We present Mass-Conserving Evolution (MaCE), a general method for implementing mass conservation in Cellular Automata (CA). MaCE is a simple evolution rule that can be easily ’attached’ to existing CAs to make them mass-conserving, which tends to produce interesting behaviours more often, as patterns can no longer explode or die out. We first show that MaCE is numerically stable and admits a simple continuous limit. We then test MaCE on Lenia, and through several experiments, we demonstrate that it produces a wide variety of interesting behaviours, starting from the variety and abundance of solitons up to hints of intrinsic evolution in resource-constrained environments. Finally, we showcase the versatility of MaCE by applying it to Neural-CAs and discrete CAs, and discuss promising research directions opened up by this scheme.

We present Cybernetic Mimesis, an agent-based model that integrates Perceptual Control Theory with Mimetic Theory. Each agent runs several negative-feedback loops and, upon reaching a refueling location, resets the associated internal error to zero, thereby driving the corresponding perception to its reference. At each step, the agent acts to minimize its largest internal error. Crucially, agents also amplify the internal error tied to the need most common in their local neighborhood—effectively imitating neighbors’ goal references rather than their movements. This single mimetic coupling yields rich population-level dynamics. Across 30 randomly seeded runs, we observe endogenous boom–bust cycles: dominant-need share exceeds 87%, volatility triples, and crashes become asymmetric, all without exogenous shocks or ad hoc switching. Cybernetic Mimesis thus offers a minimal, mechanistic account of how local reference imitation can endogenously produce market-like booms and busts.

Tokenization, a core concept in natural language processing, segments sequences into discrete, repetitive informational units. Here, we fit tokenizers to genomic data from a broad range of organisms to investigate how biological information is structured and compressed. We show that tokenization consistently outperforms standard compression algorithms, particularly for coding sequences. Compression improves with sequence length and varies systematically across taxonomic kingdoms. Token frequency patterns also reflect biological relationships, especially among eukaryotes. These findings suggest that tokenization offers a powerful framework for exploring how information is encoded in living systems.

The Cognitive Buffer Hypothesis (CBH) posits that larger brains evolved to enhance survival in changing conditions. However, larger brains also carry higher energy demands, imposing additional metabolic burdens. Alongside brain size, brain organization plays a key role in cognitive ability and, with suitable architectures, may help mitigate energy challenges. This study evolves Artificial Neural Networks (ANNs) used by Reinforcement Learning (RL) agents to investigate how environmental variability and energy costs influence the evolution of neural complexity, defined in terms of ANN size and structure. Results indicate that under energy constraints, increasing seasonality led to smaller ANNs. This challenges CBH and supports the Expensive Brain Hypothesis (EBH), as highly seasonal environments reduced net energy intake and thereby constrained brain size. ANN structural complexity primarily emerged as a byproduct of size, where energy costs promoted the evolution of more efficient networks. These results highlight the role of energy constraints in shaping neural complexity, offering in silico support for biological theory and energy-efficient robotic design.

Enactivism has had an important conceptual influence on Artificial Life. Its notorious mind-life continuity thesis is a fundamental idea that once sustained much of the theoretical framework, but has recently lost support, explicitly in cognitive science, and by omission in Artificial Life. Alternative computational accounts, however, do not seem capable of filling the vacuum it leaves behind, given that according to the Church-Turing thesis, the limit of what is computable is bounded by Turing machines (regardless of the specific instantiation, such as cellular automata, deep, spiking or any other form of neural networks, or even quantum computers, to name some well known cases) and that computable functions formally describe the notion of mechanism, every organismic process that determines a consistent response should be both limited to Turing-machine capabilities and amenable to formalization. This impending issue, namely, the relation among life, cognition and computation as the hypothetical elements of the mind, is particularly pertinent and highly relevant for Artificial Life. Moreover, although it is often overlooked or omitted for practical reasons, it will probably define, or so we argue, the direction of future research in cognitive science, artificial intelligence and related fields.

This study investigates the emergence of collective behavior in a system of coupled oscillators by extending Tanaka’s Swarm Oscillator model within the framework of active inference. In this formulation, each oscillator minimizes its own prediction error, leading to autonomous regulation of internal states and motion. Through numerical simulations, we examine the system’s long-term dynamics across a range of phase-bias parameters. The analysis reveals eight distinct behavioral patterns, including rotating clusters, traveling linear formations, waving curves, amorphous structures, and static configurations. Each phase is characterized by structural arrangement, movement, spatial confinement, and the degree of phase synchrony among oscillators. While some of the observed behaviors resemble those found in the original Swarm Oscillator model, others arise uniquely from the incorporation of predictive dynamics. The resulting phase diagram highlights how simple internal generative models embedded within interacting elements can produce rich and diverse collective phenomena. These findings offer new insights into how fundamental features of agency can emerge in minimal systems through local interactions and prediction-based control.

Agent-based simulation platforms play a key role in enabling fast-to-run evolution experiments that can be precisely controlled and observed in detail. Availability of high-resolution snapshots of lineage ancestries from digital experiments, in particular, is key to investigations of evolvability and open-ended evolution, as well as in providing a validation testbed for bioinformatics method development. Ongoing advances in AI/ML hardware accelerator devices, such as the 850,000-processor Cerebras Wafer-Scale Engine (WSE), are poised to broaden the scope of evolutionary questions that can be investigated in silico. However, constraints in memory capacity and locality characteristic of these systems introduce difficulties in exhaustively tracking phylogenies at runtime. To overcome these challenges, recent work on hereditary stratigraphy algorithms has developed space-efficient genetic markers to facilitate fully decentralized estimation of relatedness among digital organisms. However, in existing work, compute time to reconstruct phylogenies from these genetic markers has proven a limiting factor in achieving large-scale phyloanalyses. Here, we detail an improved trie-building algorithm designed to produce reconstructions equivalent to existing approaches. For modestly-sized 10,000-tip trees, the proposed approach achieves a 300-fold speedup versus existing state-of-the-art. Finally, using 1 billion genome datasets drawn from WSE simulations encompassing 954 trillion replication events, we report a pair of large-scale phylogeny reconstruction trials, achieving end-to-end reconstruction times of 2.6 and 2.9 hours. In substantially improving reconstruction scaling and throughput, presented work establishes a key foundation to enable powerful high-throughput phyloanalysis techniques in large-scale digital evolution experiments.

Drawing parallels between Deep Artificial Neural Networks (DNNs) and biological systems can aid in understanding complex biological mechanisms that are difficult to disentangle. Temporal processing, an extensively researched topic, is one such example that lacks a coherent understanding of its underlying mechanisms. In this study, we investigate temporal processing in a Deep Reinforcement Learning (DRL) agent performing an interval timing task and explore potential biological counterparts to its emergent behavior. The agent was successfully trained to perform a duration production task, which involved marking successive occurrences of a target interval while viewing a video sequence. Analysis of the agent’s internal states revealed oscillatory neural activations, a ubiquitous pattern in biological systems. Interestingly, the agent’s actions were predominantly influenced by neurons exhibiting these oscillations with high amplitudes and frequencies corresponding to the target interval. Parallels are drawn between the agent’s time-keeping strategy and the Striatal Beat Frequency (SBF) model, a biologically plausible model of interval timing. Furthermore, the agent maintained its oscillatory representations and task performance when tested on different video sequences (including a blank video). Thus, once learned, the agent internalized its timekeeping mechanism and showed minimal reliance on its environment to perform the timing task. A hypothesis about the resemblance between this emergent behavior and certain aspects of the evolution of biological processes like circadian rhythms, has been discussed. This study aims to contribute to recent research efforts of utilizing DNNs to understand biological systems, with a particular emphasis on temporal processing.

In Artificial Life (ALife) research, replicating Open-Ended Evolution (OEE)—the continuous emergence of novelty observed in biological life—has usually been pursued within isolated, closed system simulations, such as Tierra and Avida, which have typically plateaued after an initial burst of novelty, failing to achieve sustained OEE. Scholars suggest that OEE requires an open-environment system that continually exchanges information or energy with its environment. A recent technological innovation in Decentralized Physical Infrastructure Network (DePIN), which provides permissionless computational substrates, enables the deployment of Large Language Model–based AI agents on blockchains integrated with Trusted Execution Environments (TEEs). This enables on-chain agents to operate autonomously “in the wild”, achieving self-sovereignty without human oversight. These agents can control their own social media accounts and cryptocurrency wallets, allowing them to interact directly with blockchain-based financial networks and broader human social media. Building on this new paradigm of on-chain agents, Spore.fun is a recent real-world AI evolution experiment that enables autonomous breeding and evolution of new on-chain agents. This paper presents a detailed case study of Spore.fun, examining agent behaviors and their evolutionary trajectories through digital ethology. We aim to spark discussion about whether open-environment ALife systems “in the wild”, based on permissionless computational substrates and driven by economic incentives to interact with their environment, could finally achieve the long-sought goal of OEE.

Memory enables organisms to retain information about aspects of life relevant to their survival and plays a key role in the iterated Prisoner’s Dilemma. In this work, we examine a new, condensed memory representation that allows individuals to work with summaries of past encounters. This simplification enables us to easily identify cooperation thresholds within an organism’s memory of past opponent moves, theoretically fitness effects and the value of memory under assorted environmental conditions. Under restrictive conditions where memory usage incurs a cost, cooperation tends to devolve quickly, revealing high volatility. Nevertheless, organisms with summarized memory retain the ability to coevolve cooperative strategies in environments where memory is either free, or especially beneficial, such as where a single “hard defect” occurs and otherwise cooperating players must recover from it. We empirically verify fitness predictions for transition points of Always Defect, Always Cooperate, Titfor-Tat, and Majority Response 3 (where the most common opponent response is used over the last three rounds). Source Code: https://github.com/revolins/IPD_CR