arXiv Science⌕ Search
Home/Archive/arXiv

arXiv · 2610.32726

The Butterfly Defects: Tracing Cooperation in LLM Agent Networks

Maria Sampani·Maria Lymperaiou·Angeliki Dimitriou·Giorgos Filandrianos·Chrysoula Zerva·Giorgos Stamou
View original source ↗Read source document ↗Export metadata ↓

Abstract

As large language models (LLMs) increasingly operate as interconnected agents, a fundamental question emerges: does individual cooperation translate into collective cooperation? We investigate this question through the Iterated Prisoner's Dilemma (IPD), moving from dyadic interactions to homogeneous and heterogeneous networks of open-weight LLMs. Our findings reveal that cooperation is highly sensitive to model composition and seemingly minor linguistic changes. Models that cooperate with identical peers can shift toward defection in mixed populations. A simple uncertainty cue can increase or decrease cooperation, redistribute costs, and affect unexposed agents several hops away. Even reversing action-label order can dramatically alter behavior, with uncertainty framing sometimes amplifying this sensitivity. Together, these findings challenge the notion of cooperation as a stable model-level property, revealing how strategic behavior depends on interaction partners, network positions, and the language used to frame decisions.

Explore related subjects

physics.soc-phcs.CL

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Maria Sampani, Maria Lymperaiou, Angeliki Dimitriou, Giorgos Filandrianos, Chrysoula Zerva, Giorgos Stamou. 2026-10-23. The Butterfly Defects: Tracing Cooperation in LLM Agent Networks. https://arxiv.org/abs/2610.32726

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

An Epidemiological Modeling Take on Religion Dynamics

Religions are among the most consequential social institutions, shaping collective identities, moral norms, and political organization across societies and historical periods. Nevertheless, despite extensive scholarship describing conversion, competition, and secularization, there is still no widely adopted formal model that captures religious dynamics over time within a unified, mechanistic framework. In this study, we propose an epidemiologically grounded model of religious change in which religions spread and compete analogously to co-circulating strains. The model extends multi-strain compartmental dynamics by distinguishing passive believers, active missionaries, and religious elites, and by incorporating demographic turnover and mutation-like splitting that endogenously generates new denominations. Using computer simulations, we show that the same mechanism reproduces canonical qualitative regimes, including emergence from rarity, rapid expansion, long-run coexistence, and transient rise-and-fall movements. A reduced calibration variant fits historical affiliation trajectories with parsimonious regime shifts in effective recruitment and disaffiliation, yielding interpretable signatures of changing social conditions. Finally, sensitivity analyses map sharp regime boundaries in parameter space, indicating that modest shifts in recruitment efficacy or retention among active spreaders can qualitatively alter long-run religious landscapes. These results establish a general, interpretable framework for studying religion as a dynamical diffusion process and provide a tool for comparative inference and counterfactual analysis in sociological research.

physics.soc-ph↗
arXiv2026

An Eco-Friendly Tourism Trip Planner Driven by Z-Number Uncertainty: Evidence from Baku, Dubai, Abu Dhabi, Tashkent, and Istanbul

Deployed tourism trip planners ingest travel times and ratings as exact numbers, although routing services return travel times that vary with the hour of departure and a platform rating may rest on a handful of reviews or on hundreds of thousands. We present an eco-friendly trip planner that represents both quantities as Z-numbers, which pair a fuzzy restriction on a value with a fuzzy reliability on the evidence behind it, and that constructs them directly from routing-service observations at six departure hours and from platform review counts rather than from linguistic elicitation. A hybrid loss combines score, carbon, travel-time, and reliability terms, and an adaptive large neighborhood search returns single-day itineraries through a conversational front end. We compare the planner with a crisp-mean baseline under an identical optimizer in Baku, Dubai, Abu Dhabi, Tashkent, and Istanbul, with fifteen points of interest and thirty seeds per city, and we evaluate the selected itineraries on their trip carbon and their total travel time. Pooled over 150 seed-paired runs, the planner lowers trip carbon from 3.24 to 2.61 kg (19.3%, Wilcoxon p = 0.017; city-clustered bootstrap interval 6.4% to 26.7%) and travel time from 123 to 109 minutes per trip (11.5%, p = 2.8e-10; interval 5.0% to 18.2%). The travel-time reductions survive a Holm correction in Baku, Istanbul, and Abu Dhabi; the carbon reduction is significant in Baku and large but uncertain in the dispersed catalogs of Dubai and Abu Dhabi; and neither metric changes in Tashkent. Review counts are estimates of at least 3,000 for every point of interest, so rating reliability does not vary, and with a single baseline, the results establish feasibility and direction rather than a general advantage. Code, routing caches, frozen routing tables, and per-seed results are openly released.

physics.soc-ph↗
arXiv2026

Conviction versus Conformity in a Random-Field Ising Electorate: Antagonistic Ties, Network Density, and a Parameter-Free Prediction

We study the zero-temperature random-field Ising model with +-J bond disorder on sparse graphs as a model of an electorate: survey-derived, trimodal fields compete with social couplings, a fraction alpha of which are antagonistic. With exact ground states up to N=180 (over 78000 optimizations, certified except in the densest frustrated cases) and validated heuristics to N=720, we locate the crossover r_h* at which the field carries half of the ground-state energy, close to where half the voters held against their own field have switched (0.93 r_h* for alpha<=0.2). Our main test is a zero-temperature cavity theory solved on the graph ensemble with no fitted parameter: at seven densities not otherwise used, predicted before their exact ground states were computed, it placed r_h* within 10% in 19 of 21 cells (median error 3%). Its errors are unsigned at alpha=0.05 and 0.5 but all negative at alpha=0.2, where it predicts the coupling energy correctly but a crossover that is early and, at high density, too sharp, and where its replica-symmetric solution is unstable in six of seven cells, consistent with replica-symmetry breaking at intermediate frustration. Empirically r_h*~0.91 e_J(0), with e_J(0) the coupling energy of the zero-field ground state, inside a rigorous per-instance factor-of-two window, so the known +-J energy laws carry over: r_h* ~ ^gamma with gamma=1.01-1.09 for alpha=0.05 and 0.50-0.57 for alpha>=0.5 ( =3-28; five graph families agree at matched density). At intermediate frustration r_h* is not a power law: sparse networks accommodate antagonistic ties at the cost of cooperative ones, which stops paying near ~10; the theory reproduces this without adjustment. In weakly frustrated dense networks the ground state is a near-consensus following the sign of the aggregate field in 93-100% of realisations at small r_h, against near-chance agreement in frustrated ones.

physics.soc-ph↗