Large language model (LLM)-driven multi-agent systems (MAS) are transforming how humans and AIs collaboratively generate ideas and artifacts. While existing surveys provide comprehensive overviews of MAS infrastructures, they largely overlook the dimension of creativity, including how novel outputs are generated and evaluated, how creativity informs agent personas, and how creative workflows are coordinated.
This survey offers a structured framework and roadmap for advancing the development, evaluation, and standardization of creative MAS.
@ARTICLE{10531702,
author={Zhou, Xingcheng and Liu, Mingyu and Yurtsever, Ekim and Zagar, Bare Luka and Zimmer, Walter and Cao, Hu and Knoll, Alois C.},
journal={IEEE Transactions on Intelligent Vehicles},
title={Vision Language Models in Autonomous Driving: A Survey and Outlook},
year={2024},
pages={1-20},
keywords={Autonomous vehicles;Task analysis;Planning;SData models;Surveys;Computational modeling;Visualization;Vision Language Model;Large Language Model;Autonomous Driving;Intelligent Vehicle;Conditional Data Generation;Decision Making;Language-guided Navigation;End-to-End Autonomous Driving},
doi={10.1109/TIV.2024.3402136}}
MAS comprises multiple autonomous entities: software agents, robots, or human-AI hybrids. This structure enables emergent collaboration and richer exploration of open-ended creative spaces, significantly underlying the practice for computational creativity. Here, computational creativity refers to the creation of artifacts—such as ideas, behaviors, or solutions—that are both novel and valuable, demonstrating clear usefulness or appeal rather than being random.
This survey focuses on systems whose inputs and outputs span text and images. We aim to map the current landscape of techniques, datasets, evaluations, and challenges to foster and measure creativity in such multimodal and heterogeneous systems. By analyzing how different agents interact, we reveal how collaborative structures can unlock creative potentials that exceed what isolated LLMs or individuals can achieve.
| Method | Publication | Profiling Method | Agent Persona |
|---|---|---|---|
Unleashing the Emergent Cognitive Synergy in Large Language Models: A Task-Solving Agent through Multi-Persona Self-Collaboration  |
arXiv(2024) | Model-Generated | Self-Defined |
| LLM Discussion: Enhancing the Creativity of Large Language Models via Discussion Framework and Role-Play |
COLM 2024 | Model-Generated | Self-Defined |
PersonaGym: Evaluating Persona Agents and LLMs  |
arXiv(2025) | Human-defined | Self-Defined |
| Systematic Idea Refinement for Machine Learning Research Agents | Tsinghua University Course:AML 2024 | Human-Defined | Assitant |
| Sparkit: A mind map-based mas for idea generation support. | EMAS 2024 | Human-Defined | Self-Defined |
| Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate |
EMNLP 2024 | Human-Defined | Debater |
| An Interactive Co-Pilot for Accelerated Research Ideation | HCINLP 2024 | Human-Defined | Mentor & Colleague |
| ChainBuddy: An AI-assisted Agent System for Generating LLM Pipelines | CHI 2025 | Human-Defined | Mentor & Planner |
| Method | Publication | Profiling Method | Agent Persona |
|---|---|---|---|
| TheAgentCompany: Benchmarking LLM Agents on Consequential Real World Tasks |
arXiv(2025) | Model-Generated | Company Employee |
| HOLLMWOOD: Unleashing the Creativity of Large Language Models in Screenwriting via Role Playin | EMNLP 2024 | Human-Defined | Artist |
| TRIZ Agents: A Multi-Agent LLM Approach for TRIZ-Based Innovation | ICAART 2025 | Human-Defined | Problem Solver |
| Towards an AI co-scientist | arxiv(2025) | Human-Defined | Researcher |
| MaCTG: Multi-Agent Collaborative Thought Graph for Automatic Programming | arxiv(2025) | Human-Defined | Programmer |
| DesignGPT: Multi-Agent Collaboration in Design | ISCID 2023 | Human-Defined | Self-Defined |
| CoQuest: Exploring Research Question Co-Creation with an LLM-based Agent |
CHI 2024 | Human-Defined | Researcher |
| LawLuo: A Multi-Agent Collaborative Framework for Multi-Round Chinese Legal Consultation | arxiv(2024) | Human-Defined | Lawyer |
| MARG: Multi-Agent Review Generation for Scientific Papers |
arxiv(2024) | Human-Defined | Expert |
| Papers | Year | Task | Code Link |
|---|---|---|---|
| ContextCam: Bridging Context Awareness with Creative Human-AI Image Co-Creation | 2024 | Image Generation | |
| AI and the Future of Collaborative Work: Group Ideation with an LLM in a Virtual Canvas | 2024 | Idea Generation | |
| How AI Processing Delays Foster Creativity: Exploring Research Question Co-Creation with an LLM-based Agent | 2024 | Research Ideation | |
| "It Felt Like Having a Second Mind": Investigating Human-AI Co-creativity in Prewriting with Large Language Models | 2024 | Creative Writing |
| Dataset | Year | Task | Data Link |
|---|
This repository is released under the Apache 2.0 license.
