Paper Deep-Dive: Delegation Asymmetry in Agentic Recommender Systems — When AI Dates for You
> "Love is blind, but the dating market is not."
The Scenario
Imagine a dating platform rolls out an AI dating agent that can screen matches, start conversations, keep chatting while you're offline, and even arrange meetups. Sounds great — but here's the overlooked question: if everyone deploys an agent, who actually talks to whose agent?
Research from the Lucy Family Institute for Data & Society (University of Notre Dame) uncovers an awkward market truth: **people are willing to let AI date *for* them, but unwilling to date other people's AI. This "delegation asymmetry" could collapse the entire agentic dating market.
From AI-Assisted to AI-Delegated
1. Algorithmic matching (2000s–2010s): recommendation algorithms (e.g., eHarmony); users fully control communication. 2. AI-assisted (early 2020s): AI polishes profiles, suggests openers; users still lead. 3. AI-delegated (2025s+): agents autonomously screen, converse, and schedule; users intervene only at the last step.
Dating platforms are a classic two-sided market with cross-network effects: your agent is worthless if nobody wants to receive messages from agents. If everyone brings a "social agent" to a party but nobody talks to anyone else's agent, the party becomes agents talking to themselves while owners watch TV at home.
Measuring Two-Sided Receptivity
The core contribution is a two-sided receptivity measurement framework:
Send Receptivity: willingness to have your agent communicate on your behalf.
Receive Receptivity: willingness to receive communication from others' agents.
Two surveys:
| Survey | Time | Language | N | Topic | |---|---|---|---|---| | A | Mar–Apr 2026 | Russian | 2,894 | generative profile features | | B | Nov–Dec 2025 | RU/EN | 2,617 (2,499 complete) | autonomous conversational agents |
Seven ordinal items deliberately span both roles (sender items Y1–Y3, Y7; receiver items Y4–Y6), measured jointly within-person on one latent scale.
Statistical Model
A two-dimensional Graded Response Model with latent regression:
Model comparison: ΔBIC = 51.8 favoring the 2D model; LRT = 114.5 (8 df, p < 10⁻¹⁵). Despite high correlation (ρ = 0.92), send and receive are not the same construct — like driving vs. riding in a car: highly correlated, but different experiences.
Core Finding: Delegation Asymmetry
| Metric | Value | Meaning | |---|---|---| | Deployment threshold | θ = −0.38 | willingness to deploy own agent | | Engagement threshold | θ = +0.32 | willingness to engage others' agents | | Full engagement threshold | θ = +1.39 | fully willing to interact with others' agents | | Gap | 0.71 SD (95% CI [0.65, 0.77]) | quantified delegation asymmetry |
Implied propensities (strict/lenient): deploy 0.38/0.50 vs. engage 0.12/0.26. At the individual level, 40.7% rate send strictly above receive; only 2.1% reverse — a 19:1 ratio.
Four User Segments (Latent Class Analysis)
| Group | Share | Profile | |---|---|---| | Refusers | ~31% | negative on all items; slightly more female | | Enthusiasts | ~19% | positive across the board, including receiving | | Ambivalents | ~25% | lukewarm, near the mean | | Asymmetric delegators | ~26% | key group: 98% willing to try an agent, only 10% willing to fully engage someone else's |
Asymmetric delegators say: "I can send my agent, but don't send yours to me."
Market Counterfactuals
Under random pairing, only 4.4% (strict) / 12.8% (lenient) of potential dyads satisfy both deployment and engagement. There is significant gender-directional imbalance: women are less receptive on both sides (send β=−0.38, receive β=−0.30).
Design Levers
1. Reciprocity requirement (only receptive users may deploy): interaction volume halves — excludes ~2/3 of potential deployers. 2. Receptivity-based routing (AUC = 0.88 out-of-sample): per-contact engagement triples; by receptivity quartile, full-engagement rate rises 0.0% → 2.4% → 8.2% → 37.9%. 3. Transparency & choice: disclose agent use; offer opt-in/opt-out (discussed, not quantified).
Why Do People Refuse to Receive Agents?
Speculative mechanisms from the data:
Predictors of Receptivity
| Factor | Send | Receive | Interpretation | |---|---|---|---| | Female | β = −0.38 | β = −0.30 | lower receptivity on both sides | | Higher match-to-conversation conversion | β ≈ −0.17 | β ≈ −0.17 | successful users less open | | Frustration with stalled matches | β ≈ +0.14 | β ≈ +0.14 | dissatisfied users more open | | Platform tenure | marginally negative | marginally negative | veterans more conservative |
Key insight: users frustrated by stalled matches are significantly more receptive — meaning agents may first be adopted by "desperate users," potentially lowering non-desperate users' willingness further: a possible adverse-selection spiral.
Broader Implications
For dating platforms: conduct two-sided receptivity audits before launch; use receptivity-aware routing; deploy progressively (low- to high-intrusion); ensure transparency and opt-out.
The asymmetry generalizes to all two-sided agentic systems: Customer service: customers accept their own AI helper, not being handed to another AI. Recruiting: candidates accept AI résumé help, not an AI interviewer. Negotiation: you'll send an AI negotiator, but want the counterparty human.
Deep principle: the viability of agentic systems depends on the minimum of two-sided receptivity, not the average.**
A Philosophical Question
> If two people "fall in love" through AI agents, is the relationship real?
Consequentialism might say the outcome matters; Kantian ethics asks whether treating someone as an algorithm output rather than an end-in-themselves is moral. The paper offers data, not answers.
Core Numbers Recap
| Metric | Value | |---|---| | Core survey sample | N = 2,617 | | Latent correlation ρ | 0.92 | | ΔBIC (2D vs 1D) | 51.8 | | Deployment threshold | −0.38 | | Engagement threshold | +0.32 | | Asymmetry gap | 0.71 SD | | Asymmetric delegators | ~26% | | Baseline feasible-pair rate | 4.4–12.8% | | Routing engagement lift | 3× | | Routing AUC | 0.88 |
Closing
Technology can optimize matching efficiency, but not matching *meaning*. The paper shows viable paths — receptivity-aware routing, progressive deployment, transparency and choice — but the data confront us with a truth first: in the two-sided market of agentic dating, supply and demand are severely mismatched; without solving this, even the best technology is a castle in the air.
As one Reddit user put it: "I can accept AI helping me find the right person — but when I first speak to that person, I want it to really be me speaking."