Decentralized Value Systems Agreements

A research team from Universitat Politècnica de València and King's College London has published a groundbreaking paper, 'Decentralized Value Systems Agreements,' introducing a new AI framework for multi-agent systems. The core challenge they address is the subjective nature of human values: individuals not only prioritize values differently but also interpret them uniquely. The team's novel method allows AI agents to indicate their personal value systems and their willingness to concede, then uses a decentralized optimization approach to find not one, but a set of possible societal agreements. This approach better accommodates realistic, heterogeneous societies where a single forced consensus is neither possible nor desirable.

In two real-world case studies—using data from a Participatory Value Evaluation process and the European Value Survey—the method demonstrated a substantial improvement in individual utilities compared to existing aggregation techniques. The results show that by allowing for multiple, distinct agreements, the system can find solutions that leave participants significantly better off than traditional methods that enforce a single compromise. This work, accepted at the prestigious AAMAS 2026 conference, represents a major step toward building AI systems and social simulations that can navigate complex, value-laden decisions in a way that respects human diversity.