifteen years. They did not program the system with a fixed set of rules about how farmers or politicians would behave. Instead, they provided the AI with a starting point—the current state of the crisis—and a list of key players, including a livestock farmer union representative, a European Commission official, an environmental scientist, a right-wing media editor, and the Dutch Minister of Agriculture.
The system then began to run in cycles, mimicking the passage of time in five-year steps. In each cycle, the stakeholder agents would discuss the situation and propose policies, such as new regulations or financial support for farmers. Once a decision was made, the expert agents would analyze the likely consequences of that decision over the next five years. They would look for evidence, identify uncertainties, and generate different possible futures based on how those uncertainties might be resolved. For example, if a new policy was introduced, the system might ask: "What if the public accepts this policy?" versus "What if the public rejects it?" This process created a branching tree of scenarios, allowing the researchers to explore many different paths forward without having to build a new computer model from scratch for each one.
The results of the simulation revealed two distinct pathways for the nitrogen crisis. In the first path, the system assumed that social acceptance of new policies remained low. This led to a scenario of conflict, where monitoring equipment was sabotaged and reporting was falsified, causing the situation to deteriorate. However, even in this difficult scenario, the system found a way to stabilize the situation. A new agreement was proposed that offered financial support to farmers if they could restore the integrity of the monitoring system. When the system simulated the successful restoration of this monitoring, the conflict began to ease, and the focus shifted from political fighting to technical problem-solving.
In a second, alternative path, the researchers explored what would happen if the attempt to restore the monitoring system failed. In this scenario, the failure led to a collapse of trust, with the European Union imposing heavy fines and cutting off agricultural payments. This external pressure radicalized the farming community further, creating a deadlock that was much harder to break. By comparing these two paths, the system highlighted a crucial insight: the success of any policy depends heavily on the ability to maintain trust and data integrity. If the data is seen as trustworthy, cooperation is possible; if it is not, even the best-designed policies may fail.
The researchers also built a special tool to evaluate the quality of the stories the AI generated. This tool checked the logic of the scenarios, the strength of the evidence used, and the consistency of the characters' behavior. It found that the system produced coherent and well-supported narratives, but it also pointed out limitations. For instance, the simulation showed that the system sometimes assumed a level of stability that might not be realistic, and it noted that the specific policies generated were illustrative examples rather than perfect blueprints for action. The evaluation emphasized that the value of the system lies in identifying the mechanisms that drive success or failure, rather than in predicting the exact outcome.
This study demonstrates that artificial intelligence can be used to create a flexible, transparent, and collaborative way to explore complex problems. Unlike traditional models that are often locked in rigid structures, this approach allows for the integration of diverse types of knowledge, from scientific data to political narratives. The researchers argue that this method does not replace human experts but rather augments their ability to think through complex situations. By making the reasoning process visible and open to scrutiny, the system helps decision-makers understand the trade-offs involved in their choices. While the technology is still in its early stages, it offers a promising new way to navigate the uncertainties of the future, turning the crystal ball of the fortune teller into a clear, auditable map of possibilities.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.