Delegation Asymmetry in Agentic Recommender Systems: Measuring Two-Sided Receptivity in Online Dating
Using latent variable graded response models, this study measures users' willingness to deploy and accept agent-mediated communication in online dating, revealing a significant asymmetry with deployment willingness three times higher.
Key Findings
Methodology
This paper employs Item Response Theory (IRT), specifically the Graded Response Model (GRM), combined with latent regression to quantify user receptivity towards agent-mediated communication. Two large-scale surveys collected data on users' attitudes in both roles: deploying their own agents (send receptivity) and engaging with others' agents (receive receptivity). The model estimates item parameters via marginal maximum likelihood (MLE), with model comparison based on Bayesian Information Criterion (BIC) differences to validate the distinction between send and receive dimensions. Partial measurement invariance tests across languages ensure cross-lingual comparability. The study further uses counterfactual simulations based on model-derived probabilities to quantify the systemic asymmetry, revealing that deploying an agent has a lower threshold (−0.38) than engaging with an agent (0.32), with a difference of 0.71 standard deviations. The model predicts that only 4-13% of dyads would both deploy and engage, highlighting a significant barrier to mutual acceptance in agent-mediated interactions.
Key Results
- Model comparison shows that the thresholds for deploying one's own agent (−0.38) and engaging with another's agent (0.32) are statistically distinct (ΔBIC=52), despite a high correlation (ρ=0.92).
- Analysis indicates that only 4% to 13% of user pairs are mutually willing to deploy and accept agents, with a pronounced gender and directional imbalance. Counterfactual simulations suggest that enforcing reciprocity reduces interaction volume by over 50%, while routing based on receive receptivity can triple engagement per contact.
- Predictive validation demonstrates that routing signals based on receive receptivity achieve an AUC of 0.88, with a 3.1-fold lift in engagement, confirming the practical utility of the model for platform optimization.
Significance
This research provides the first large-scale, joint measurement of users' willingness to both deploy and accept agent-mediated communication within a single framework. The findings reveal a systemic asymmetry that could hinder the adoption of autonomous agents in online dating, emphasizing the importance of designing systems that account for both sides' receptivity. Quantifying these dynamics offers platform designers actionable insights, such as leveraging reciprocity norms and routing strategies to enhance mutual engagement, ultimately improving match quality and user trust. The methodological approach also establishes a scalable framework for assessing two-sided market readiness for AI-mediated interactions across various domains.
Technical Contribution
The paper introduces a novel application of multidimensional graded response models with latent regression to jointly measure send and receive receptivity at the individual level, overcoming limitations of traditional attitude surveys. It rigorously tests measurement invariance across languages, ensuring cross-cultural validity. The integration of item response theory with counterfactual market simulations provides a powerful toolkit for quantifying systemic asymmetries and designing targeted interventions. The high predictive accuracy (AUC=0.88) and out-of-sample validation demonstrate the robustness of the approach, paving the way for real-time routing and matchmaking algorithms that are receptivity-aware.
Novelty
This study is the first to simultaneously quantify and compare users’ willingness to deploy and accept agent-mediated communication within a large-scale, real-world dating platform. Its core innovation lies in modeling these two facets as distinct but correlated latent constructs, validated across languages, and linking them to market-level interaction metrics via counterfactual simulations. Unlike prior work that either measures AI acceptance in isolation or focuses solely on user behavior, this research bridges the gap by providing a joint, psychometrically grounded measurement framework that directly informs system design and policy. The integration of IRT with market simulation is a pioneering step in AI-mediated interaction research.
Limitations
- The model assumes static acceptance thresholds, but real-world attitudes are dynamic and context-dependent, which could affect the accuracy of predictions over time.
- Survey data are limited to specific cultural and platform contexts (Russian and English users), potentially restricting cross-cultural generalizability. Further validation in diverse settings is needed.
- Reliance on self-reported attitudes may introduce response biases, such as social desirability or hypothetical bias, which could distort the true willingness levels.
Future Work
Future research should incorporate behavioral tracking data to validate the correspondence between stated willingness and actual engagement patterns. Extending the measurement framework to other domains like professional networking or customer service could test its universality. Additionally, developing dynamic models that capture temporal fluctuations in receptivity and integrating real-time routing algorithms could further enhance platform efficiency. Exploring user-specific customization and privacy-preserving mechanisms will also be crucial for practical deployment.
AI Executive Summary
In the rapidly evolving landscape of digital communication, autonomous agents powered by large language models (LLMs) are beginning to transform how people interact on online platforms. Particularly in online dating, these AI agents can initiate conversations, craft messages, and even negotiate match terms on behalf of users, promising to significantly boost interaction volume and efficiency. However, this technological leap raises fundamental questions about user acceptance and trust, especially since successful deployment depends on mutual willingness: users must not only delegate their communication but also be receptive to agent-mediated contact from others.
This study addresses this critical gap by conducting two large-scale surveys on active users of a major dating platform, capturing attitudes towards deploying personal agents and receiving agent-initiated messages. Using advanced psychometric modeling—specifically, a multidimensional graded response model within the item response theory framework—the authors measure these two facets of receptivity simultaneously at the individual level. The results reveal a striking asymmetry: deploying one’s own agent requires a lower threshold of receptivity (−0.38) compared to engaging with another’s agent (+0.32), with a difference of approximately 0.71 standard deviations. Despite a high correlation (ρ=0.92), statistical tests confirm that these are distinct constructs.
The implications of this asymmetry are profound. The analysis shows that only about 4% to 13% of user dyads are mutually willing to both deploy and accept agent-mediated communication, indicating a systemic barrier to widespread adoption. To quantify the impact of platform design choices, the authors simulate counterfactual market scenarios. They find that enforcing reciprocity—requiring mutual high receptivity—can reduce interaction volume by over 50%, but strategic routing based on receive receptivity can triple engagement rates per contact. These findings suggest that platforms can significantly improve interaction outcomes by incorporating receptivity-aware routing mechanisms.
Beyond the technical insights, the study highlights important social and ethical considerations. It uncovers gender differences in receptivity, with women and long-tenured users showing lower acceptance levels, and emphasizes the importance of transparency and opt-in mechanisms for deploying autonomous agents. The predictive accuracy of the model (AUC=0.88) and its validation in out-of-sample tests demonstrate its robustness and practical utility.
Overall, this research pioneers a rigorous, scalable approach to measuring two-sided market readiness for AI-mediated communication. Its findings inform the design of more receptivity-sensitive matchmaking algorithms, fostering trust and mutual engagement in AI-enhanced online dating. As autonomous agents become more prevalent, understanding and addressing the systemic asymmetries in user acceptance will be essential for realizing their full potential in creating meaningful digital connections.
Deep Dive
Abstract
Autonomous LLM agents that converse on a user's behalf are an emerging design pattern in matching platforms, yet their viability depends on a condition rarely examined: users must accept not only delegating conversation to an agent, but also receiving agent-mediated communication from others. We study this condition using two large-scale surveys of active users of a major dating platform (N=2,894 on generative profile features; N=2,617 on autonomous conversational agents, fielded in two languages). We develop a latent-variable measurement model of agent receptivity based on graded response models with latent regression, and show via model comparison that willingness to send and willingness to receive agent communication are distinct constructs: highly correlated (rho=0.92) but separable (Delta BIC=52), with partial measurement invariance across languages. The model quantifies a systematic delegation asymmetry: deploying one's own agent requires far lower receptivity (threshold -0.38) than engaging a counterpart's agent (+0.32; full engagement +1.39), and mean deployment propensity exceeds engagement propensity roughly threefold. Under a random-pairing counterfactual derived from stated receptivity, only 4-13% of directed dyads combine agent deployment with receiver engagement, with a pronounced gender-directional imbalance. Design counterfactuals quantify the levers: a reciprocity requirement cuts interaction volume by half or more by excluding nearly two-thirds of would-be deployment, while routing agent contacts on receive receptivity triples per-contact engagement, a lift that survives out-of-sample validation with the target item held out (AUC 0.88, 3.1x quartile lift under respondent-level cross-validation). We discuss implications for agentic recommender design, including disclosure, opt-in mechanics, and receptivity-aware matchmaking.
References (20)
The Cyrano effect: LLM-assisted impression management and authenticity in online dating
Lennart Ante
Disintermediating your friends: How online dating in the United States displaces other ways of meeting
Michael J. Rosenfeld, Reuben J. Thomas, Sonia Hausen
Latent variables in psychology and the social sciences.
K. Bollen
Reciprocal Recommender Systems: Analysis of state-of-art literature, challenges and opportunities towards social recommendation
I. Palomares, C. Porcel, L. Pizzato et al.
The General Attitudes towards Artificial Intelligence Scale (GAAIS): Confirmatory Validation and Associations with Personality, Corporate Distrust, and General Trust
Astrid Schepman, P. Rodway
Recommending people to people The nature of reciprocal recommenders with a case study in online dating
L. Pizzato, Tomek Rej, Joshua Akehurst et al.
Estimation of latent ability using a response pattern of graded scores
F. Samejima
Attitudes towards AI: measurement and associations with personality
Jan-Philipp Stein, Tanja V. Messingschlager, T. Gnambs et al.
Fair Reciprocal Recommendation in Matching Markets
Yoji Tomita, Tomohiko Yokoyama
Marginal maximum likelihood estimation of item parameters: Application of an EM algorithm
R. D. Bock, M. Aitkin
iAgent: LLM Agent as a Shield between User and Recommender Systems
Wujiang Xu, Yunxiao Shi, Zujie Liang et al.
Matching and Sorting in Online Dating
Günter J. Hitsch, Ali Hortaçsu, D. Ariely
Generative AI and Perceptual Harms: Who's Suspected of using LLMs?
Kowe Kadoma, D. Metaxa, Mor Naaman
Perceived Usefulness, Perceived Ease of Use, and User Acceptance of Information Technology
Fred D. Davis
Predicting User Replying Behavior on a Large Online Dating Site
Peng Xia, Hua Jiang, Xiaodong Wang et al.
Generalized Latent Variable Modeling: Multilevel,Longitudinal, and Structural Equation Models
S. Lipovetsky
User Acceptance of Information Technology: Toward a Unified View
V. Venkatesh, Michael G. Morris, G. Davis et al.
The V–Dem Measurement Model: Latent Variable Analysis for Cross-National and Cross-Temporal Expert-Coded Data
Daniel Pemstein, Kyle L. Marquardt, Eitan Tzelgov et al.
Development and validation of the AI attitude scale (AIAS-4): a brief measure of general attitude toward artificial intelligence
Simone Grassini
Thinking twice about sum scores
Daniel M. McNeish, M. Wolf