Decentralized Value Systems Agreements

Arturo Hernandez-Sanchez; Natalia Criado; Stella Heras; Miguel Rebollo; Jose Such

cs.MAMar 26, 2026

Decentralized Value Systems Agreements

Authors: Arturo Hernandez-Sanchez, Natalia Criado, Stella Heras, Miguel Rebollo, Jose Such

Organizations: VRAIN, Universitat Politècnica de València, València, Spain · INGENIO (CSIC-Universitat Politècnica de València), València, Spain

Abstract

One of the biggest challenges of value-based decision-making is dealing with the subjective nature of values. The relative importance of a value for a particular decision varies between individuals, and people may also have different interpretations of what aligning with a value means in a given situation. While members of a society are likely to share a set of principles or values, their value systems--that is, how they interpret these values and the relative importance they give to them--have been found to differ significantly. This work proposes a novel method for aggregating value systems, generating distinct value agreements that accommodate the inherent differences within these systems. Unlike existing work, which focuses on finding a single value agreement, the proposed approach may be more suitable for a realistic and heterogeneous society. In our solution, the agents indicate their value systems and the extent to which they are willing to concede. Then, a set of agreements is found, taking a decentralized optimization approach. Our work has been applied to identify value agreements in two real-world scenarios using data from a Participatory Value Evaluation process and a European Value Survey. These case studies illustrate the different aggregations that can be obtained with our method and compare them with those obtained using existing value system aggregation techniques. In both cases, the results showed a substantial improvement in individual utilities compared to existing alternatives.

Figures & tables

Panel of Figure 1 — Figure 1. Individual utility functions for value safety (S) and alternative Private Car (PC). Utility functions for four agents evaluating the safety of traveling by private car. Each curve corresponds to one agent and reaches its maximum at that agent’s original evaluation of the safety of travelling by a private car. Agents with higher weights assigned to safety have narrower parabolas, indicating a faster utility decrease when moving away from their preferred evaluation. Agents with lower safety weights have wider parabolas, showing greater tolerance to deviations. The figure illustrates how value weights affect sensitivity to changes in a specific decision-matrix entry.

Policy option	Description
$o_{1}$	The municipality takes the lead and unburdens
	you: The municipality will stay in charge and
	endorse what residents think is important.
$o_{2}$	Inhabitants do it themselves: Residents generate
	their own energy.
$o_{3}$	The market decides: The municipality waits and

Table 1. Policy options for Sùdwest-Fryslân PVE ( Delft University of Technology, 2020 ) .

Value	Description
$v_{1}$	Cost-effectiveness
$v_{2}$	Nature and landscape
$v_{3}$	Leadership
$v_{4}$	Cooperation
$v_{5}$	Self-determination

Table 2. Values inferred from participants’ motivations

Panel of Figure 2 — Figure 2. Utility distributions for the agreed - upon value systems in the PVE case study. Red points denote mean utilities and the violin widths reflect the density of individual utilities.

Confidence Bound	$Q_{1}$	$Q_{2}$	$Q_{3}$	Max
Number of Groups	$45$	$8$	$3$	$1$
Value-option Matrices Avg. Distance:	$0.29$	$0.84$	$1.86$	$2.47$
Weight Vectors Avg. Distance:	$0.02$	$0.08$	$0.12$	$0.25$

Table 3. Description of the partitions obtained in the PVE case study. Average distances are computed as the mean pairwise distances between group members’ value–option matrices (

\{X_{i}\}_{i\in Ag}

) and weight vectors (

\{\Omega_{i}\}_{i\in Ag}

Group	Consensual Options Ranking
$P_{1}$	$o_{4}\prec o_{6}\prec o_{5}\prec o_{3}\prec o_{1}\prec o_{2}$
$P_{2}$	$o_{2}\sim o_{1}\prec o_{4}\prec o_{5}\prec o_{6}\prec o_{3}$
$P_{3}$	$o_{4}\sim o_{1}\sim o_{2}\sim o_{3}\prec o_{5}\prec o_{6}$
$P_{4}$	$o_{2}\prec o_{3}\prec o_{5}\sim o_{4}\sim o_{6}\sim o_{1}$
$P_{5}$	$o_{3}\prec o_{4}\prec o_{6}\prec o_{5}\prec o_{1}\sim o_{2}$
$P_{6}$	$o_{6}\prec o_{1}\sim o_{3}\prec o_{2}\prec o_{4}\sim o_{5}$

Table 4. Rankings obtained using TOPSIS from the agreed-upon value systems when considering the confidence bound

Q_{2}

in the PVE case study.

Panel of Figure 3 — Figure 3. Distribution of individual utilities for agreed-upon value weight vectors in the EVS case study. Red points denote mean utilities and the violin widths reflect the density of individual utilities.

Confidence Bound	$Q_{1}$	$Q_{2}$	$Q_{3}$
Number of Groups	$11$	$2$	$1$
Decision Matrices Avg. Distance:	$0.10$	$0.52$	$0.84$
Weight Vectors Avg. Distance:	$0.02$	$0.16$	$0.33$

Table 5. Description of the partitions obtained in the EVS case study. Average distances are computed as the mean pairwise distances between group members’ decision matrices (

\{X_{i}\}_{i\in Ag}

) and weight vectors (

\{\Omega_{i}\}_{i\in Ag}

Figure 4. Groups formed when taking $Q_{2}$ as the global confidence bound in the decentralized value aggregation method. Geographical map of Europe showing the groups of countries formed using the decentralized value aggregation method with the second quartile confidence bound. Countries are partitioned into two groups based on their converged value systems. One group, consisting mainly of Eastern European countries together with Italy and Portugal, is more aligned with religiosity, while the other group, composed primarily of Western European countries, is more aligned with permissiveness. The figure visually illustrates how value-based clustering corresponds to regional patterns.

Multi-Criteria Decision Analysis

Multi-Criteria Decision Analysis (MCDA) ranks or selects alternatives by balancing competing criteria, preferences, uncertainty, and constraints.

See all 35 papers

Beyond Alignment: Value Diversity as a Collective Property in Multicultural Agent Systems
Jun 4, 2026Shaoyang Xu, Jingshen Zhang, Long P. Hoang +2Multi-Agent LLM Systems Multi-Agent Systems
Singapore University of Technology and Design · Washington University in St. Louis
Modelling Human Values for Value-Aware Multi-Agent Systems
Feb 9, 2024Nardine Osman, Mark d'InvernoAI Alignment Multi-Agent Systems
Artificial Intelligence Research Institute (IIIA-CSIC), Barcelona, Spain. · Academy of Music, Hong Kong Baptist University, Hong Kong, China.
VACS: Value-Aligned Compositional Shielding for Multi-Agent Reasoning
Aug 7, 2026Yiyao Zhang, Diksha Goel, Hussain AhmadAI Agent Safety Multi-Agent Reasoning
School of Computing and Information Technology, University of Wollongong, Wollongong, NSW, Australia · CSIRO’s Data61, Clayton, VIC, Australia · School of Computer Science and Information Technology, Adelaide University, Adelaide, SA, Australia

Jun 4, 2026J/K move · Enter open · S save

Beyond Alignment: Value Diversity as a Collective Property in Multicultural Agent Systems

Shaoyang Xu, Jingshen Zhang, Long P. Hoang, Jinyuan Li+1

Singapore University of Technology and Design · Washington University in St. Louis

Multi-Agent LLM Systems Multi-Agent Systems Pluralistic Alignment

Open paper AskPDF arXiv

Multicultural multi-agent systems are increasingly deployed in globally diverse settings, where different agents are grounded in different cultural backgrounds. Existing cultural evaluation focuses on value alignment: how closely a single agent matches a target culture. Yet alignment is a per-agent property and cannot reveal whether a system, taken as a whole, preserves the cultural plurality it is meant to represent. We propose value diversity as a system-level evaluation axis for multicultural agent systems, defined through the dissimilarity between culturally conditioned agents' responses on a shared value survey. Using the World Values Survey, we evaluate 19 cultures and 18 backbone models across a wide range of system configurations. We find that diversity is largely uncorrelated with alignment, indicating that the two capture complementary system properties, and that current multicultural agent systems fall substantially below human societies in value diversity. Mixed-backbone systems narrow this gap but do not close it, and the gap persists across culture compositions and agent scales. Social interaction further erodes diversity by driving agents toward consensus, and a participatory budgeting case study shows that this homogenization narrows the breadth of collective decision-making. Together, our results establish value diversity as a distinct evaluation axis for multicultural multi-agent systems and reveal a persistent homogenization tendency in current LLM-based societies. Our code and data are publicly available at https://github.com/iNLP-Lab/MultiAgent-Diversity.

Abstract

Multi-Criteria Decision Analysis

Explore similar work

Beyond Alignment: Value Diversity as a Collective Property in Multicultural Agent Systems

Modelling Human Values for Value-Aware Multi-Agent Systems

VACS: Value-Aligned Compositional Shielding for Multi-Agent Reasoning