All Futures at Once: Supporting Speculative Design for Placemaking with Multi-Agent Social Simulation

Paper Title

All Futures at Once: Supporting Speculative Design for Placemaking with Multi-Agent Social Simulation

Publication Info

• Topic area: Speculative design for placemaking using multi-agent social simulation.
• Keywords: Placemaking, speculative design, multi-agent systems, large language models, social simulation, dynamic speculative objects, diversity, evolvability, design support tools.

Background and Problem

• Problem / challenge: Existing speculative design approaches for placemaking often produce static representations that emphasize spatial form over the evolving dynamics of human activity. They fail to capture diversity and evolvability in how communities interact with spaces over time.
• Significance: Understanding the temporal and social dynamics of placemaking is crucial for creating inclusive and adaptive spaces that meet the needs of diverse stakeholders.
• Motivation and related work: Prior research has explored speculative objects and generative AI for envisioning future scenarios, but these approaches lack the ability to simulate dynamic, evolving human activities. This paper addresses this gap by proposing a system that models and visualizes such dynamics.

Solution

• Proposed approach: ParaScape, a design support system that generates dynamic speculative objects using a multi-agent social simulation framework powered by large language models (LLMs).
• Novelty:
  1. Introduction of dynamic speculative objects, which represent evolving sequences of scenarios depicting diverse activities over time.
  2. Development of an LLM-based multi-agent social simulation framework that models group-specific preferences and sensitivities to generate believable, diverse, and evolvable activities.
  3. Integration of measures for diversity and evolvability to guide designers in analyzing and reflecting on placemaking proposals.
  4. Implementation of an Inspect-then-Discuss mechanism to foster critical reflection on trade-offs among stakeholder needs.
• Procedure and key techniques:
  1. Curated a dataset of place preferences and sensitivities for nine user groups across six dimensions (e.g., accessibility, quietness) and three time periods.
  2. Developed a multi-agent simulation framework with modules for initialization, planning, action, place evaluation, memory, dynamic updating, and visualization.
  3. Generated dynamic speculative objects as image sequences with textual descriptions, enabling designers to analyze and reflect on evolving scenarios.
  4. Evaluated the system through experiments, a user study, and case studies.

Results

• Concrete findings:
  • Dataset validation showed that the simulation framework improved believability across behavioral consistency (48.8%), emotional consistency (56.2%), and emotional appropriateness (65.4%).
  • Ablation study demonstrated that the full framework outperformed reduced versions in diversity (scores >4 vs. <3.3) and evolvability (scores >4 vs. <3.5).
  • User study participants rated the system highly for usability (median = 4–5) and its ability to enhance speculative thinking (median = 4 across five dimensions).
• Advantage over baselines:
  • The full framework achieved significantly higher scores in diversity and evolvability compared to ablation versions.
  • Dynamic speculative objects enabled richer and more nuanced reflections compared to static speculative methods.
• Experiments / evaluation:
  • Dataset validation with 18 evaluators compared believability under dataset and non-dataset conditions.
  • Ablation study with 29 evaluators assessed the impact of key modules on diversity and evolvability.
  • User study with 25 participants evaluated usability, speculative thinking, and system features.
  • Two case studies illustrated how designers used the system to reflect on temporal and spatial dynamics in placemaking.
• Limitations and future work:
  • Dataset coverage is limited to specific groups and time periods; future work will expand group categories and sample sizes.
  • Lack of long-term deployment and comparative studies with similar tools.
  • Future plans include integrating real-time public feedback, amplifying conflicts for anti-solutionist strategies, and applying the approach to other socio-technical contexts.

Summary

This paper introduces ParaScape, a design support system that uses LLM-based multi-agent social simulation to generate dynamic speculative objects for placemaking. The system enables designers to explore evolving scenarios, analyze diversity and evolvability, and reflect on trade-offs among stakeholder needs. Experimental results demonstrate the framework’s ability to produce believable, diverse, and evolving activities, while user studies and case studies highlight its effectiveness in fostering critical reflection and inclusive design. Future work will expand dataset coverage, explore real-world deployment, and refine the system for broader applications in socio-technical design.