Value Preferences Estimation and Disambiguation in Hybrid Participatory Systems

TL;DR

This work tackles the problem of inferring individual value preferences in hybrid participatory systems by combining participants’ choices with their value-laden motivations. It introduces five value-preference estimation methods and a disambiguation strategy that uses an Active Learning–style loop to resolve conflicts between choices and motivations, evaluated on a large energy-transition Participatory Value Evaluation dataset. Empirical results show that incorporating motivations improves alignment with human judgments, while the disambiguation strategy yields NLP performance on par with baselines and mixed gains in value-estimation; the approach provides a scalable pathway toward more legitimate and context-sensitive policy deliberation. Overall, the framework advances citizen-centric policy design by enabling more accurate, interpretable, and interactive estimation of individual value preferences, with potential for population-level aggregation and value-aligned decision-making.

Abstract

Understanding citizens' values in participatory systems is crucial for citizen-centric policy-making. We envision a hybrid participatory system where participants make choices and provide motivations for those choices, and AI agents estimate their value preferences by interacting with them. We focus on situations where a conflict is detected between participants' choices and motivations, and propose methods for estimating value preferences while addressing detected inconsistencies by interacting with the participants. We operationalize the philosophical stance that "valuing is deliberatively consequential." That is, if a participant's choice is based on a deliberation of value preferences, the value preferences can be observed in the motivation the participant provides for the choice. Thus, we propose and compare value preferences estimation methods that prioritize the values estimated from motivations over the values estimated from choices alone. Then, we introduce a disambiguation strategy that combines Natural Language Processing and Active Learning to address the detected inconsistencies between choices and motivations. We evaluate the proposed methods on a dataset of a large-scale survey on energy transition. The results show that explicitly addressing inconsistencies between choices and motivations improves the estimation of an individual's value preferences. The disambiguation strategy does not show substantial improvements when compared to similar baselines--however, we discuss how the novelty of the approach can open new research avenues and propose improvements to address the current limitations.

Figures (8)

• Figure 1: A hybrid participatory system where human participants make choices and motivate those choices, and AI agents estimate and disambiguate participants' value preferences.
• Figure 2: Each PVE participant makes choices $C$ (i.e., distributes points to the policy options) and provides motivations $M$ to their choices. The participant's value system is defined as the ranking $R$ over a set of values $V$. Our proposed value preferences estimation methods estimate $R$ based on (1) a given value list $V$, (2) the choices $C$, (3) the values annotated in the motivations, and (4) an initial estimate of their value-option matrix $VO$.
• Figure 3: Overview of the five proposed value preferences estimation methods.
• Figure 4: Overview of the proposed disambiguation strategy, guided by the detected inconsistencies between value preferences estimated from participants' choices and motivations.
• Figure 5: Performance of the value preferences estimation methods, measured as the overlap with the evaluators' answers.