Artificial Neural Networks have been crowned with tremendous successes in recent years, with ever wider and more complex ranges of applications. However, they, too often, result from a costly human design process relying as much on expertise as on trial and error. While the field of NeuroEvolution provides a complementary view point through emergent, self-designing ANNs, the “black-box” properties of the resulting networks is further magnified. Still, by once more taking inspiration from biology, we may extract meaningful information from ANNs by using similar approaches as those used for biological brains. In this work, we study the emergence and functional allocation of neurons in a light communication task. By having a robot transmit visual information, through vocal channels, we enrich the existing literature with new types of stimuli, namely those related to role (emitter/ receiver). Through Virtual functional Magnetic Resonance Imaging (VfMRI), we observe that evolution only favored specific kind of input-processing modules. Combined with a strong presence of jack-of-alltrades modules, this demonstrates the balancing act between specialization and generalization in Artificial Neural Networks with emergent topologies.

Biological systems exhibit hierarchical and intricate mechanisms that enable self-sustenance and open-ended behavior. This organizational closure is arguably one of life’s hallmarks, and it is facilitated by the widespread utilization of enzymes. Enzymes enhance improbable pathways, enabling the formation of complex structures and functions. Here, we propose a model to characterize artificial enzymes within an artificial “soup” of functions. We contend that these enzymes can emerge from elementary interactions among functions, and they should foster rapid complexity growth, owing to their ability to construct auto-catalitic networks.

Cooperation is essential for both human and artificial life societies, yet understanding how to promote it remains a complex challenge. Indirect reciprocity, where individuals cooperate to maintain a good reputation, is one mechanism to encourage cooperation. To promote stable cooperation, society needs social norms that stipulate how individuals should behave and how they should evaluate others. Previous research has identified a set of effective social norms, called the “leading eight”, for achieving evolutionarily stable cooperation. In this study, we expand on a classical framework in two significant ways. First, we include norms that update the reputations of passive receivers. Second, we introduce stochasticity to social norms. We theoretically derived the necessary and sufficient conditions for evolutionarily stable norms that result in full cooperation within this generalized model. Our findings offer a new perspective on prior research and provide a foundation for future studies in this field.

This paper presents an energy-based approach for simulating virtual creatures, advocating for a shift from traditional monolithic physics engines to a more flexible implementation approach centered on energy minimization and automatic differentiation. By integrating insights from established disciplines alongside emerging concepts such as scale-free cognition, this approach enables a comprehensive modeling of behaviors, where everything from basic physical phenomena, such as inertia and elasticity, to more complex behaviors, such as robust locomotion, can be interpreted as goal-directed behavior.

Cartesian Genetic Programming (CGP) literature repeatedly reports that crossover operators hinder CGP search compared to a 1 + λ strategy based on mutation only. Though there have been efforts in making CGP crossover operators work, the literature is relatively evasive on why the phenomenon is observed at all. This contrasts with what happens in Linear Genetic Programming (LGP), where we know that crossover works well. While both CGP and LGP individuals can be represented as directed acyclic graphs (DAGs), changing a single connection gene in a CGP individual can drastically alter the activeness of nodes in the entire graph, as opposed to LGP where crossover changes are much more beneficial. In this contribution, we demonstrate the phenomenon and show that LGP evolution produces children that are far more similar to their parents than in CGP. This lets us propose that the design of LGP, namely the inclusion of steady-state memory registers and program size regulation, serves to protect highfitness substructures from perturbation in a way that is not provided for in CGP.

Plants are complex organisms, showing collective adaptive behavior. Plant behavior is often defined by shoot growth, yet root systems exhibit equally complex, less visible, behaviors. Roots have to navigate in a particular environment, while optimizing nutrient and water uptake as well as avoiding exposure to harmful elements. In this paper, we introduce an interactive, agent-based simulation model of root growth. It supports the exploration of the interplay of different root models within different environments. To this end, we resort to Swarm Grammars (SGs) which combine the interactivity of spatial agents with the generative perspectives of LSystems. We pursue a point-based representation of the environment due to its versatility with respect to modeling and rendering possibilities. SGs and point-based environments are combined in a simulation that enables interactions between a user, agents and the environment at runtime. We validate the model by recreating and analysing several established root model configurations, and validate the benefits of the interactive simulation by an expert interview.

A common subject: Evolution through a computational lens. Two different communities: on the one hand, artificial life researchers use computational systems to understand emergent evolutionary processes and patterns such as complexity, robustness, evolvability and open-endedness; on the other hand, evolutionary bioinformatics researchers decipher patterns and processes in diverse domains of life on Earth using computational methods based on biological data. Both communities use simulations of living organisms but with different aims, objects, and methods, resulting in disjoint research corpuses. We propose Aevol 4b, an artificial life evolution simulator, and show that the data it produces can be successfully and interestingly processed using bioinformatics methods. This bridges the gap between the two fields and paves the way for fruitful exchanges between artificial life models and bioinformatic analysis methods.

Human culture relies on collective innovation: our ability to continuously explore how existing elements in our environment can be combined to create new ones. Language is hypothesized to play a key role in human culture, driving individual cognitive capacities and shaping communication. Yet the majority of models of collective innovation assign no cognitive capacities or language abilities to agents. Here, we contribute a computational study of collective innovation where agents are Large Language Models (LLMs) that play Little Alchemy 2, a creative video game originally developed for humans that, as we argue, captures useful aspects of innovation landscapes not present in previous test-beds. We, first, study an LLM in isolation and discover that it exhibits both useful skills and crucial limitations. We, then, study groups of LLMs that share information related to their behaviour and focus on the effect of social connectivity on collective performance. In agreement with previous human and computational studies, we observe that groups with dynamic connectivity out-compete fully-connected groups. Our work reveals opportunities and challenges for future studies of collective innovation that are becoming increasingly relevant as Generative Artificial Intelligence algorithms and humans innovate alongside each other.

In this study, we investigated the extent to which Vision Language Models (VLMs) possess sensibilities similar to those of humans by focusing on color impressions, which have a significant impact on the sensory aspects of vision, and sound symbolism, which constitutes linguistic and auditory sensibilities. For the experiments, we newly constructed an evolving image generation system based on the CONRAD algorithm, which evolves images based on human evaluations. Our system can also reflect the evaluations of VLMs in addition to humans. Using this system, we analyzed the sensibilities of VLMs. The experimental results suggested similarities between human and VLM sensibilities in both color impressions and sound symbolism. In sound symbolism, VLMs demonstrated sound-symbolic sensibilities similar to those of humans, even for the pseudo-words we newly generated, yielding intriguing results. These findings suggest that VLM evaluations and feedback may have a certain level of effectiveness in tasks that have previously required human evaluations or annotations related to sensibility.

ALife is primed to address the biggest challenges in astrobiology by simulating systems which capture the most general and fundamental features of living systems. One such challenge is how to detect life outside of the solar system— especially without making strong assumptions about how life would manifest and interact with its planetary environment. Here we explore an ALife model meant to overcome this problem, by focusing on what life may do, rather than what life may be: life can spread between planetary systems (panspermia) and can modify planetary characteristics (terraformation). Our model shows that as life propagates across the galaxy, correlations emerge between planetary characteristics and location, and these correlations can function as a biosignature. This biosignature is agnostic because it is independent of strong assumptions about any particular instantiation of life or planetary characteristic. We demonstrate (and evaluate) a way to prioritize specific planets for further observation—based on their potential for containing life. We consider obstacles that must be overcome to practically implement our approach, including identifying specific ways in which better understanding astrophysical and planetary processes would improve our ability to detect life.

The idea that the Earth system self-regulates itself in a habitable state was proposed in the 1970s by James Lovelock, later on formalized as the Daisyworld model using a two-species system interacting with their environment. The potential for testing this conceptual framework in an experimental way is limited by its scale. To fill this gap, here we propose an explicit test tube-scale implementation for a microbial synthetic Daisyworld using an engineered community where pH as the external, abiotic control parameter. The computational modelling of this system shows robust self-regulation within a broad range of conditions, limited by tipping points, This synthetic Daisyworld allows exploring multiple scenarios of self-regulation that include the role of parasites, fluctuations or biodiversity and can help developing an experimental path to Earth Systems Science.

This paper introduces Alter3, a humanoid robot that demonstrates spontaneous motion generation through the integration of GPT-4, Large Language Model (LLM). This overcomes challenges in applying language models to direct robot control. By translating linguistic descriptions into actions, Alter3 can autonomously perform various tasks. The key aspect of humanoid robots is their ability to mimic human movement and emotions, allowing them to leverage human knowledge from language models. This raises the question of whether Alter3+GPT-4 can develop a “minimal self” with a sense of agency and ownership. This paper introduces mirror self-recognition and rubber hand illusion tests to assess Alter3’s potential for a sense of self. The research suggests that even disembodied language models can develop agency when coupled with a physical robotic platform.

In many situations, communication between agents is a critical component of cooperative multi-agent systems, however, it can be difficult to learn or evolve. In this paper, we investigate a simple way in which the emergence of communication may be facilitated. Namely, we explore the effects of when agents can mimic preexisting, externally generated useful signals. The key idea here is that these signals incentivise listeners to develop positive responses, that can then also be invoked by speakers mimicking those signals. This investigation starts with formalising this problem, and demonstrating that this form of mimicry changes optimisation dynamics and may provide the opportunity to escape non-communicative local optima. We then explore the problem empirically with a simulation in which spatially situated agents must communicate to collect resources. Our results show that both evolutionary optimisation and reinforcement learning may benefit from this intervention.

Behavioral changes that result from rapid environmental shifts such as those brought about by human activity are some of the most immediate and consequential responses in the biotic world. However, deriving community models that take into account non-trivial behavior and therefore the ecological significance of those changes is an ongoing challenge in ecology. Here, we propose methods for deriving community models from populations of evolved agents who both forage and avoid predation. We exemplify those methods with a case study of sensory pollution by manipulating the sensor of the agents and deriving functions that characterize resulting interaction rates with both food and predators. We believe these methods can apply to any question regarding how a specific behavior or behavioral change will affect community structure and population dynamics. Spatial limitations of the method are discussed as an area for future work.

We gather many perspectives on Capital and synthesize their commonalities. We provide a characterization of Capital as a historical agential system and propose a model of Capital using tools from computer science. Our model consists of propositions which, if satisfied by a specific grounding, constitute a valid model of Capital. We clarify the manners in which Capital can evolve. We claim that, when its evolution is driven by quantitative optimization processes, Capital can possess qualities of Artificial Intelligence. We find that Capital may not uniquely represent meaning, in the same way that optimization is not intentionally meaningful. We find that Artificial Intelligences like modern day Large Language Models are a part of Capital. We link our readers to a web-interface where they can interact with a part of Capital.

To achieve autonomous cooperation among heterogeneous agents, we propose the artificial minimal self. The artificial minimal self is implemented based on the free energy principle (FEP), which all living organisms are supposed to follow. In a standard FEP, the generative model configured for each physical agent consists only of its own observations and actions. The key to autonomous cooperation among heterogeneous agents is to extend the sense of self beyond to others to include interaction with others to self. In this study, focusing on Gallagher’s minimal self where the self is extended to others, the generative model integrates its own observations and actions and those of other heterogeneous agents. Multiple heterogeneous physical agents have different embodiments, resulting in different types of observations and actions. To integrate different types of observations and actions, we define observations and actions based on environmental information independent of the agent’s embodiment. Integrating the observations and actions of multiple physical agents into a generative model allows extending the sense of self beyond to others, leading to autonomous cooperation between heterogeneous agents. We demonstrated cooperative object transport by two heterogeneous robot arms implementing the artificial minimal self. The results confirmed that each robot arm can autonomously share tasks and transport objects by simply providing a common goal of placing the object into the goal. Furthermore, the process of self-extension to others reproduced the behavior of observing the characteristics of others that is seen in human cooperation.

Cellular automata (CA) are discrete dynamical systems with a prominent place in the history and study of artificial life. Here, we focus on the density classification task (DCT) in which a 1-dimensional lattice of Boolean (on/off) automata must perform a form of rudimentary quorum sensing. Typically, the ring lattice consists of 149 cells (though we consider other sizes as well) that update their state according to their own state and its six nearest neighbors in the previous time step. The goal is obtaining Boolean CA rules whose dynamics converges to the majority state of the entire lattice for a given initial configuration of the lattice. This is a nontrivial task because cells have access only to local information, and thus need to integrate and coordinate information across the lattice to converge to the correct collective state. Because initial conditions are random, they have very similar proportions of on and off states, which makes the problem very difficult. This problem has hitherto been studied with the assumption that input to each cell is perfectly stable. Since biological systems that solve similar problems (e.g. bacterial quorum sensing) must operate in noisy environments, here we study the impact of noise on DCT accuracy for the 13 highest-accuracy CA rules from the literature. We use cubewalkers, a recently released GPU-accelerated Boolean simulator to conduct large-scale random experiments. We uncover a trade-off between maximum accuracy without noise and robustness to noise among these high-performance CAs. Moreover, there is no significant difference between rules that were human-designed or evolved computationally.

Navigating the intricacies of digital environments demands effective strategies for fostering cooperation and upholding norms. Banning and moderating the content of influential users on social media platforms is often met with shock and awe, yet clearly has profound implications for online behaviour and digital governance. Drawing inspiration from these actions, we model broadcasting retributive measures. Leveraging analytical modeling and extensive agent-based simulations, we investigate how signaling punitive actions can deter anti-social behaviour and promote cooperation in online and multi-agent systems. Our findings underscore the transformative potential of threat signaling in cultivating a culture of compliance and bolstering social welfare, even in challenging scenarios with high costs or complex networks of interaction. This research offers valuable insights into the mechanisms of promoting pro-social behaviour and ensuring behavioural compliance across diverse digital ecosystems.

The softness of robots that navigate the real world is frequently hamstrung by rigid elements such as traditional computers. One avenue that may reduce the reliance on these types of devices is the application of physical reservoir computing (PRC). Previous studies have shown that by leveraging the under-actuated nonlinear dynamics of soft mechanisms, complex computing tasks can be achieved. In this study we present a new octopus-inspired walking and swimming robot: Takorobo. We investigated the degree to which locomotion significant tasks (including body motion prediction and direct actuator control) can be embedded into this robot using its four soft sensory tentacles. The robot was found to be able to accurately compute its body motions and successfully exercise direct closed-loop PRC control on both land and water for some control signals.

Neural Cellular Automata (NCA) is a class of Cellular Automata where the update rule is parameterized by a neural network that can be trained using gradient descent. In this paper, we focus on NCA models used for texture synthesis, where the update rule is inspired by partial differential equations (PDEs) describing reaction-diffusion systems. To train the NCA model, the spatio-temporal domain is discretized, and Euler integration is used to numerically simulate the PDE. However, whether a trained NCA truly learns the continuous dynamic described by the corresponding PDE or merely overfits the discretization used in training remains an open question. We study NCA models at the limit where spacetime discretization approaches continuity. We find that existing NCA models tend to overfit the training discretization, especially in the proximity of the initial condition, also called ”seed”. To address this, we propose a solution that utilizes uniform noise as the initial condition. We demonstrate the effectiveness of our approach in preserving the consistency of NCA dynamics across a wide range of spatio-temporal granularities. Our improved NCA model enables two new test-time interactions by allowing continuous control over the speed of pattern formation and the scale of the synthesized patterns. We demonstrate this new NCA feature in our interactive online demo. Our work reveals that NCA models can learn continuous dynamics and opens new venues for NCA research from a dynamical system’s perspective.