TLDR: A new multi-agent deep reinforcement learning model helps improve participatory budgeting by addressing “choice overload” and promoting fair compromises. By simulating voter behavior and optimizing voting strategies, the AI demonstrates that prioritizing lower-cost projects leads to higher voter satisfaction and more equitable distribution of public funds, offering a valuable decision support tool for both citizens and policymakers.
Participatory budgeting is a powerful way for citizens to directly influence how public money is spent. It allows communities to collectively decide on spending priorities, aiming to distribute public funds more fairly. However, this process can sometimes overwhelm voters with too many project options, a phenomenon known as “choice overload.”
To address this challenge, new research explores how artificial intelligence, specifically a multi-agent deep reinforcement learning approach, can support decision-making in participatory budgeting. This innovative method helps voters identify strategies that increase the impact of their votes and assists policymakers in designing elections that encourage fair compromises on projects.
A Novel AI Approach to Decision Support
The paper introduces a unique, ethically aligned framework for decision support. It models voters as individual “agents” that learn and adapt their voting strategies. A significant technical hurdle in applying multi-agent reinforcement learning to large-scale voting scenarios is scalability, due to the vast number of possible voting combinations. The researchers overcome this by using a novel branching neural network architecture, which allows the AI to process complex ballot formats in a decentralized way.
Fair compromises are achieved by optimizing voter actions to better represent their preferences in the final selection of projects. The AI learns what voters are willing to compromise on, such as project cost, rather than just how much they are willing to compromise.
Real-World Experiments and Key Findings
The research tested its model using real-world participatory budgeting data from two elections: the “Stadtidee” election in Aarau, Switzerland (2023), and the “Budget Participatif” in Toulouse, France (2019). The multi-agent reinforcement learning models were then compared to the actual election outcomes in terms of voting behavior and the fairness of the collective choices.
The experiments revealed several key insights. Firstly, the AI-supported voting agents achieved higher vote satisfaction compared to voters in the actual elections. This means a greater proportion of the projects they favored ended up in the winning set. Secondly, the collective choices produced by the AI model were demonstrably fairer. Measures of inequality (Gini coefficient) were lower, and in some cases, every voter had at least some of their preferred projects win, indicating broader inclusion.
A crucial finding was the pattern observed in achieving fair compromise: it is often achievable through projects with smaller costs. The AI agents learned to attribute a higher proportion of their “tokens” (votes) to lower-cost projects compared to actual voters. This suggests that a shift towards supporting more affordable initiatives can lead to more equitable and satisfying outcomes for the entire community.
Also Read:
- Smart Energy Trading: How AI Helps Homes Share Power More Efficiently
- LLMs Learn to Think Smarter with Hierarchical Budget Policy Optimization
Implications for the Future of Public Spending
This work offers significant benefits for various stakeholders. For researchers and election designers, it provides a framework to understand how voter motivations and decision-making criteria relate to compromises and the quality of collective choices. For policymakers, the modeling can inform the design of future participatory budgeting elections, ensuring that lower-cost projects are adequately represented to foster greater fairness and public legitimacy.
Ultimately, for citizens, this type of AI decision support has the potential to recommend more effective voting strategies, helping them express their preferences for wider social issues in a way that maximizes the impact of their vote and ensures a larger share of the budget is allocated to projects they care about. You can read the full research paper for more details here.
