1. médialab Sciences Po
  2. Activities
  3. AI4AD: AI for Augmented Deliberation

AI4AD: AI for Augmented Deliberation

How can post-generative AI systems improve deliberation and collective decision-making?

Research

Research project

Participatory democracy is often seen as a possible remedy for combating citizens' growing mistrust of our representatives. However, the advent of democratic mechanisms presents citizens with problems such as unbalanced participation, the cost of organising large-scale consultations, etc. While the prospect of AI-enhanced deliberative and collective decision-making processes has not yet fully materialised, the emergence of large language models is seen as an opportunity to broaden participation, facilitate deliberation, suggest common ground, or infer unexpressed preferences. On the contrary, the use of AI technologies in such sensitive contexts also presents risks due to the biases inherent in their responses and their potential exploitation for decision manipulation, challenging the fundamental assumptions of traditional frameworks.

The aim of this project is therefore to explore how post-generative AI systems can improve deliberation and collective decision-making in order to promote fair and collectively acceptable policies. The aim is to design AI mechanisms and tools that can be used in deliberative democratic processes, and to experimentally evaluate and assess their real value at different scales and in different contexts (e.g. civic assemblies, online deliberative platforms, international or local negotiations, etc.). To achieve this goal, we envisage a combination of:

  • Descriptive approaches, based either on large-scale quantitative approaches or on localised studies. The aim is to analyse and visualise how people actually deliberate and issue early warnings about potential risks of manipulation or coercion (mapping controversies, coalition dynamics, diagnosing influence between individuals, etc.);
  • Descriptive-normative approaches based on multi-agent simulations, enabling complex systems based on empirically validated behaviours to be run in silico and counterfactual scenarios to be tested;
  • Normative approaches based on idealised models, which explore the properties of deliberation and collective decision-making mechanisms. More specifically, computational social choice, formal argumentation, multi-criteria decision support and preference learning offer a range of new techniques for designing mechanisms whose desirable properties are theoretically guaranteed. We intend to experiment with these AI-augmented deliberative frameworks in a variety of real-world situations.

Project funding

The project is funded by ANR (France 2003).

Partner institutions

LIPS6 (Sorbonne Université), Nukk.AI, CESE