Pith. sign in

REVIEW 4 major objections 8 minor 77 references

Empowering Economic Simulation for Massively Multiplayer Online Games through Generative Agent-Based Modeling

T0 review · 4 major / 8 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper argues that LLM-driven agents, built as MMOAgent with profile, perception, reasoning, memory, and action modules, reproduce emergent MMO economic phenomena such as role specialization and market-conforming price fluctuations.

desk verdict Useful industrial sandbox with a real architecture contribution, but the 'emergent economics' claim is largely prompt-conditioned and needs controls. read the letter →

arxiv 2506.04699 v1 pith:RTDRFYLH submitted 2025-06-05 cs.AI

Bihan Xu , Shiwei Zhao , Runze Wu , Zhenya Huang , Jiawei Wang , Zhipeng Hu , Kai Wang , Haoyu Liu
show 5 more authors
Tangjie Lv Le Li Changjie Fan Xin Tong Jiangze Han
This is my paper
classification cs.AI
keywords MMOeconomysimulationgenerativeagent-basedmodelinglargelanguagemodelagentsplayer-to-playertradinggameeconomicsemergentbehaviorsupplyanddemandrolespecialization
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper argues that replacing rule-based or reinforcement-learning agents with LLM-driven generative agents makes MMO economy simulations more human-like. It builds MMOAgent, whose modules for profile, perception, reasoning, memory, and action let agents role-play, perceive the environment, remember successful experiences, and explain their choices. The authors extend an MMO economy environment with direct player-to-player trading through public and private chat, enabling verbal negotiation and bargaining. In a 200-step simulation with 30 agents, they report role specialization, P2P prices below auction prices, a 0.67 correlation between auction price and the demand-supply gap, and an equality-profitability tradeoff. A sympathetic reading is that this is a first demonstration that LLM agents can serve as a usable generative agent-based model for studying and testing virtual economies.

What carries the argument

The central object is MMOAgent, an LLM-powered agent with five modules: a data-driven profile built by k-means clustering of real player records and rendered by an LLM into text personas; a perception module that parses inventory, auction orders, nearby resources, and chat messages into text; structured actions with execution feedback, so failed actions return an explicit explanation to the reasoning module; feedback-enhanced reasoning using zero-shot chain-of-thought prompting plus periodic self-reflection; and a numeric-aware long short-term memory where successful trajectories are stored as numerical embeddings, retrieved by similarity and importance, and forgotten along an exponential curve. The newly added mechanism is direct P2P trading with public and private chatting, which lets agents broadcast offers and bargain privately, with auction order information feeding their price negotiation.

What would settle it

A concrete check: run the same 30-agent simulation with the 'Auction info' line removed from the parsed observation, or with the price-setting request phrased neutrally, and test whether the auction-price/demand-supply correlation of 0.67 survives and whether P2P negotiation still settles below auction prices. If the correlation collapses to near zero, the reported market behavior is prompt-following, not emergence.

Watch

Extended reading notes

Core claim

The central claim is that LLM-empowered agents can simulate MMO economic activity with human-like reliability, sociability, and interpretability, and that the macro-level outcomes they produce are economically meaningful. Concretely, the paper claims that agents whose profiles are built from real player data via clustering and LLM-generated personas specialize into economic roles, negotiate in natural language to close P2P trades at prices below auction prices, and generate auction prices that track the demand-supply gap with a Pearson correlation of 0.67 ($p < 0.001$). The paper also claims that the simulation reproduces the equality-profitability tradeoff: resource-rich environments equalize outcomes but reduce recharge-driven profit, while resource-scarce environments do the opposite. The authors present these results as evidence that generative agent-based modeling can overcome the reliability, sociability, and interpretability limitations of rule-based and RL-based MMO economy simulations.

Load-bearing premise

The simulation's claim that price fluctuations follow market rules rests on a 0.67 correlation between auction price and the demand-supply gap, but agents are explicitly shown the auction's selling and bidding orders and asked to set bid and ask prices, so the correlation could reflect agents following the prompt rather than an emergent market dynamic.

Editorial extensions

If this is right

  • Game studios could use MMOAgent to test economic policy changes in simulation before deploying them to live players, since the agents adapt strategies through reflection and memory.
  • The P2P negotiation result implies that verbal bargaining measurably lowers transaction prices, so MMO economic models should treat negotiation as a price-forming mechanism rather than assuming auction-only clearing.
  • The equality-profitability tradeoff finding suggests designers can tune resource density to choose a desired point on the fairness-versus-revenue frontier.
  • The modular five-component design implies the framework can be ported to other simulated economies by redefining the environment, tasks, action space, and prompts.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The reported role specialization may partly be prompt-following: agents receive profiles that say who spends and who grinds, so the cleanest test of emergence would shuffle or blank profiles and see whether specialization persists.
  • The 0.67 supply-demand correlation is reported for one 30-agent auction setting; testing across resource densities, agent counts, and LLM backbones would show whether the market regularity is a property of the simulation or of the single configuration.
  • If the framework transfers to non-game economies, the same profile-perception-reasoning-memory-action loop could simulate policy interventions in markets where negotiation matters, with LLM call cost as the main scaling constraint.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 8 minor

Summary. The paper extends the MMO Economist environment with direct player-to-player (P2P) linguistic negotiation and proposes MMOAgent, an LLM-driven generative agent with profile, perception, reasoning, memory, and action modules. Experiments compare MMOAgent against Random, rule-based, RL, ReAct, and Reflexion baselines on capability and diversity, test profile consistency, and claim to observe emergent role specialization and auction prices obeying supply-demand rules. The paper concludes that LLM-empowered agents can promote emergent economic phenomena in MMO simulations.

Significance. If the results hold, the paper provides a useful generative-ABM system for MMO economy simulation, with an explicitly released codebase and a plausible modular architecture. The capability gains over the RL baseline are credible in direction, and the P2P negotiation extension addresses a real gap relative to the MMO Economist. However, the headline evidence for 'emergent phenomena' is currently weakened by a conditioning problem: the auction price correlation in Section 5.3.3 is likely inflated because agents are handed exactly the auction order information they are asked to price. The role-specialization evidence is similarly partly designed-in through profile texts. These issues affect the central claim as stated, so the paper needs revision rather than acceptance in its present form.

major comments (4)
  1. [Section 5.3.3, with Section 4.2 and Appendix B] The Pearson correlation r=0.67 between auction price and the demand-supply gap is not convincing evidence of emergent market-rule-following. The parsed observation in Section 4.2 explicitly includes 'Auction info' with selling and bidding orders and the past average deal price, and Appendix B (Table 3) states that Auction_Buy and Auction_Sell prices are set by LLM calls. Thus the LLM sees the exact supply and demand variables before choosing bid and ask prices, so a prompt-following agent would produce a positive correlation even in the absence of any market mechanism; the past average price also injects autocorrelation. Please add a non-LLM pricing control that maps observed auction information to prices with a simple rule, an ablation that removes or perturbs the 'Auction info' field, the full price-setting prompt, and a residual or lagged analysis that separates the effect of the provided order book from genuinely emergent price dynamics.
  2. [Section 5.2.2, with Section 4.1 and Table 4] The claim that role specialization is an emergent phenomenon is overstated. The profiles in Table 4 explicitly encode time and money preferences (e.g., 'Invest substantial time but little money' for Engaged Grinder), and the agent decisions are evaluated for consistency with these profiles in Section 5.2.1. Under these conditions, observing that cost-averse agents favor tasks while spending-prone agents recharge is largely a check of prompt/profile compliance, not of unscripted economic role formation. Please add control experiments: shuffled profiles, a single homogeneous profile assigned to all agents, and no-profile agents; then measure whether specialization patterns shift accordingly.
  3. [Section 5.1.2 and Table 1] The capability comparison reports only mean values over five repetitions, without standard deviations, confidence intervals, or per-trial data. The asterisk in Table 1 indicates p<0.05 for MMOAgent versus the MMO-economist baseline, but no dispersion is reported for any method, and it is unclear whether the t-test is paired or what the pairing unit is. Because the central performance claim rests on these numbers, please report the full distribution (or at least mean, std, and CI) for each baseline and each ablation, and state the exact test procedure.
  4. [Section 5.3.2] The comparison of average transaction prices for MAT (6.86 tokens in auctions vs 6.46 in P2P trades) is reported without sample sizes, variance, or a significance test. As this is the only quantitative evidence for a distinct P2P price channel, please add error bars or a statistical comparison, and clarify how transaction prices are aggregated across agents and over the 200-step simulation.
minor comments (8)
  1. [Section 4.5.2 and Appendix E] The long-term memory is described without an explicit capacity bound in the main text; only Appendix E introduces an LTM size of 20. Since Section 4.5.2 says similar records have their importance scores accumulated, it is unclear how new records are inserted or evicted when the memory reaches capacity. Please specify the capacity mechanism and the merging/eviction rule in the main text or formalize the pseudocode.
  2. [Section 6] The limitation statement acknowledges that agents 'may still generate illegal actions during decision-making, like upgrading without resources or overbidding in auction without enough tokens,' and that limited game knowledge 'reduces action sequence variety.' These limitations are relevant to the interpretability and realism claims made in Sections 5.1.4 and 5.2, so please quantify how often illegal or conservative actions occur and discuss how they affect the reported capability and role-specialization results.
  3. [Section 5.3.3] The 'Demand Supply Gap' variable is used to compute the headline correlation but is never precisely defined in the main text. Please provide its formula (e.g., total bid quantity minus total ask quantity, or number of orders), the time window over which it is aggregated, and whether the correlation is computed on raw prices or averaged series; the current Figure 7(a) caption does not specify these details.
  4. [Section 5.3.4 and Equation (7)] The equality-profitability trade-off result is interesting but presented only as a qualitative directional pattern. Please report the actual profit and equality values for the three scenarios, with error bars, so that the claimed monotonic relationship is verifiable from the text.
  5. [Section 5.2.1 and Appendix D.3] The GPT-4-based consistency rating is said to have 95% agreement with three human evaluators on a 20% sample, but the exact number of evaluated subsequences, the distribution of ratings, and the inter-rater agreement among human annotators are not reported. Please provide these details to make the evaluation reproducible.
  6. [General reproducibility] The paper provides a source-code link, which is a strength, but it does not provide the exact prompts used for profile generation, reasoning, auction pricing, or P2P negotiation. Since prompt design is a major component of the system, please release the prompts or include them in an appendix to make the results reproducible and the conditioning concern in the first major comment assessable.
  7. [Section 3, caption of Figure 6] The caption 'The role specification of agents' in Figure 6 should say 'role specialization'; 'specification' is a different concept. Also, the sentence in Section 3 that says 'Both buyers and sellers will utilize selling and bidding orders in the auction for price negotiations' appears to conflate P2P negotiation with auction behavior and should be rewritten for clarity.
  8. [Branding and affiliation] The first-page affiliation contains a typo: 'Hefei Comprehensive National Science Centerce' should be 'Center'. Please correct this in the camera-ready version.

Circularity Check

2 steps flagged · score 6.0 of 10

Price-rule correlation and role specialization are largely forced by the LLM's observation/profile design; the capability benchmark remains independent.

  1. self definitional [Section 4.2 (Perception) + Appendix B (Structured Actions) + Section 5.3.3 (Supply and Demand Rule)]
    "Auction infoincludes details of the selling and bidding orders with their prices in the auction. ... For executing complex structured actions, operational capabilities are realized by invoking LLMs, such as determining auction pricing and generating P2P chat responses. ... We introduce the "Demand Supply Gap" variable to represent the disparity between bidding orders (demand) and selling orders (supply) in the auction. ... we assess the relationship between the auction price and the demand-supply disparity, obtaining a coefficient of 0.67 with a significance level of p < 0.001."

    The headline evidence for 'price fluctuations in line with market rules' is a correlation between auction price and the demand-supply gap. But the demand-supply gap is exactly the 'selling orders' and 'bidding orders' that the agent's parsed observation (Auction info) feeds to the LLM, and the auction prices themselves are produced by LLM calls (Auction_Buy/Auction_Sell, 'determining auction pricing'). So the LLM is asked to set prices from the very variables later correlated with price; r=0.67 largely measures prompt-following/internal consistency, not an emergent market law. The paper provides no ablation without Auction info and no non-LLM pricing baseline, and the included 'past average deal price' also injects autocorrelation.

  2. self definitional [Section 4.1 / Appendix D.2 (Profiles) + Section 5.2.2 (Role Specialization)]
    "Engaged Grinder You are highly active in the game, focusing on gameplay rather than spending money. ... Spending Enthusiast ... You’re not averse to spending money to advance in the game... (Invest substantial time and money.) ... cost-averse agents like Agents 1 and 5 tend to engage in labor-intensive activities such as task completion and trading. ... agents willing to spend money evolve into Pay-to-Win players (i.e., Agents 2 and 3)... This emergent behavior stems entirely from the agents' diverse characteristics."

    Role specialization is presented as an emergent phenomenon, but the agent profiles explicitly encode the specialization: an 'Engaged Grinder' is told to focus on gameplay rather than spending money, a 'Spending Enthusiast' is told he is not averse to spending. Section 5.2.1 shows GPT-4-rated consistency of decisions with profiles is ~3.8-3.9, i.e., the agents follow the profile text. The observed division into task/trade-heavy and recharge-heavy agents is therefore the profile prompt being obeyed, not a phenomenon that develops bottom-up from homogeneous economic incentives. The paper does not test whether specialization arises from unlabeled agents or from economic interaction alone, so the claimed emergence is by construction rather than demonstrated.

full rationale

The paper's strongest non-circular content is its capability benchmark: MMOAgent vs Random, Rule-based, MMO-economist, ReAct, and Reflexion (Table 1) is an external comparison not derived from the framework's own outputs, and the P2P negotiation and equality-profitability results are not forced by construction. Self-citations to the MMO Economist environment and utility function are standard reuse and are not load-bearing for the paper's new claims; no uniqueness theorem is imported. However, the two headline 'emergent phenomena' in the abstract and conclusion are not independent: the supply-demand price correlation (Section 5.3.3) is measured between variables that are both input to and output of the LLM pricing call (via Auction info and Auction_Buy/Auction_Sell), and role specialization (Section 5.2.2) is a restatement of the assigned profile preferences. Both reduce by construction to the observation/action design, so the central evidence for emergent economic behavior is partially circular. Honest non-finding for the rest.

Assumptions & free parameters 7 free parameters · 4 assumptions · 0 invented entities

The framework inherits the virtual environment and resource taxonomy from MMO Economist, assumes textual profiles elicit human-like LLM choices, and relies on several hand-set memory hyperparameters. No new physical or conceptual entities are introduced beyond the agent architecture itself.

free parameters (7)
  • Memory strength S = 20 (tested 10, 30)
    Exponential forgetting rate in Eq. (6); chosen by hand, swept in Appendix E.
  • Short-term memory size = 10
    Number of recent trajectories retained; swept 2-18 in Appendix E.
  • LTM forgetting threshold = 0.2
    Records with importance score below this threshold are removed from long-term memory.
  • LTM merging similarity threshold = 0.9
    Similar memories accumulate importance scores when similarity exceeds this threshold.
  • Memory writing window m = 5
    Number of preceding trajectories credited for a positive-reward action in Eq. (5).
  • Discount factor gamma = not reported
    Used in Eq. (5) to discount past trajectories; no value is given in the paper.
  • Number of profile clusters k = 5
    k-means produces five player archetypes; chosen for interpretability, not validated against a model-selection criterion.
assumptions (4)
  • domain assumption The MMO Economist environment and its resource/activity taxonomy faithfully represent real MMO economies
    The paper uses the environment from [67] as the simulation substrate without independent validation of its economic fidelity.
  • domain assumption LLM role-playing from textual profiles yields decisions resembling real players
    Section 4.1: assumes GPT-4-generated profile text plus GPT-3.5 reasoning produces human-like behavior; only heuristic consistency ratings support this.
  • domain assumption Numerical embedding similarity (Eq. 3) adequately measures decision-relevant experience
    Memory retrieval relies on max-relative-difference over numeric observation features; no ablation of this retrieval design beyond memory size.
  • standard math Statistical methods (k-means, Pearson correlation, t-test) are valid for the data
    Standard tools; no distributional assumptions discussed for the correlation test.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Empowering Economic Simulation for Massively Multiplayer Online Games through Generative Agent-Based Modeling." pith.science (2026). https://pith.science/paper/RTDRFYLH

@misc{pith2026250604699,
  author       = {Pith},
  title        = {Pith review of: Empowering Economic Simulation for Massively Multiplayer Online Games through Generative Agent-Based Modeling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RTDRFYLH}},
  note         = {Machine review of arXiv:2506.04699}
}
read the original abstract

Within the domain of Massively Multiplayer Online (MMO) economy research, Agent-Based Modeling (ABM) has emerged as a robust tool for analyzing game economics, evolving from rule-based agents to decision-making agents enhanced by reinforcement learning. Nevertheless, existing works encounter significant challenges when attempting to emulate human-like economic activities among agents, particularly regarding agent reliability, sociability, and interpretability. In this study, we take a preliminary step in introducing a novel approach using Large Language Models (LLMs) in MMO economy simulation. Leveraging LLMs' role-playing proficiency, generative capacity, and reasoning aptitude, we design LLM-driven agents with human-like decision-making and adaptability. These agents are equipped with the abilities of role-playing, perception, memory, and reasoning, addressing the aforementioned challenges effectively. Simulation experiments focusing on in-game economic activities demonstrate that LLM-empowered agents can promote emergent phenomena like role specialization and price fluctuations in line with market rules.

Figures

Figures reproduced from arXiv: 2506.04699 by the authors.

Figure 1
Figure 1. Example of the popular MMO game, World of War [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. SBM and ABM simulation in MMO economies. SBM [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. An overview of our expanded virtual environment for MMO economies. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: An overview of our GABM simulation framework for MMO economies. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Case study of decision-making and P2P trading. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: The role specification of agents. (a) Diagram of the price and supply and demand of MAT in the auction. (b) Diagram of the profitability and equality across three scenarios [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Two system-level emergent economic phenomena. [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: The 5-tier rating system for evaluating the consis [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

77 extracted references · 45 canonical work pages

  1. [1]

    Sahar Abdelnabi, Amr Gomaa, Sarath Sivaprasad, Lea Schönherr, and Mario Fritz

    work page

  2. [2]

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Floren- cia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. 2023. Gpt-4 technical report.arXiv preprint arXiv:2303.08774 (2023)

    arXiv 2023

  3. [3]

    Mohammad Ahmadi Achachlouei and Lorenz M Hilty. 2015. System Dynamics vs. agent-based modeling—comparing models and approaches: A literature review and a transformation procedure. (2015)

    work page 2015

  4. [4]

    S Nageeb Ali, Navin Kartik, and Andreas Kleiner. 2023. Sequential veto bargaining with incomplete information.Econometrica91, 4 (2023), 1527–1562

    work page 2023

  5. [5]

    Lisa P Argyle, Ethan C Busby, Nancy Fulda, Joshua R Gubler, Christopher Rytting, and David Wingate. 2023. Out of one, many: Using language models to simulate human samples.Political Analysis31, 3 (2023), 337–351

    2023

  6. [6]

    Jamal Jokar Arsanjani, Marco Helbich, and Eric de Noronha Vaz. 2013. Spatiotem- poral simulation of urban growth patterns using agent-based modeling: The case of Tehran.Cities32 (2013), 33–42

    work page 2013

  7. [7]

    Lee Averell and Andrew Heathcote. 2011. The form of the forgetting curve and the fate of memories.Journal of mathematical psychology55, 1 (2011), 25–35

    work page 2011

  8. [8]

    Robert L Axtell and J Doyne Farmer. 2022. Agent-based modeling in economics and finance: Past, present, and future.Journal of Economic Literature(2022), 1–101

    work page 2022

Show all 77 references
  1. [9]

    Aleksandr Mikhaylovich Batkovskiy, Alina Valerevna Konovalova, Elena Georgievna Semenova, Valeriy Jaroslavovich Trofimets, and Alena Vladimirovna Fomina. 2015. Study of economic systems using the simulation-based statistical modeling method.Mediterranean Journal of Social Scie...

    2015

  2. [10]

    Tanla E Bilir. 2009. Real economics in virtual worlds: A massively multiplayer online game case study: Runescape.A vailable at SSRN 1655084(2009)

    2009

  3. [11]

    Yngvi Bjornsson and Hilmar Finnsson. 2009. Cadiaplayer: A simulation-based general game player.IEEE Transactions on Computational Intelligence and AI in Games1, 1 (2009), 4–15

    2009

  4. [12]

    Sam Devlin, Anastasija Anspoka, Nick Sephton, Peter Cowling, and Jeff Rollason

  5. [13]

    Hermann Ebbinghaus. 2013. Memory: A contribution to experimental psychology. Annals of neurosciences20, 4 (2013), 155

    2013

  6. [14]

    Philipp Eisen. 2017. Simulating human game play for level difficulty estimation with convolutional neural networks

    2017

  7. [15]

    Yao Fu, Hao Peng, Tushar Khot, and Mirella Lapata. 2023. Improving language model negotiation with self-play and in-context learning from ai feedback.arXiv preprint arXiv:2305.10142(2023)

    2023 arXiv

  8. [16]

    David Gale. 1955. The law of supply and demand.Mathematica scandinavica (1955), 155–169. KDD ’25, August 3–7, 2025, Toronto, ON, Canada Bihan Xu et al

    1955

  9. [17]

    Chen Gao, Xiaochong Lan, Zhihong Lu, Jinzhu Mao, Jinghua Piao, Huandong Wang, Depeng Jin, and Yong Li. 2023. 𝑆 3: Social-network Simulation System with Large Language Model-Empowered Agents.arXiv preprint arXiv:2307.14984 (2023)

    2023 arXiv

  10. [18]

    Weibo Gao, Qi Liu, Linan Yue, Fangzhou Yao, Rui Lv, Zheng Zhang, Hao Wang, and Zhenya Huang. 2025. Agent4Edu: Generating Learner Response Data by Generative Agents for Intelligent Education Systems. InProceedings of the AAAI Conference on Artificial Intelligence. 23923–23932

    2025

  11. [19]

    Etan A Green and E Barry Plunkett. 2022. The science of the deal: Optimal bargaining on ebay using deep reinforcement learning. InProceedings of the 23rd ACM Conference on Economics and Computation. 1–27

    2022

  12. [20]

    Stefan Freyr Gudmundsson, Philipp Eisen, Erik Poromaa, Alex Nodet, Sami Pur- monen, Bartlomiej Kozakowski, Richard Meurling, and Lele Cao. 2018. Human- like playtesting with deep learning. In2018 IEEE Conference on Computational Intelligence and Games (CIG). IEEE, 1–8

    2018

  13. [21]

    Veronika Hackl, Alexandra Elena Müller, Michael Granitzer, and Maximilian Sailer. 2023. Is GPT-4 a reliable rater? Evaluating consistency in GPT-4’s text ratings. InFrontiers in Education, Vol. 8. Frontiers Media SA, 1272229

    2023

  14. [22]

    John A Hartigan and Manchek A Wong. 1979. Algorithm AS 136: A k-means clustering algorithm.Journal of the royal statistical society. series c (applied statistics)28, 1 (1979), 100–108

    1979

  15. [23]

    Zhipeng Hu, Yu Ding, Runze Wu, Lincheng Li, Rongsheng Zhang, Yujing Hu, Feng Qiu, Zhimeng Zhang, Kai Wang, Shiwei Zhao, et al. 2023. Deep learning applications in games: a survey from a data perspective.Applied Intelligence53, 24 (2023), 31129–31164

    2023

  16. [24]

    Zhenya Huang, Qi Liu, Yuying Chen, Le Wu, Keli Xiao, Enhong Chen, Haiping Ma, and Guoping Hu. 2020. Learning or forgetting? A dynamic approach for tracking the knowledge proficiency of students.ACM Transactions on Information Systems (TOIS)38, 2 (2020), 1–33

    2020

  17. [25]

    Takahisa Imagawa and Tomoyuki Kaneko. 2015. Enhancements in monte carlo tree search algorithms for biased game trees. In2015 IEEE Conference on Compu- tational Intelligence and Games (CIG). IEEE, 43–50

    2015

  18. [26]

    Di Jin, Shikib Mehri, Devamanyu Hazarika, Aishwarya Padmakumar, Sungjin Lee, Yang Liu, and Mahdi Namazifar. 2023. Data-efficient alignment of large language models with human feedback through natural language.arXiv preprint arXiv:2311.14543(2023)

    2023 arXiv

  19. [27]

    Shi Jinxin, Zhao Jiabao, Wang Yilei, Wu Xingjiao, Li Jiawen, and He Liang. 2023. Cgmi: Configurable general multi-agent interaction framework.arXiv preprint arXiv:2308.12503(2023)

    2023 arXiv

  20. [28]

    Samuel Kaminski. 2006. The impacts of farming and crafting on MMO economies

    2006

  21. [29]

    Guohao Li, Hasan Abed Al Kader Hammoud, Hani Itani, Dmitrii Khizbullin, and Bernard Ghanem. 2023. Camel: Communicative agents for" mind" exploration of large scale language model society. (2023)

    2023

  22. [30]

    Jing Li, Zhaohan Sheng, and Huimin Liu. 2010. Multi-agent simulation for the dominant players’ behavior in supply chains.Simulation Modelling Practice and Theory18, 6 (2010), 850–859

    2010

  23. [31]

    Mingjia Li, Hong Qian, Jinglan Lv, Mengliang He, Wei Zhang, and Aimin Zhou

  24. [32]

    Nian Li, Chen Gao, Yong Li, and Qingmin Liao. 2023. Large language model- empowered agents for simulating macroeconomic activities.arXiv preprint arXiv:2310.10436(2023)

    2023 arXiv

  25. [33]

    Siyu Li, Jin Yang, and Kui Zhao. 2023. Are you in a masquerade? exploring the behavior and impact of large language model driven social bots in online social networks.arXiv preprint arXiv:2307.10337(2023)

    2023 arXiv

  26. [34]

    Tian Liang, Zhiwei He, Wenxiang Jiao, Xing Wang, Yan Wang, Rui Wang, Yujiu Yang, Zhaopeng Tu, and Shuming Shi. 2023. Encouraging divergent thinking in large language models through multi-agent debate.arXiv preprint arXiv:2305.19118(2023)

    2023 arXiv

  27. [35]

    Fu-ren Lin and Shyh-ming Lin. 2006. Enhancing the supply chain performance by integrating simulated and physical agents into organizational information systems.Journal of Artificial Societies and Social Simulation9, 4 (2006)

    2006

  28. [36]

    Fu-ren Lin and You-yu Lin. 2006. Integrating multi-agent negotiation to resolve constraints in fulfilling supply chain orders.Electronic Commerce Research and Applications5, 4 (2006), 313–322

    2006

  29. [37]

    Jiaju Lin, Haoran Zhao, Aochi Zhang, Yiting Wu, Huqiuyue Ping, and Qin Chen

  30. [38]

    Jiayu Liu, Zhenya Huang, Qi Liu, Zhiyuan Ma, Chengxiang Zhai, and Enhong Chen. 2025. Knowledge-Centered Dual-Process Reasoning for Math Word Prob- lems with Large Language Models.IEEE Transactions on Knowledge and Data Engineering37, 6 (2025), 3457–3471

    2025

  31. [39]

    Jiayu Liu, Zhenya Huang, Tong Xiao, Jing Sha, Jinze Wu, Qi Liu, Shijin Wang, and Enhong Chen. 2024. SocraticLM: Exploring socratic personalized teaching with large language models.Advances in Neural Information Processing Systems 37 (2024), 85693–85721

    2024

  32. [40]

    Jie Ma, Zhitao Gao, Qi Chai, Wangchun Sun, Pinghui Wang, Hongbin Pei, Jing Tao, Lingyun Song, Jun Liu, Chen Zhang, et al. 2025. Debate on graph: a flexible and reliable reasoning framework for large language models. InProceedings of the AAAI Conference on Artificial Intelligen...

    2025

  33. [41]

    arXiv preprint arXiv:2308.04026(2023)

    Agentsims: An open-source sandbox for large language model evaluation. arXiv preprint arXiv:2308.04026(2023)

    2023 arXiv

  34. [42]

    Oskar Milik. 2017. Persona in MMO games: Constructing an identity through complex player/character relationships.Persona Studies3, 2 (2017), 66–78

    2017

  35. [43]

    Ben Naismith, Phoebe Mulcaire, and Jill Burstein. 2023. Automated evaluation of written discourse coherence using GPT-4. InProceedings of the 18th Workshop on Innovative Use of NLP for Building Educational Applications (BEA 2023). 394–403

    2023

  36. [44]

    Joon Sung Park, Joseph O’Brien, Carrie Jun Cai, Meredith Ringel Morris, Percy Liang, and Michael S Bernstein. 2023. Generative agents: Interactive simulacra of human behavior. InProceedings of the 36th Annual ACM Symposium on User Interface Software and Technology. 1–22

    2023

  37. [45]

    Aman Madaan, Niket Tandon, Prakhar Gupta, Skyler Hallinan, Luyu Gao, Sarah Wiegreffe, Uri Alon, Nouha Dziri, Shrimai Prabhumoye, Yiming Yang, et al

  38. [46]

    Chen Qian, Xin Cong, Cheng Yang, Weize Chen, Yusheng Su, Juyuan Xu, Zhiyuan Liu, and Maosong Sun. 2023. Communicative agents for software development. arXiv preprint arXiv:2307.07924(2023)

    2023 arXiv

  39. [47]

    Jingqing Ruan, Yihong Chen, Bin Zhang, Zhiwei Xu, Tianpeng Bao, Guoqing Du, Shiwei Shi, Hangyu Mao, Xingyu Zeng, and Rui Zhao. 2023. Tptu: Task planning and tool usage of large language model-based ai agents.arXiv preprint arXiv:2308.03427(2023)

    2023

  40. [48]

    Florian Rupp and Kai Eckert. 2024. GEEvo: Game Economy Generation and Balancing with Evolutionary Algorithms.arXiv preprint arXiv:2404.18574(2024)

    2024 arXiv

  41. [49]

    2021.Game balance

    Ian Schreiber and Brenda Romero. 2021.Game balance. CRC Press

    2021

  42. [50]

    Siyuan Qi, Shuo Chen, Yexin Li, Xiangyu Kong, Junqi Wang, Bangcheng Yang, Pring Wong, Yifan Zhong, Xiaoyuan Zhang, Zhaowei Zhang, et al. 2024. CivRealm: A Learning and Reasoning Odyssey in Civilization for Decision- Making Agents.arXiv preprint arXiv:2401.10568(2024)

    2024 arXiv

  43. [51]

    Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. 2024. Reflexion: Language agents with verbal reinforcement learning. Advances in Neural Information Processing Systems36 (2024)

    2024

  44. [52]

    Conor Stephens and Chris Exton. 2021. Measuring Inflation within Virtual Economies using Deep Reinforcement Learning.. InICAART (2). 444–453

    2021

  45. [53]

    Guanzhi Wang, Yuqi Xie, Yunfan Jiang, Ajay Mandlekar, Chaowei Xiao, Yuke Zhu, Linxi Fan, and Anima Anandkumar. 2023. Voyager: An open-ended embodied agent with large language models.arXiv preprint arXiv:2305.16291(2023)

    2023 arXiv

  46. [54]

    Lei Wang, Chen Ma, Xueyang Feng, Zeyu Zhang, Hao Yang, Jingsen Zhang, Zhiyuan Chen, Jiakai Tang, Xu Chen, Yankai Lin, et al. 2024. A survey on large language model based autonomous agents.Frontiers of Computer Science18, 6 (2024), 1–26

    2024

  47. [55]

    Yongliang Shen, Kaitao Song, Xu Tan, Dongsheng Li, Weiming Lu, and Yueting Zhuang. 2024. Hugginggpt: Solving ai tasks with chatgpt and its friends in hugging face.Advances in Neural Information Processing Systems36 (2024)

    2024

  48. [56]

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al. 2022. Chain-of-thought prompting elicits reasoning in large language models.Advances in neural information processing systems35 (2022), 24824–24837

    2022

  49. [57]

    2011.The economic and social impact of MMO virtual markets and identities

    Christopher Wilkinson. 2011.The economic and social impact of MMO virtual markets and identities. University of Massachusetts Lowell

    2011

  50. [58]

    Shuang Wu, Liwen Zhu, Tao Yang, Shiwei Xu, Qiang Fu, Yang Wei, and Haobo Fu. 2024. Enhance Reasoning for Large Language Models in the Game Werewolf. arXiv preprint arXiv:2402.02330(2024)

    2024 arXiv

  51. [59]

    Zhiheng Xi, Wenxiang Chen, Xin Guo, Wei He, Yiwen Ding, Boyang Hong, Ming Zhang, Junzhe Wang, Senjie Jin, Enyu Zhou, et al. 2023. The rise and potential of large language model based agents: A survey.arXiv preprint arXiv:2309.07864 (2023)

    2023 arXiv

  52. [60]

    Lei Wang, Jingsen Zhang, Hao Yang, Zhiyuan Chen, Jiakai Tang, Zeyu Zhang, Xu Chen, Yankai Lin, Ruihua Song, Wayne Xin Zhao, et al. 2024. User behavior sim- ulation with large language model based agents.arXiv preprint arXiv:2306.02552 (2024)

    2024 arXiv

  53. [61]

    Yuzhuang Xu, Shuo Wang, Peng Li, Fuwen Luo, Xiaolong Wang, Weidong Liu, and Yang Liu. 2023. Exploring large language models for communication games: An empirical study on werewolf.arXiv preprint arXiv:2309.04658(2023)

    2023 arXiv

  54. [62]

    Shangzi Xue, Zhenya Huang, Jiayu Liu, Xin Lin, Yuting Ning, Binbin Jin, Xin Li, and Qi Liu. 2024. Decompose, analyze and rethink: Solving intricate problems with human-like reasoning cycle.Advances in Neural Information Processing Systems37 (2024), 357–385

    2024

  55. [63]

    Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2022. React: Synergizing reasoning and acting in language models. arXiv preprint arXiv:2210.03629(2022)

    2022 arXiv

  56. [64]

    Xinli Yu, Zheng Chen, Yuan Ling, Shujing Dong, Zongyi Liu, and Yanbin Lu. 2023. Temporal Data Meets LLM–Explainable Financial Time Series Forecasting.arXiv preprint arXiv:2306.11025(2023). LLM-based Agents for Simulating Game Economics KDD ’25, August 3–7, 2025, Toronto, ON, Canada

    2023 arXiv

  57. [65]

    Bihan Xu, Zhenya Huang, Jiayu Liu, Shuanghong Shen, Qi Liu, Enhong Chen, Jinze Wu, and Shijin Wang. 2023. Learning behavior-oriented knowledge tracing. InProceedings of the 29th ACM SIGKDD conference on knowledge discovery and data mining. 2789–2800

    2023

  58. [66]

    Hongke Zhao, Likang Wu, Yuqing Shan, Zonghan Jin, Yuanpei Sui, Zipeng Liu, Nan Feng, Minqiang Li, and Wei Zhang. 2024. A comprehensive survey of large language models in management: Applications, challenges, and opportunities. Challenges, and Opportunities (August 14, 2024)(2024)

    2024

  59. [67]

    Shiwei Zhao, Xi Yuan, Runze Wu, Zhipeng Hu, Haoyu Liu, Kai Wang, Yujing Hu, Tangjie Lv, Changjie Fan, Xin Tong, et al. 2024. The MMO Economist: AI Empowers Robust, Healthy, and Sustainable P2W MMO Economies. InCompanion Proceedings of the ACM on Web Conference 2024. 443–452

    2024

  60. [68]

    Denny Zhou, Nathanael Schärli, Le Hou, Jason Wei, Nathan Scales, Xuezhi Wang, Dale Schuurmans, Claire Cui, Olivier Bousquet, Quoc Le, et al. 2022. Least-to- most prompting enables complex reasoning in large language models.arXiv preprint arXiv:2205.10625(2022)

    2022 arXiv

  61. [69]

    Xizhou Zhu, Yuntao Chen, Hao Tian, Chenxin Tao, Weijie Su, Chenyu Yang, Gao Huang, Bin Li, Lewei Lu, Xiaogang Wang, et al. 2023. Ghost in the minecraft: Generally capable agents for open-world enviroments via large language models with text-based knowledge and memory.arXiv pre...

    2023 arXiv

  62. [70]

    Zhiyuan Zeng, Jiatong Yu, Tianyu Gao, Yu Meng, Tanya Goyal, and Danqi Chen

  63. [71]

    Evaluating large language models at evaluating instruction following.arXiv preprint arXiv:2310.07641(2023)

    2023 arXiv

  64. [76]

    Rule" indicates applying actions per predefined human regulations. The

    Alexander Zook, Brent Harrison, and Mark O Riedl. 2019. Monte-carlo tree search for simulation-based strategy analysis.arXiv preprint arXiv:1908.01423 (2019). A Economic Resources and Activities Six kinds of economic resources form the material foundation and the carrying enti...

    2019 arXiv

  65. [77]

    When STM size > 10, impact on agent performance is limited

    Results show an overall saturation trend. When STM size > 10, impact on agent performance is limited. In Scarce scenario, capabil- ity even drops slightly. This may be because the game environment changes in real time. The current state and a few recent decisions let the agent...

  66. [2016]

    InProceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, Vol

    Combining gameplay data with monte carlo tree search to emulate human play. InProceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, Vol. 12. 16–22

  67. [2023]

    Llm-deliberation: Evaluating llms with interactive multi-agent negotiation games.arXiv preprint arXiv:2309.17234(2023)

    2023 arXiv

  68. [2024]

    Self-refine: Iterative refinement with self-feedback.Advances in Neural Information Processing Systems36 (2024)

    2024

  69. [2025]

    Foundation model enhanced derivative-free cognitive diagnosis.Frontiers of Computer Science19, 1 (2025), 191318

    2025

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.