Coalition Formation Demo
Given agents with differing policy positions, this algorithm finds a majority coalition that agrees on an AI-generated compromise. Read the paper
How it works:
Each document is mapped to a 512-dimensional metric space using semantic embeddings.
A mediator identifies two coalitions based on their size-weighted distance from the system's centroid
and uses an LLM to generate candidate compromise sentences.
The proposal that minimizes the cosine dissimilarity from the groups' weighted mathematical average is selected.
Agents vote to join the new coalition if this proposal is closer to their ideal than the fixed status quo.
This iterative process repeats until a single coalition representing a majority converges on a final consensus.