Advances in artificial intelligence (AI), particularly in large language models (LLMs), increasingly enable AI to serve as an active counterpart in negotiation. Drawing on negotiation theory and mind perception research, this study examines whether attributed counterpart identity, that is, the belief that a negotiation counterpart is human rather than AI, influences relational evaluations, satisfaction, and outcomes in a mixed-motive bargaining context. In a randomized laboratory experiment, 205 participants negotiated apartment rent with the same LLM-based broker but were led to believe the counterpart was either human or AI. After correction across the five Hypothesis 1 comparisons, participants who believed they negotiated with a human reported higher perceived warmth and integrity; perceived competence was higher only in the unadjusted comparison. Participants in the Person condition also reported higher relational satisfaction and agreed to less favorable economic outcomes, paying higher final rents than those who believed they negotiated with AI. Exploratory analyses further suggested that attributed counterpart identity shaped how participants organized and described their post-negotiation experiences, with human-attributed counterparts eliciting more differentiated evaluations and more partnership-oriented descriptions. As AI becomes more common in negotiation, these results highlight the importance of accounting for the relational value people derive from believing they are negotiating with another person.
Understanding how much a group of individuals disagrees in their preferences is a fundamental problem in collective decision-making. When the number of participants and available options is large, measuring this disagreement rigorously and efficiently becomes particularly demanding. The present study addresses this challenge by introducing a formally grounded measure of preference heterogeneity for large-scale group decision-making settings, in which individual preferences are expressed as complete rankings over a set of alternatives, allowing for ties. The proposed measure is derived from Shannon’s entropy and quantifies the degree of diversity in the distribution of individual preferences across a group. A value of zero indicates full agreement, whereas higher values reflect greater dispersion of opinions. The measure is shown to satisfy a set of desirable normative properties drawn from Social Choice Theory, including anonymity, neutrality, and conditions governing how heterogeneity changes when the group or the set of alternatives is modified. An axiomatic characterisation establishes that the proposed measure is the unique index satisfying these properties jointly. The computational simplicity of the measure makes it applicable to very large populations without loss of tractability. Its practical performance is evaluated through ten real-world ranked-choice elections from a publicly available preference database, encompassing electorates of up to nearly three hundred thousand voters and up to eleven candidates. Simulation experiments conducted under diverse probabilistic scenarios further confirm the robustness and sensitivity of the measure to variations in preference dispersion.
In many group decision-making problems, the interval multiplicative preference relation (IMPR) is widely used to express uncertain information. To effectively consider the stochastic characteristics of preference relations and expert weights, this paper proposes a novel framework for two-stage stochastic group preference analysis based on the combined compromise solution (CoCoSo) method. First, the stochastic multiplicative preference relations are generated from IMPRs according to a specified probability distribution, and experts’ preference priority weight vectors are determined with the logarithmic least squares method. Then, a comprehensive preference priority weight matrix is constructed, which serves as the decision matrix in the CoCoSo method to compute the comprehensive preference ranking degree of alternatives. Meanwhile, three descriptive measures are proposed to analyze the ranking results, including the stochastic preference rank acceptability index, the stochastic preference central weight vector, and the stochastic preference confidence level. Moreover, the proposed framework combines stochastic preference information and stochastic expert weights, thereby avoiding information loss and reducing the subjective influence of experts. Finally, a logistics decision case of food transportation is analyzed, and sensitivity and comparative analyses are performed to verify the potential effectiveness of the proposed method.
This study presents MGDM-Sy, a multicriteria group decision-making system that supports heterogeneous preference elicitation within a unified computational environment. The system allows evaluators participating in the same decision problem to express their preferences using different preference elicitation formats, including ordering of alternatives, direct scoring, and pairwise comparison. Heterogeneous preference representations are automatically transformed into a common score representation suitable for aggregation. For pairwise comparisons, MGDM-Sy incorporates a consistency-improvement mechanism that assists decision-makers in identifying and correcting inconsistent judgments while preserving the original assessments as much as possible. The system also performs consensus analysis at both the individual and group levels and generates decision-support reports. In addition, MGDM-Sy incorporates two methodological contributions: the Card Sorting method, which combines ordering and pairwise comparison while allowing the assignment of zero importance to alternatives, and the Consensus Degree-Enhancement tool, which improves group consensus by filtering assessments according to their individual consensus degree. The system is illustrated through two multicriteria group decision-making applications: (i) the prioritization of regions for agrivoltaic system deployment and (ii) the weighting of the criteria composing the Cost of Doing Business Index. The results demonstrate the applicability of the proposed framework and show that the Consensus Degree-Enhancement tool increases the group consensus degree while preserving representative assessments that would otherwise be unnecessarily excluded.
