Participatory Evolution of Artificial Life Systems via Semantic Feedback DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation email; June 03–05, 2018; Woodstock, NY ISBN: 978-1-4503-XXXX-X/2018/06 CCS: Computing methodologies Artificial life

Shuowen Li, Kexin Wang, Minglu Fang, Danqi Huang, Ali Asadipour, Haipeng Mi, Yitong Sun · 2025

Read Participatory Evolution of Artificial Life Systems via Semantic Feedback DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation email; June 03–05, 2018; Woodstock, NY ISBN: 978-1-4503-XXXX-X/2018/06 CCS: Computing methodologies Artificial life by Shuowen Li, Kexin Wang, Minglu Fang, Danqi Huang, Ali Asadipour, Haipeng Mi, Yitong Sun (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-03839.txt.

Introduction

Artificial life software grows creatures by blind mutation and selection, which leaves the human user with little say. Shuowen Li and six co-authors add language to the loop in a 2025 paper. Their framework lets a sentence of intent steer the evolution of a simulated ecosystem.

Three parts do the work. An encoder turns prompts into parameter settings. An evolution strategy optimizer breeds and tests variants. A vision-language model scores how well each result matches the words. Users can refine prompts mid-run, agents interact, and new behavioral rules emerge from the process. In user studies, the language-guided runs matched what people asked for better than hand tuning did. The authors offer the system as a platform for shared, open-ended design between people and evolving software.

Catalog record

CREATORShuowen Li, Kexin Wang, Minglu Fang, Danqi Huang, Ali Asadipour, Haipeng Mi, Yitong Sun
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE27 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-03839

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Participatory Evolution of Artificial Life Systems via Semantic Feedback DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation email; June 03–05, 2018; Woodstock, NY ISBN: 978-1-4503-XXXX-X/2018/06 CCS: Computing methodologies Artificial life CCS: Applied computing Media arts Shuowen Li, Kexin Wang, Minglu Fang, Danqi Huang, Ali Asadipour, Haipeng Mi, Yitong Sun Abstract . We present a semantic feedback framework that enables natural language to guide the evolution of artificial life systems. Integrating a prompt-to-parameter encoder, a CMA-ES optimizer, and CLIP-based evaluation, the system allows user intent to modulate both visual outcomes and underlying behavioral rules. Implemented in an interactive ecosystem simulation, the framework supports prompt refinement, multi-agent interaction, and emergent rule synthesis. User studies show improved semantic alignment over manual tuning and demonstrate the system’s potential as a platform for participatory generative design and open-ended evolution. Figure 1. Evolutionary Pathways of Digital Lifeforms and Schematic Illustration of Ecological States in a Digital Life System. Enjoying the baseball game from the third-base seats. Ichiro Suzuki preparing to bat. 1. Introduction Human interaction with complex systems, including ecological environments, social dynamics, and artificial simulations, presents ongoing challenges for both modeling and creative intervention. While agent-based models and artificial life systems provide formal tools for simulating emergent behavior, they often rely on fixed rule structures and limited input modalities. This makes it difficult to embed high-level conceptual intent into evolving system dynamics. Recent advances in multimodal AI, particularly vision-language models such as CLIP, have expanded the potential of using natural language to guide generative systems. However, most implementations adopt a static prompt-to-output paradigm in which language acts as a trigger rather than a continuous modulator. This restricts semantic iteration and reduces a system’s ability to adapt behavior in response to evolving user input. We introduce a semantic feedback-driven framework for evolving artificial life systems, where natural language serves as both expressive input and regulatory signal. Grounded in the Swarm model, our architecture integrates a BERT-based prompt-to-parameter encoder, a CMA-ES evolutionary optimizer, and a CLIP-based semantic evaluation module. This closed-loop pipeline allows language to influence both the visual output and the behavioral rules of the simulation, enabling interpretable, adaptive, and semantically aligned evolution. To demonstrate the framework’s capabilities, we implemented it in an interactive platform that enables users to co-construct a semantic ecosystem. Participants generate and evolve synthetic agents using language prompts. These agents coexist and interact in a shared artificial environment, exhibiting behaviors shaped by both local rules and emergent collective dynamics. As prompts accumulate, the system analyzes linguistic histories and behavioral trajectories to derive meta-rules that recursively influence subsequent evolution. We evaluated the system in two settings: a controlled study assessing its alignment with user intent, and a public deployment supporting open-ended co-creation. Results suggest that semantic input enhances controllability, interpretability, and user engagement, positioning the framework as a novel platform for participatory artificial life and expressive complex system design. 2. Related Work 2.1. Consciousness and the Emergence of Artificial Life Contemporary theories of consciousness, including relational, ecological, and generative perspectives, challenge traditional assumptions that define consciousness as uniquely human and a priori. Frameworks such as animism ( Harvey 2005 ) , the extended mind ( Clark and Chalmers 1998 ) , and embodied cognition ( Varela Francisco et al. 1991 ) reconceptualize consciousness as a regulatory process emerging from system–environment coupling. Within this view, consciousness operates as a feedback mechanism that enables self-organization, boundary modulation, and adaptive response to perturbations. These principles have inspired the design of artificial systems that embed similar dynamics of self-regulation and emergent behavior. Agent-based models, particularly those within the artificial life (ALife) tradition, such as Reynolds’ Boids ( Reynolds 1987 ) , cellular automata, and swarm-based ecologies ( Sayama 2009 ) , offer formal tools for studying emergent complexity through decentralized local interaction. These systems exemplify how simple rules at the individual level can give rise to lifelike collective phenomena. While ALife models have been widely applied in generative art, ecological simulation, and swarm robotics, most rely on predefined rules and do not incorporate semantically adaptive input, limiting their responsiveness to human conceptual intent ( Park et al. 2023 ; Wang et al. 2024 ) . This opens new possibilities for language-integrated artificial life systems. Rather than treating language as a symbolic descriptor of output, we explore its function as a continuous regulatory input that is functionally analogous to consciousness. In this role, language actively shapes system evolution in real time. Our work positions semantic feedback as a mediator between user intent and behavioral adaptation, extending the scope of ALife through participatory, linguistically guided evolution. 2.2. Language Models as Evolutionary Drivers Recent advances in multimodal embedding models, particularly CLIP, have greatly improved systems’ capacity to interpret and respond to natural language within generative contexts ( Tian and Ha 2022 ) . Platforms such as Midjourney, DALL·E, and Stable Diffusion translate textual prompts into rich visual outputs, while systems like DreamFusion ( Poole et al. 2022 ) and Text2Mesh ( Michel et al. 2022 ) extend these capabilities into the realm of 3D generation. Despite their sophistication, these platforms follow a predominantly static prompt - to - output paradigm, in which user input does not influence the underlying generative mechanisms over time. At the same time, foundational research in artificial life has begun to integrate language-aligned foundation models. Kumar et al.’s Automated Search for Artificial Life (ASAL) framework leverages vision-language models to explore diverse lifelike behaviors across multiple ALife substrates, such as Boids, Lenia, and cellular automata, discovering open-ended novelty and human-aligned emergence ( Kumar et al. 2024 ) . This work exemplifies how semantic evaluation tools can direct emergent simulations toward biologically evocative patterns, underscoring the potential of language-grounded open-ended search within ALife. Building on this trajectory, our system enables semantic input to govern not only the visual output but also the behavioral rules in a swarm-based artificial life simulation. Natural language prompts are encoded into parameter vectors that act as both seeds and directional priors. A CLIP-based fitness function quantifies alignment between simulation behaviors and user intent, guiding the CMA-ES optimizer to adapt system dynamics progressively. This closed-loop process realizes a form of language-curated evolution, dynamically evolving agent-based systems in response to continuous semantic signals. Figure 2. System architecture of the semantic feedback-driven evolution framework. Users input a natural language prompt, which is encoded via a text encoder and transformed into guiding parameters for a population-based simulation. A CMA-ES optimizer evaluates each population using CLIP-based semantic similarity between the prompt and simulation outputs. The optimization iteratively updates the parameters to align system behavior with the conceptual intent. The simulation loop is GPU-accelerated, enabling real-time visual feedback over successive generations. The transformer-based embedding model aligns image and text embeddings in a shared semantic space to compute fitness values. 2.3. Participatory Artificial Ecosystems and Generative Art Artificial life has long intersected with generative art through systems that embed user input into emergent behavioral frameworks. Pioneering works such as Sommerer and Mignonneau’s Interactive Plant Growing (1994) and A-Volve (1995) ( Sommerer and Mignonneau 1997 ; Sommerer and Mignonneau 2002 ) enabled participants to physically engage with biological or digital organisms, prompting growth and transformation in evolving life forms through touch and gesture-based interaction.

Read the complete text