[Paper] Evolving Symbiosis, from Barricelli's Legacy to Collective Intelligence: a simulated and conceptual approach

Source: arXiv - 2603.08463v1

Overview

The paper revisits Nils Aall Barricelli’s 1953 experiments on “symbiogenesis” – the emergence of cooperative, self‑replicating patterns in simple cellular automata – and extends them into modern computational settings. By recreating Barricelli’s 1‑D worlds, scaling to 2‑D “symbio‑organisms”, and probing DNA‑style norm constraints, the authors explore how low‑level interactions can give rise to open‑ended evolution and collective intelligence, with concrete take‑aways for artificial life (ALife) and AI research.

Key Contributions

• Faithful replication of Barricelli’s original 1‑D cellular automaton experiments, confirming historic results on numerical organism symbiosis.
• Extension to 2‑D environments, introducing richer spatial dynamics and more complex symbio‑organism structures.
• Introduction of “DNA‑norms”, a lightweight genotype‑phenotype mapping that guides organism behavior without full genetic encoding.
• Conceptual bridge between symbiogenesis, origins‑of‑life theories, and modern collective intelligence frameworks.
• Roadmap for future substrates, outlining how the ideas could be ported to neural networks, neural cellular automata, and other differentiable models.

Methodology

1. Re‑implementation of Barricelli’s 1‑D CA – The authors coded the original rule set (binary strings evolving under competition and cooperation) and ran large ensembles to verify emergent symbiotic patterns.
2. Design of a 2‑D cellular automaton – Cells hold integer values; local update rules combine replication, mutation, and a “resource sharing” operator that encourages neighboring cells to co‑operate.
3. DNA‑norms layer – Each organism carries a small vector of norm values (e.g., preferred density, alignment). During updates, cells evaluate whether local conditions satisfy these norms; compliance boosts replication fitness.
4. Experimental protocol – Simulations were executed on commodity hardware (CPU‑only) for up to 10⁶ timesteps, with metrics such as organism count, average lifespan, and cooperation index logged.
5. Analysis – Visual inspection of emergent patterns, statistical comparison between 1‑D and 2‑D runs, and ablation studies removing DNA‑norms to assess their impact.

Results & Findings

• Symbiotic clusters reliably re‑emerged in both 1‑D and 2‑D settings, confirming that simple competitive‑cooperative rules can sustain open‑ended growth.
• 2‑D worlds produced richer structures (e.g., branching colonies, ring‑like formations) that persisted longer than their 1‑D counterparts.
• DNA‑norms increased stability: organisms adhering to norm constraints showed a 27 % higher average lifespan and a 15 % boost in replication rate, indicating that lightweight “genetic” guidance can steer emergent dynamics without heavy encoding.
• Collective intelligence proxies (e.g., coordinated movement toward resources) emerged spontaneously, suggesting that symbiogenesis can be a bottom‑up route to coordinated behavior.
• Scalability: The simulations ran efficiently on a single CPU core, demonstrating that the approach is computationally cheap and suitable for rapid prototyping.

Practical Implications

• AI research – The DNA‑norm concept offers a minimalist way to embed soft constraints into evolving agents, which could be leveraged for curriculum learning, multi‑agent coordination, or safe exploration in reinforcement learning.
• Procedural content generation – Game developers can use the 2‑D symbio‑organism engine to generate organic, self‑evolving terrains, flora, or enemy swarms that adapt over gameplay sessions.
• Distributed systems – The underlying principles of local cooperation leading to global robustness may inspire new protocols for peer‑to‑peer networks, swarm robotics, or edge‑computing clusters where nodes self‑organize without central control.
• Educational tools – The low‑cost simulation can serve as an interactive teaching aid for concepts like emergence, evolution, and collective behavior in computer science curricula.
• Cross‑disciplinary research – By providing a concrete computational platform, the work invites collaborations with synthetic biology, where similar rule‑based symbiosis could be explored in wet‑lab settings.

Limitations & Future Work

• Simplified physics – The cellular automata lack explicit energy or resource budgets, limiting realism for biological analogues.
• Fixed rule set – While the authors experimented with a few rule variations, a systematic exploration of the rule space (e.g., via meta‑evolution) remains open.
• Scalability to high‑dimensional substrates – Porting the framework to neural networks or neural cellular automata will require careful handling of differentiability and training stability.
• Quantitative metrics for intelligence – Current measures focus on lifespan and cooperation; developing rigorous benchmarks for emergent collective intelligence is a next step.

The authors outline a roadmap that includes integrating differentiable CA, testing on GPU clusters, and collaborating with neuroscientists to map symbiogenesis concepts onto brain‑inspired architectures.

Authors

• James Ashford
• Marko Cvjetko
• Richard Löffler
• Berfin Sakallioglu
• Alessandro Valerio
• Marta Tataryn
• Benedikt Hartl
• Léo Pio-Lopez
• Stefano Nichele

Paper Information

• arXiv ID: 2603.08463v1
• Categories: cs.NE
• Published: March 9, 2026
• PDF: Download PDF