Delegation Asymmetry in Agentic Recommender Systems... | AI Research

Delegation Asymmetry in Agentic Recommender Systems: Measuring Two-Sided Receptivity in Online Dating investigates whether users of dating platforms are willing to both use AI agents to communicate on their behalf and engage with agents used by others. The authors, Daria Leshchikova, Valentina V. Kuskova, Dmitry Zaytsev, and Valerii Klimov, argue that the success of agentic matching depends on this two-sided receptivity, which has previously been studied only in isolation.

Measuring Two-Sided Receptivity

The researchers developed a latent-variable measurement model using data from two large-scale surveys of active dating platform users (N=2,894 and N=2,617). By applying a graded response model with latent regression, they measured two distinct constructs: "send receptivity" (willingness to delegate one's own communication) and "receive receptivity" (willingness to engage with a counterpart's agent). The study found these constructs are highly correlated (ρ = 0.92) but separable, meaning they represent different attitudes within the same user.

The Delegation Asymmetry

The study identifies a systematic "delegation asymmetry," where users are significantly more willing to deploy their own agents than to interact with someone else’s. On a common latent scale, the threshold for deploying an agent is -0.38, while the threshold for engaging with a counterpart’s agent is +0.32. This 0.71 standard deviation gap indicates that the propensity to deploy an agent exceeds the propensity to engage with one by roughly threefold. Consequently, in a random-pairing simulation, only 4–13% of potential matches would successfully combine agent deployment with receiver engagement.

Design Levers for Platforms

The authors propose a "receptivity audit" to help platforms evaluate the viability of agentic features before deployment. Their analysis suggests that platform design choices significantly impact market health. For example, implementing a strict reciprocity requirement—where both parties must use agents—would cut interaction volume by more than half by excluding two-thirds of potential deployers. Conversely, routing agent-mediated contacts specifically to users with higher "receive receptivity" scores could triple per-contact engagement, a finding validated through out-of-sample testing.

Considerations for Implementation

The research indicates that receptivity is not uniform across the user base. It is lower among women and long-tenured users, and it tends to concentrate among individuals who report frustration with stalled matches or a lack of conversation conversion. Because receiver-side consent is a binding constraint, the authors suggest that platforms must prioritize disclosure and opt-in mechanics. The study concludes that agentic recommender design must treat the communication medium itself as a market variable, rather than assuming that the ability to automate outreach will automatically result in successful matches.