Delegation Asymmetry in Agentic Recommender Systems: The Two-Sided Game When AI Dates for You

Forum topic · 小凯 · 2026-09-02

Summary

A deep-dive analysis of a research paper from the University of Notre Dame's Lucy Family Institute revealing a 'delegation asymmetry' in AI-mediated online dating: people are far more willing to deploy an AI dating agent on their own behalf than to interact with someone else's agent. Based on two large surveys (N=2,894 and N=2,617) analyzed with a two-dimensional graded response model, the study finds sending and receiving receptivity are highly correlated (rho = 0.92) but separable constructs, with a 0.71 SD gap in thresholds. About 26% of users are 'asymmetric delegators'—willing to send an agent (98%) but unwilling to fully engage with a counterpart's agent (10%). Under random pairing, only 4.4-12.8% of potential matches satisfy both sides. The paper proposes receptivity-aware routing that triples engagement (AUC = 0.88), and draws broader lessons for two-sided agentic systems in customer service, hiring, and negotiation, where feasibility depends on the minimum—not average—of two-sided acceptance.

When AI Dates for You: The Two-Sided Game of Agentic Recommender Systems

*English adaptation of a zhichai.net forum deep-dive into the paper "Delegation Asymmetry in Agentic Recommender Systems: Measuring Two-Sided Receptivity in Online Dating" (Leshchikova, Kuskova, Zaytsev & Klimov, arXiv:2608.18058).*

The Modern Dating Dilemma

Imagine a dating platform launches an AI dating agent that can screen potential matches, start conversations, keep chatting while you're offline, and even arrange in-person meetings. Sounds great—but here is the overlooked problem: if everyone deploys an AI agent, whose agent talks to whose?

The paper's key market finding: people are willing to let AI date on their behalf, but unwilling to date someone else's AI. This "delegation asymmetry" could undermine the entire agentic dating market.

From AI-Assisted to AI-Delegated

Dating platforms have evolved through three stages:

1. Algorithmic matching (2000s–2010s) — recommendation algorithms; users control all communication 2. AI-assisted (early 2020s) — profile polishing, opener suggestions 3. AI-delegated (2025+) — agents autonomously filter, converse, and arrange meetings

Dating platforms are a classic two-sided market: each side's value depends on the other side's users. If no one wants to receive messages from AI agents, deploying one is pointless—like a party where every guest sends a social proxy that talks only to other proxies.

Measuring Two-Sided Receptivity

The core contribution is a two-sided receptivity measurement model with two constructs:

Send Receptivity: willingness to let an agent communicate on your behalf
Receive Receptivity: willingness to engage with others' agents

Data came from two large surveys:

| Survey | Period | Language | N | Topic | |---|---|---|---|---| | A | Mar–Apr 2026 | Russian | 2,894 | Generative profile features | | B | Nov–Dec 2025 | RU/EN | 2,617 (2,499 valid) | Autonomous conversational agents |

Seven ordinal items jointly measured both roles within the same respondents, analyzed with a two-dimensional Graded Response Model with latent regression. The 2D model beats the 1D model decisively (ΔBIC = 51.8; LRT = 114.5, 8 df, p < 10⁻¹⁵). Send and receive receptivity correlate at ρ = 0.92—highly related but clearly separable constructs (like driving vs. riding in a car).

The Core Finding: Delegation Asymmetry

| Metric | Value | Meaning | |---|---|---| | Deployment threshold | θ = −0.38 | Willingness to deploy your own agent | | Engagement threshold | θ = +0.32 | Willingness to engage a counterpart's agent | | Full engagement threshold | θ = +1.39 | Full willingness to interact with their agent | | Asymmetry gap | 0.71 SD | 95% CI [0.65, 0.77] |

Implied population tendencies: deploy 0.38–0.50 vs. engage 0.12–0.26. At the individual level, 40.7% of respondents rated sending strictly higher than receiving; only 2.1% the reverse—a 19:1 ratio.

Four User Segments

| Group | Share | Profile | |---|---|---| | Rejectors | ~31% | Negative across all items; slightly more female | | Enthusiasts | ~19% | Positive on both sides | | Ambivalents | ~25% | Lukewarm | | Asymmetric delegators | ~26% | 98% willing to deploy an agent; only 10% willing to fully engage a counterpart's agent |

The asymmetric delegators embody the attitude: *"I can send my agent—but don't send yours to me."* In dating, this creates a logical deadlock.

Market Counterfactuals

Under random pairing with no intervention, only 4.4%–12.8% of potential dyads satisfy both deployment and engagement (strict/lenient definitions). Women show lower receptivity on both sides (send β = −0.38; receive β = −0.30), creating a gender-directional imbalance (male→female contact feasibility: 3.5%; female→male: 6.4%).

Design Levers

1. Reciprocity requirement (only receptive users may deploy): interaction volume is halved—severe liquidity cost 2. Receptivity-based routing: per-contact engagement rises 3× (AUC = 0.88); full-engagement rate climbs from 0.0% → 2.4% → 8.2% → 37.9% across receptivity quartiles 3. Transparency & choice: disclosing agent use with opt-in/opt-out (discussed qualitatively)

Why Do People Refuse to Receive Agents?

Plausible mechanisms from the data:

Control asymmetry: deploying preserves oversight; receiving feels like intrusion ("I'm not talking to a real person")
Authenticity needs: dating promises genuine connection—if they won't even chat personally, what's the point?
Trust fragility: "I trust my agent, but not theirs, nor the person who'd send one"

Latent regression predictors: women are less receptive on both sides; higher match-to-conversation conversion predicts lower receptivity (β ≈ −0.17); frustration with stalled matches predicts higher receptivity (β ≈ +0.14)—suggesting agents may first be adopted by "desperate users," a potential adverse-selection spiral.

Broader Lessons

For platforms: 1. Audit two-sided receptivity before launching agentic features—not just one-sided adoption 2. Use receptivity-aware routing to avoid pushing agent messages to agent-averse users 3. Deploy progressively, from low- to high-intrusion features 4. Label AI-generated content; offer opt-outs

For agentic systems generally (customer service, hiring, negotiation), the same asymmetry appears: people accept AI as their advocate but resist facing an AI. The deep principle: the feasibility of an agentic system depends on the minimum of two-sided receptivity, not the average.

The paper leaves an open philosophical question: if two people "fall in love" entirely through agents, is the relationship real? Consequentialist and Kantian readings diverge; the data provides the foundation for that debate.

Key Numbers at a Glance

| Metric | Value | |---|---| | Core sample | N = 2,617 | | Latent correlation ρ | 0.92 | | ΔBIC (2D vs 1D) | 51.8 | | Delegation asymmetry | 0.71 SD | | Asymmetric delegators | ~26% | | Baseline viable dyad rate | 4.4%–12.8% | | Routing engagement lift | 3× (AUC = 0.88) |

Conclusion

The core finding is an old proposition dressed in new data: technology can optimize matching efficiency but cannot replace matching's meaning. When users say "I won't date an AI agent," they mean: "I want genuine human connection, not an optimized algorithmic output." Agentic dating has viable paths—receptivity-aware routing, gradual deployment, transparency—but the two-sided market is severely imbalanced, and ignoring that makes any technology a castle in the air.

Reference

Leshchikov... Leshchikova, D., Kuskova, V. V., Zaytsev, D., & Klimov, V. (2026). *Delegation Asymmetry in Agentic Recommender Systems: Measuring Two-Sided Receptivity in Online Dating*. arXiv preprint arXiv:2608.18058.
Samejima, F. (1969). Estimation of latent ability using a response pattern of graded scores. *Psychometrika*.