pith. sign in

arxiv: 2606.31038 · v1 · pith:6EEOQZGXnew · submitted 2026-06-30 · 💻 cs.GR · cs.AI

LLM-Driven Personalities for Decision Making in Emergency Simulations

Stefano Calzolari , Rubens Montanha , Gabriel Schneider , Gustavo Wide , Paulo Knob , Francesco Strada , Andrea Bottino , Soraia Raupp Musse This is my paper

Pith reviewed 2026-07-01 03:15 UTC · model grok-4.3

classification 💻 cs.GR cs.AI
keywords LLMOCEAN personality traitsevacuation simulationvirtual agentsdecision makingcrowd simulationagent behavior
0
0 comments X

The pith

Encoding OCEAN personality traits in LLM prompts produces distinct decision patterns among virtual agents in evacuation simulations.

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

The paper tests whether large language models can drive believable decision-making in virtual humans by injecting OCEAN personality traits through natural language prompts. It runs these agents in a simulated evacuation and measures how traits alter choices such as route selection or group behavior. If the approach works, virtual crowds could gain realistic variation without programmers defining every possible response in advance. Sympathetic readers would see this as a step toward more flexible simulations for training or entertainment.

Core claim

LLM-driven personality profiles significantly impact agents' decisions, leading to distinct behavioral patterns across different traits. These findings suggest that heterogeneous crowds composed of LLM-guided agents can enhance the realism and variability of simulated environments, offering a flexible alternative to traditional rule-based approaches.

What carries the argument

OCEAN personality traits encoded as language prompts that steer LLM outputs for agent actions in the simulation.

If this is right

  • Agents exhibit trait-specific behaviors in emergency decisions.
  • Heterogeneous personality compositions increase overall simulation variability.
  • LLM-based guidance serves as an alternative to predefined rules for agent intelligence.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Similar prompting could introduce personality variation into agents in non-emergency scenarios such as social interactions.
  • Testing whether the same traits produce consistent patterns across different base LLMs would clarify the robustness of the method.
  • Real-world validation might involve matching simulated behaviors to observed human responses under personality assessments.

Load-bearing premise

Behavioral differences arise specifically from the personality trait encodings in the prompts and not from other uncontrolled factors in the LLM or environment.

What would settle it

Running identical simulations with personality prompts removed or scrambled and finding no reduction in behavioral distinctions would falsify the claim that the profiles drive the patterns.

Figures

Figures reproduced from arXiv: 2606.31038 by Andrea Bottino, Francesco Strada, Gabriel Schneider, Gustavo Wide, Paulo Knob, Rubens Montanha, Soraia Raupp Musse, Stefano Calzolari.

Figure 1
Figure 1. Figure 1: Overview of our method. The agent comprises three elements: Perception, Reasoning, and [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Layout of our test scene. The green arrows represent the evacuation exits, while the blue ones represent the route the agents should follow to evacuate the building. to Action. The system was implemented in Unity 3D1 engine and connected with our Decision￾Making pipeline through ZeroMQ2 with a request-response architecture. In the simulation, anytime an agent is prompted with a Call for Ac￾1 https://unity.… view at source ↗
Figure 3
Figure 3. Figure 3: Number of agents evacuated after each alert message, simulated with 600 agents. Most of [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Cumulative agents’ decision state over the alert steps Call for Actions by personality. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Final alert Call for Actions response ratios per personality. All Neutral and Conscientious [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Example of three possible agent behaviors during evacuation. From left to right, the agents [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Rescue decision rate by helping agent personality, being the helped agent always a neurotic [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
read the original abstract

For virtual humans to appear believable, they must exhibit agency and spatial awareness while interacting with their environment in ways that reflect competence and intelligence. At the core of these capabilities lies effective decision-making, which strongly shapes agent behavior. With the rapid advancement of artificial intelligence, Large Language Models (LLMs) have increasingly been explored as a mechanism to support such decision-making processes. In this work, we investigate the use of LLMs to drive decision-making in virtual humans within a simulated evacuation scenario, incorporating OCEAN personality traits into agent representations. Our goal is to evaluate how personality, expressed through language-based prompts, influences both individual behaviors and collective simulation outcomes. Our results demonstrate that LLM-driven personality profiles significantly impact agents' decisions, leading to distinct behavioral patterns across different traits. These findings suggest that heterogeneous crowds composed of LLM-guided agents can enhance the realism and variability of simulated environments, offering a flexible alternative to traditional rule-based approaches.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript investigates the integration of OCEAN personality traits into LLM-driven decision-making for virtual agents in an evacuation simulation. It claims that personality prompts produce significantly different individual behaviors and collective outcomes compared to non-personality baselines, providing a flexible alternative to rule-based crowd simulation methods.

Significance. If the central claim were supported by quantitative evidence isolating personality effects, the work could contribute to more variable and realistic agent behaviors in graphics and simulation applications. The approach addresses a relevant gap in believable virtual humans, but the absence of metrics, controls, and statistical validation prevents assessment of its actual significance or reproducibility.

major comments (2)
  1. [Abstract] Abstract: The claim that 'results demonstrate that LLM-driven personality profiles significantly impact agents' decisions, leading to distinct behavioral patterns across different traits' is unsupported by any reported quantitative metrics, statistical tests, baseline comparisons, ablation studies, or controls for prompt sensitivity and LLM stochasticity. This directly undermines the attribution of behavioral differences to OCEAN traits rather than base model variance or encoding details.
  2. [Abstract] Abstract/Results (inferred): No description of experimental controls such as multiple independent runs with fixed seeds per trait, comparison to neutral prompts, or variation in LLM parameters (e.g., temperature) is provided. Without these, the weakest assumption—that observed differences are caused by the personality prompts—cannot be evaluated, rendering the data-to-claim link unevaluable.
minor comments (1)
  1. [Abstract] The abstract would benefit from specifying the exact LLM model, simulation environment details, number of agents, and evacuation scenario parameters to allow readers to contextualize the approach.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thorough review and constructive criticism. We agree that the current manuscript's claims require stronger quantitative support and explicit experimental controls to be rigorously evaluated. We will revise the paper to incorporate these elements.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The claim that 'results demonstrate that LLM-driven personality profiles significantly impact agents' decisions, leading to distinct behavioral patterns across different traits' is unsupported by any reported quantitative metrics, statistical tests, baseline comparisons, ablation studies, or controls for prompt sensitivity and LLM stochasticity. This directly undermines the attribution of behavioral differences to OCEAN traits rather than base model variance or encoding details.

    Authors: We agree that the abstract's claim is not supported by quantitative evidence in the current manuscript, which presents primarily descriptive and qualitative observations of agent behaviors. In the revision, we will add quantitative metrics (e.g., evacuation completion times, decision counts per trait, path deviation measures), statistical comparisons (ANOVA or Kruskal-Wallis tests across traits with post-hoc analysis), baseline runs using neutral prompts and non-LLM agents, ablation on prompt phrasing, and multiple runs with fixed seeds to quantify variance. These changes will enable proper attribution of effects to the OCEAN prompts. revision: yes

  2. Referee: [Abstract] Abstract/Results (inferred): No description of experimental controls such as multiple independent runs with fixed seeds per trait, comparison to neutral prompts, or variation in LLM parameters (e.g., temperature) is provided. Without these, the weakest assumption—that observed differences are caused by the personality prompts—cannot be evaluated, rendering the data-to-claim link unevaluable.

    Authors: We concur that the manuscript does not describe these controls. The revision will include a new Experimental Design subsection specifying: (i) 10+ independent runs per trait with fixed random seeds, (ii) direct comparisons to neutral-prompt and rule-based baselines, and (iii) sensitivity sweeps over temperature (0.0–1.0) and top-p values, with results reported as means and standard deviations. This will allow readers to assess whether differences are attributable to the personality prompts. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical simulation study with no derivations or fitted predictions

full rationale

The paper describes an empirical investigation of LLM agents in evacuation simulations using OCEAN personality prompts. No equations, parameters, or derivations are present in the provided text. The central claim rests on observed behavioral differences in simulation runs rather than any self-referential construction, self-citation chain, or renaming of known results. No load-bearing steps reduce to inputs by definition. This is a standard experimental setup whose validity can be assessed via replication and controls, independent of any circularity analysis.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Only the abstract is available, so the ledger is necessarily incomplete. The central claim implicitly rests on the domain assumption that LLM outputs can be treated as faithful proxies for human personality-driven decisions.

axioms (1)
  • domain assumption LLM responses to personality prompts produce stable, interpretable differences in agent decision-making that reflect the intended OCEAN traits.
    This premise is required for the observed behavioral differences to be attributed to personality rather than to LLM stochasticity or prompt artifacts.

pith-pipeline@v0.9.1-grok · 5706 in / 1143 out tokens · 41920 ms · 2026-07-01T03:15:24.587852+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

18 extracted references

  1. [1]

    Virtual humans for anima- tion, ergonomics, and simulation

    Norman Badler. Virtual humans for anima- tion, ergonomics, and simulation. InPro- ceedings IEEE Nonrigid and Articulated Motion Workshop, pages 28–36. IEEE, 1997

    1997

  2. [2]

    Developing a scale for mea- suring the believability of virtual agents

    Siqi Guo, Nicoletta Adamo, and Chris- tos Mousas. Developing a scale for mea- suring the believability of virtual agents. InInternational Conference on Artificial Reality and Telexistence & Eurograph- ics Symposium on Virtual Environments (ICAT-EGVE). Eurographics Digital Li- brary, 2023

    2023

  3. [3]

    Crowd simulation incorporating agent psycholog- ical models, roles and communication

    Nuria Pelechano, Kevin O’brien, Barry Silverman, and Norman Badler. Crowd simulation incorporating agent psycholog- ical models, roles and communication. 2005

    2005

  4. [4]

    Embodied agent interface: Benchmarking llms for embodied decision making.Advances in Neural Information Processing Systems, 37:100428–100534, 2024

    Manling Li, Shiyu Zhao, Qineng Wang, Kangrui Wang, Yu Zhou, Sanjana Sri- vastava, Cem Gokmen, Tony Lee, Li E Li, Ruohan Zhang, et al. Embodied agent interface: Benchmarking llms for embodied decision making.Advances in Neural Information Processing Systems, 37:100428–100534, 2024

    2024

  5. [5]

    Towards human-like virtual beings: Sim- ulating human behavior in 3d scenes

    Chen Liang, Wenguan Wang, and Yi Yang. Towards human-like virtual beings: Sim- ulating human behavior in 3d scenes. In Proceedings of the IEEE/CVF Interna- tional Conference on Computer Vision, pages 10753–10763, 2025

    2025

  6. [6]

    de- scription of personality

    Lewis R Goldberg. An alternative “de- scription of personality”: The big-five fac- tor structure. InPersonality and person- ality disorders, pages 34–47. Routledge, 2013

    2013

  7. [7]

    Smart-dream: To condi- tion or not to condition; a study on the im- pact of llm conditioning on motivational interview dialog virtual agent

    Lucie Galland, Catherine Pelachaud, and Florian Pecune. Smart-dream: To condi- tion or not to condition; a study on the im- pact of llm conditioning on motivational interview dialog virtual agent. InProceed- ings of the 25th ACM International Con- ference on Intelligent Virtual Agents, pages 1–9, 2025

    2025

  8. [8]

    Bin Han, Deuksin Kwon, Spencer Lin, Kaleen Shrestha, and Jonathan Gratch. Can llms generate behaviors for embodied virtual agents based on personality traits? InProceedings of the 25th ACM Inter- national Conference on Intelligent Virtual Agents, pages 1–10, 2025

    2025

  9. [9]

    How does a virtual agent decide where to look? symbolic cognitive reasoning for embod- ied head rotation

    Juyeong Hwang, Seong-Eun Hong, JaeY- oung Seon, and HyeongYeop Kang. How does a virtual agent decide where to look? symbolic cognitive reasoning for embod- ied head rotation. InProceedings of the SIGGRAPH Asia 2025 Conference Papers, pages 1–12, 2025

    2025

  10. [10]

    Modelling and interpret- ing pre-evacuation decision-making using machine learning.Automation in Con- struction, 113:103140, 2020

    Xilei Zhao, Ruggiero Lovreglio, and Daniel Nilsson. Modelling and interpret- ing pre-evacuation decision-making using machine learning.Automation in Con- struction, 113:103140, 2020

    2020

  11. [11]

    Phan- tom: Persona-based prompting has an ef- fect on theory-of-mind reasoning in large language models

    Gerard Yeo, Fiona Tan An Ting, Kokil Jaidka, Shaz Furniturewala, Wu Fanyou, Weijie Xu, Vinija Jain, Aman Chadha, Yang Liu, and See Kiong Ng. Phan- tom: Persona-based prompting has an ef- fect on theory-of-mind reasoning in large language models. InProceedings of the In- ternational AAAI Conference on Web and Social Media, volume 19, pages 2124– 2142, 2025

    2025

  12. [12]

    A psychome- tric framework for evaluating and shaping personality traits in large language models

    Gregory Serapio-Garc ´ıa, Mustafa Safdari, Cl´ement Crepy, Luning Sun, Stephen Fitz, Peter Romero, Marwa Abdulhai, Aleksan- dra Faust, and Maja Matari´c. A psychome- tric framework for evaluating and shaping personality traits in large language models. Nature Machine Intelligence, pages 1–15, 2025

    2025

  13. [13]

    Large-language-model- driven agents for fire evacuation simula- tion in a cellular automata environment

    Pei Dang, Jun Zhu, Weilian Li, Yakun Xie, and Heng Zhang. Large-language-model- driven agents for fire evacuation simula- tion in a cellular automata environment. Safety Science, 191:106935, 2025

    2025

  14. [14]

    When agents learn to think: Large language model-enhanced agent-based modeling for crowd evacuation in disaster scenarios

    Sen Yang, Luis Ceferino, Yi Zhang, Chen Gu, Tong Guo, and Gen Kondo. When agents learn to think: Large language model-enhanced agent-based modeling for crowd evacuation in disaster scenarios. Reliability Engineering & System Safety, page 112056, 2025

    2025

  15. [15]

    IOS Press, 2008

    Christian Becker-Asano.WASABI: Affect simulation for agents with believable inter- activity, volume 319. IOS Press, 2008

    2008

  16. [16]

    [Accessed 27-03-2026]

    LangChain overview - Docs by LangChain — docs.langchain.com.https: //docs.langchain.com/oss/ python/langchain/overview. [Accessed 27-03-2026]

    2026

  17. [17]

    https://developers.openai

    OpenAI Harmony Response For- mat — developers.openai.com. https://developers.openai. com/cookbook/articles/ openai-harmony. [Accessed 27- 03-2026]

    2026

  18. [18]

    Simulating crowds based on a space colonization algorithm.Computers & Graphics, 36(2):70–79, 2012

    Alessandro de Lima Bicho, Rafael Ara ´ujo Rodrigues, Soraia Raupp Musse, Cl´audio Rosito Jung, Marcelo Par- avisi, and L´eo Pini Magalh˜aes. Simulating crowds based on a space colonization algorithm.Computers & Graphics, 36(2):70–79, 2012. [19]https://openai.com/safety/. [Accessed 27-03-2026]

    2012