ESSA@work is a session that is open to all interested modellers at any career stage. The only (and key!) requirement is to have a work-in-progress model on which participants would like expert and peer feedback. Models can be at any stage of development. Those in the early stages of development (e.g., at the design stage) should still have a clear description of the purpose, scope, structure and logic of the model. It is intended to be less ‘rigourous’ than a traditional conference-based paper presentation, and accepts work that is still rough.

The intention is to provide a friendly venue for participants that may not have access to experts/modellers, and do not know whom to turn to. Participants will present a work-in-progress, and experts (usually two) will analyze the key aspects of the model, ask probing questions, and offer advice on things to try, or traps to avoid.

The modelling experts that provided feedback on the proposed projects are: Cristina Chueca Del Cerro (Independent Researcher), more TBC.

LLM4ABM is a Special Interest Group (SIG) of ESSA focusing on discussing topics related to GenAI-augmented ABM. During our discussions around the question “Who We Are and Where We Want to Go”, we concluded that this topic should be explored more broadly within the social simulation community, as this emerging transformative approach is likely to influence how research is conducted in the future.

The panel will examine the advantages, disadvantages, and risks of incorporating GenAI as a new tool within the social simulation research toolkit, with regards to scientific integrity, including scientific rigour, transparency, reproducibility, accountability, and adherence to ethical standards, considering the views of the individual panel members as well as the audience.

Before the panel takes place, several questions are defined to structure the discussion, and each panel member is asked to prepare a five-minute presentation to express their view on the question they are given. During the panel, each panel member presents their view on the question given to them, before passing it to the other panel members to add their views. The question is then passed on to the audience to express their thoughts and discuss with the panel members,
before moving on to the next question. After the panel, we will ask each panel member to write down their individual conclusions from the discussion.

AGENDA

1. Introduction

What is GenAI-augmented ABM and how does it differ from traditional ABM? Clarification and definition of key terminology?

2. Potential questions to discuss (perhaps only 4 out of the 6 listed):

• Q1: To what extent are LLMs genuinely useful for social simulation?

• Q2: What can social simulation learn from how GenAI is validated and governed in other disciplines??

• Q3: What barriers currently prevent GenAI-augmented ABM from achieving broader or universal acceptance?

• Q4: What standards, safeguards, and oversight mechanisms are needed to ensure that GenAI-augmented ABM is scientifically rigorous, transparent, reproducible, and ethically sound?

• Q5: As GenAI becomes embedded across the full ABM lifecycle, do we risk losing the research skills and critical judgement that underpin good science?

• Q6: Who should govern the development of GenAI-augmented ABM and the use of GenAI across the full ABM lifecycle, and through what mechanisms

We plan to publish the outcomes of this panel discussion as a paper, capturing not only the views of the panel members but also the key insights emerging from the audience discussion. To ensure accuracy, the session will be recorded with the consent of all participants.

Acknowledgement: The questions have been derived through discussions with Claude Sonnet 4.6

The “Agent-Verse” — the vast ecosystem of Agent-Based Simulation (ABS) and multi-agent systems — has expanded from its 1970s origins into a truly “broad church” spanning the natural sciences, economics, policy, and the humanities. However, this breadth often leads to silos, where practitioners in different fields remain unaware of the commonalities and innovations happening right next door.

This interactive workshop invites participants to step back and explore the scope of the agent-verse through three distinct lenses: scholarly influence (highly cited researchers), academic discourse (published literature), and industrial application (government and private institutions). Participants will explore why there is surprisingly low overlap between these spheres and what that means for the future of the field through a mix of guided presentations and online survey tools.

We are particularly interested in a wide variety of participants, whether you are a social scientist, an engineer, or a humanities scholar. The workshop format will be adaptive depending on the number of participants. If the number of participants is low, then the workshop will focus on understanding everyone’s agent-based modeling story. If there are many participants, the workshop will focus on collecting sound bites using online survey tools like Slido, which the workshop team has used successfully in the past, which the participants will be able to access using their mobile phones.

This workshop builds on the research paper, entitled “Exploring the Agent-verse: Shedding Light on Three Spheres of Agentic Activity,” which was presented at the Compuational Social Science and Simulation Association (CSSSA) 2025. The intention is to foster further collaboration between European and American societies.

AGENDA

15.15 – 15.20 Introduction to Session and overview of format

15.20 – 15.25 Presentation on Scholarly Influence

15.25 – 15.30 How to use the Slido tool with tool example

15.30 – 15.40 Question Session (three questions)

15.40 – 15.45 Reflect on Answer

15.45 – 15.55 Presentation on academic discourse

15.55 – 16.05 Question Session

16.05 – 16.10 Reflect on Answer

16.10 – 16.15 Break

16.15 – 16.20 Presentation on industrial application

16.20 – 16.30 Question Session

16.30 – 16.40 Reflect on Answer

Presenter: Christophe Le Page, 

Title: Participatory Agent-Based Modelling and Simulation: Engaging Local Communities to Support Social Learning in Socio-Ecological Systems

What lies behind the illusion of the crystal ball? Drawing on nearly three decades of experience, this talk argues that modelling is not about predicting the future, but about making it discussable. Inspired by Joshua Epstein’s view that models serve many purposes beyond prediction (explaining, questioning, and structuring dialogue), it presents a participatory approach to agent-based modelling and simulation grounded in the KILT posture (Keep It a Learning Tool). Here, the primary “end-users” are not policy-makers but local communities often with limited formal education. Models are deliberately simple and stylized, co-designed to make complex socio-ecological dynamics tangible and open to discussion. Through interactive simulations and role-playing, they become shared artefacts that foster dialogue, mutual understanding, and collective exploration of possible futures.

 Recent advances in artificial intelligence (AI) have intensified discussions on how AI and large language models can be used across domains. Evidence suggests the most promising approach is not replacing human intelligence, but fostering human-AI synergies, where AI supports humans and humans compensate for AI’s limitations. These synergies are, for instance, particularly effective in generative tasks, where AI produces alternatives and humans evaluate them.

Leveraging such synergies is especially relevant in crisis environments, which are volatile, information-rich, and coordination-intensive. Rapidly evolving events hinder information detection and response. Information overload, along with misinformation and disinformation cascades, further complicates identifying reliable data. At the same time, changing conditions challenge collective sensemaking and coordination. AI can support crisis response by filtering large volumes of information and providing timely and tailored insights.

However, how to effectively use these capabilities remains underexplored, as most research focuses on stable settings. Risks such as Ai-induced bias, knowledge homogenization, and accountability also persist. Agent-based modelling and simulation offer useful tools to study information flows, decision-making, and coordination in human-AI systems.

This workshop examines these challenges and explores how modelling and simulation can address them and it is delivered by the ESSA Special Interest Group (SIG) BRICSS – Building Resilience with Social Simulations. It includes a talk by Professor Tina Comes (TU Delft, The Netherlands) and interactive discussions on agent-based approaches. The goal is to advance research and foster exchange on human-AI collaboration in crisis management.

AGENDA

1) 5 minutes: Introduction of the ESSA-SIG BRICSS & of the speaker (Tina Comes)

2) 30 minutes: Presentation by the speaker on current trends in crisis management, opportunities and challenges associated with AI, and open questions that may be addressed through agent-based modelling and simulation.

3) 10 minutes: Selection of research questions to be discussed in breakout rooms, with the number of topics and rooms depending on participant numbers.

4) 30 minutes: Breakout room discussions on how to address the selected research questions using agent-based modelling (one breakout room per challenge). Outcome: initial ideas on more refined research questions and drafts of research designs to address them.

5) 15 minutes: Plenary session to discuss the outcomes of the breakout room discussions & closing remarks.

A wider discussion within the ESSA community and some of ESSA’s Management Committee members. Bring us your questions and thoughts.

The transition from ABM results to a relatable narrative concerning what the modelling is about, is a “dark art” in the sense that: (a) there is little or no published method for this, (b) this process is not transparent (e.g. the simplifications or assumptions involved are not documented) and (c) there are no criteria for what narratives might be justified in any particular context. And yet, the stories that are told about models can be far more influential than the models and their results that they originated from.

The Science Communication and Qual2Rule ESSA Special Interest Groups (SIGs) have thus come together to organise a discussion at SimSocFest to discuss this transition. Whether a full-blown science of “narratisation” (like that of visualisation) is achievable, we would like to start to map out the main problems and issues, to bring to attention any methods and good examples out there and to encourage a collective focus on this.

AGENDA

After topic selection  

15.15 – 15.45 Panel discussion between 3 speakers + moderator

15.45 – 16.15 Discussion in the breakout rooms

16.15 – 16.45 Summary from the breakout rooms’ discussions