Artificial Intelligence Models for Inclusive Participation in Policy Decision Making

Imagine future generations watching us right now and seeing what we are doing to our planet, people, and nature. Imagine that they could have a say in the decisions we’re making today (or lack of them), and on how they are affecting their future. What would they tell us? We will never know for sure. But it may already be possible to understand the potential worries and needs of future generations, and use that knowledge to represent them in current decision making. In “Could AI Speak on Behalf of Future Humans?”, Angela Aristidou and Konstantin Scheuermann argue that “Artificial Intelligence (AI) systems can give voice to previously unheard stakeholders and make collective decision-making processes more inclusive —but only if they are designed thoughtfully and deployed responsibly”. 

The idea of using AI systems to represent future peoples can be extended to many other unheard groups and perspectives of society. The use of AI to assist in filling data gaps or blind spots in decision making is rapidly proliferating in several areas of society and governance. 

Why does it matter for sustainable economic development? 

AI models and other digital technologies can be used as a tools for inclusive and sustainable economic development worldwide. However, in employing these technologies we must be aware that they present as many opportunities as risks, both at the local and global levels.

The Potential – Using AI Models for Policy Design, Impact Evaluation, and Policy Making

We are off track to meeting the Sustainable Development Goals (SDGs) by 2030 (2024 Sustainable Development Goals Report) – on some targets we are stagnant, whereas in others we are regressing. Only 17% of the targets are currently on track and we urgently need to address progress. Renown scholars and policy makers worldwide have come to an important conclusion that urges change in how we see the SDGs, and a new roadmap of the SDGs to 2050 (Nerini et al, 2024): we have been considering the SDG targets in isolation from each other even though there are strong and ever-evolving interlinkages between them, often in the form of trade-offs. Sustainability requires “focus on overall system outcomes rather than simplistic targets” (Fisher et al, 2024). This means that we need new ways to evaluate the impact of policy making using a complex systems approach. 

Using complex systems tools and modelling, based on mathematical and computational methods, allows us to:

• understand complex interrelationships, interdependencies, synergies and trade-offs;
• identify critical real-world challenges and prioritize solutions;
• simulate policies and action, as well as their juxtaposition with each other across different socio-economic-environmental contexts;
• estimate future impacts of alternative policies (or lack of them);
• explore tipping points; and support effective design and implementation.

For instance, we can use virtual representations (Digital Twins) of real world challenges to understand their complexity, how they relate to each other, their synergies and tradeoffs, and perform simulation models/exercises to estimate past or future policy impacts. Also, Agent Based Modelling (ABM) can project complex situations to support the achievement of the SDGs, for instance, modelling the impact of consumption/energy use to greenhouse gas emission and global warming. This can help explore social tipping points to counter-balance environmental tipping points. Overall, complex systems tools and modelling have great potential to support the design and implementation of policies and actions aimed at meeting the SDGs across diverse contexts.

The Risks – Data Validity in AI Models

The operational validity of results derived from complex systems tools and modelling are at stake due to issues of inclusion and representation of the data and algorithms. These are especially pronounced under the current digital transition, where data divides – wide differences between and within countries in their capacity to harness data – lead to “data invisibility” of marginalized communities, including women, cultural groups, socioeconomic, religious, and linguistic minorities, and migrant workers (UNCTAD, 2022).

For data-informed policy decisions to be effective, the data needs to reflect collective perspectives and be representative of diverse populations. This means that subjective social knowledge should be collected and balanced with external technical or scientific inputs. For that, we need to ensure inclusiveness in participatory modelling (Ferrand and al, 2024), which are often qualitative yet formalized and repeatable.

Also, we need to ensure responsible modelling and simulation, meaning that they are used to support decision making in ways that are evidence-based, transparent, and accountable. This requires analysts to consider the ethical, legal, and social implications of the use of data and algorithms in sustainability contexts. How can we ensure that the models and the subsequent decision-making processes that they inform are inclusive and responsible?

An Innovative Solution – AI Models for Inclusive Participation

AI simulations are innovative ways of ensuring the inclusion of diverse groups’ perspectives in complex modelling, by creating virtual representations of groups who are excluded from society. This means that AI modelling can be applied in two interdependent tasks: using AI to mitigate bias resulting from lack of data, and thus lack of representation, in combination with using AI to model real world challenges in all their complexity and interdependencies, trade-offs, and synergies, and simulating solutions. This is particularly relevant for inclusive policy design.

AI simulations can also be useful in decision making. A strong example is the use of AI system proxies for representing future generations, for instance, in climate related decision making. But any kind of marginalised group could be represented by AI, on any kind of matter of relevance for the SDGs by using AI-generated outputs to “speak” for under-represented groups. This can allow for inclusion opportunities for under-represented groups, such as including an AI proxy of their perspectives in decision-making scenarios. Or using these perspectives as a prompt in discussions, allowing stakeholders to explore and debate the potential consequences of inclusion and exclusion. 

However, this solution is only truly inclusive as long as we manage to ensure (i) stakeholder and domain expert involvement, (ii) transparency in AI deployment, and (iii) AI and data literacy in the process (Scheuermann & Aristidou, 2024). In this way, AI-powered human participation can legitimize the virtual representations mentioned above but also have significant and positive impacts in each of the SDGs. It can be used as a tool for democracy, empower minority workers to improve work conditions, reduce inequalities (income, gender, health, education, etc.), harness climate action, promote decentralized partnerships, among many other new ways of advancing the SDGs.

The Way Forward

Along with development and implementation of AI models for advancing the SDGs, and even ahead of it, we need to actively discuss and seek solutions of the risks associated with using AI modelling for policy making. Nurturing dialogue on the dissemination of AI solutions for inclusive participation, including all kinds of stakeholders, is the way forward.

This brought us together to the UNU’s Bonn AI and Climate Expert Meeting on the 2nd July 2024. Organized by Serge Stinckwich, our work group 1HideSerge Stinckwich (UNU-Macau), Liu Jia An (UNU-Macau), Angela Aristidou (Stanford University/UCL), Teresa Farinha (UNU-MERIT), Nils Ferrand (INRAE), Ronald Musizvingoza (UNU-Macau), Ally Nyamawe (UNU-Macau), Jonghwi Park (UNU-IAS), Jaimee Stuart (UNU-Macau), Min Yang (UNU-Macau) presented and discussed the use of participatory modelling approaches to establish trust with the public, engage citizens and stakeholders, and give them more opportunities for inclusive participation in Climate Action. We also invited experts 2HideGeorgina Curto (UNU-Macau), Gabriela Zuniga (UNSAAC), Caleb Gichuhi (UNSSC), Titiksha Vashist (The Pranava Institute), Girmaw Abebe Tadesse (Microsoft AI for Good), Ernest Mwebaze (Sunbird AI) in the topic to participate as panellists. The result was a fascinating sharing of concrete examples on how to use complex system modelling and AI to:

• understand the interplay between climate change, migration and basic social services, such as education and health care, among other pressing global issues;
• understand the role of synthetic data in building more representative synthetic users;
• develop model evaluation practices for enhancing transparency and quality in model creation;
• identify contextual factors and underlying assumptions in the modelling process;
• examining the chain of responsibility in the model's production workflows from the model commissioner to the creator, and the final user;
• analysing and deciphering minority positions;
• trigger alternative expressions in deliberative contexts;
• assess the construction and impact of argumentative contents representing unheard stakeholders. 

Our inspiring conversation in Bonn last summer developed further to become a series of blog articles, workshops, and scientific work to nurture dialogue and dissemination of AI solutions for inclusive participation, beyond Climate Action to all SDGs. 

In the upcoming series of blog articles, due to be published throughout 2025, we will be sharing the experiences and views of experts in this innovative field. We will also be hosting a workshop in late October, in the upcoming UNU-Macau AI conference 2025, to share and discuss lessons learned and future contributions. Looking forward to having you too participating in our talks! Please feel free to reach out for more information, or for sharing your comments, ideas, etc.

Keep tuned on AI for inclusive participation!