Advanced biological intelligence learns efficiently from an in-formation-rich stream of stimulus information, even when feed-back on behaviour quality is sparse or absent. Such learning ex-ploits implicit assumptions about task domains. We refer to such learning as Domain-Adapted Learning (DAL). In contrast, AI learning algorithms rely on explicit externally provided measures of behaviour quality to acquire fit behaviour. This im-poses an information bottleneck that precludes learning from di-verse non-reward stimulus information, limiting learning effi-ciency. We consider the question of how biological evolution circumvents this bottleneck to produce DAL. We propose that species first evolve the ability to learn from reward signals, providing inefficient (bottlenecked) but broad adaptivity. From there, integration of non-reward information into the learning process can proceed via gradual accumulation of biases induced by such information on specific task domains. This scenario pro-vides a biologically plausible pathway towards bottleneck-free, domain-adapted learning. Focusing on the second phase of this scenario, we set up a population of NNs with reward-driven learning modelled as Reinforcement Learning (A2C), and allow evolution to improve learning efficiency by integrating non-re-ward information into the learning process using a neuromodu-latory update mechanism. On a navigation task in continuous 2D space, evolved DAL agents show a 300-fold increase in learning speed compared to pure RL agents. Evolution is found to elimi-nate reliance on reward information altogether, allowing DAL agents to learn from non-reward information exclusively, using local neuromodulation-based connection weight updates only. Code available at github.com/aislab/dal.

This paper aims at considering novel and practical applications of ALife techniques to design a co-creative social dynamics in an online and virtual space, which is becoming important because of the recent emergence of various types of online communication platforms due to outbreaks of COVID-19. Recently, spatial and online communication services, such as SpatialChat, have attracted more attention. Each participant is represented as an avatar or icon in a virtual 2D space. She can move it around in the space and listen to neighbors’ voices of which volume become louder as they get closer to her. However, the overall structure of communications tends to be deadlocked, which might make participants lose chances to communicate with many other people. We design and investigate a virtual agent, called “facilitator agent,” as a study towards realization of practical agents that facilitate novel and cooperative interactions in a spatial and online communication by giving human participants opportunities to communicate with many others cooperatively. We adopt a Social Particle Swarm (SPS) model to simulate group dynamics in this type of communication service. We assume several behavioral patterns of a facilitator agent with fixed game-theoretical strategies and several movement strategies. We discuss how incorporating a single facilitator agent into the space can increase novel and cooperative interactions in several behavioral settings of the facilitator agent. We also report on a preliminary experiment on designing a facilitator agent using a deep reinforcement learning technique.

Niche construction is a process in which organisms modify the selection pressures on themselves and others through their ecological activities. While effects of niche construction on evolution have been discussed using simple theoretical models, not much is known about the evolution of complex and physically-grounded niche construction such as beaver dams. Our purpose is to clarify what conditions and what kind of complex and various niche-constructing behaviors evolve in physically-grounded environments. We focus on a predator-prey relationship because it is one of the most fundamental ecological relationships among species and had brought about a wide variety of anti-predator adaption. We constructed an evolutionary model in which a prey has to prevent itself from being captured by a predator through construction of a structure composed of objects in a 2D physically simulated environment using the LiquidFun. Moreover, we used a deep learning technique for further analysis of emerged adaptive structures. We show that there was a large diversity in the emerged adaptive structure in the case that the number of available resources was intermediate. It also turned out that there was a positive relationship between the number of available resources and the average fitness. The detailed analysis shows that there were three typical types of adaptive structures. A “shell strategy” encloses the whole body of a prey with many objects, while “barnacles strategy” encloses it by moving toward left and using objects and field tiles. A “wall strategy” creates a tall wall between a prey and a predator.

Researchers in the field of artificial life or complex systems grapple with abstract, complex and large data sets in general. We believe that virtual experience based on sonification (the transformation of data relations into perceived relations in an acoustic signal) has great potential to understand the dynamics in complex systems. We focus on the complex dynamics emerging in a social particle swarm model closely related to social psychology, and realize virtual experience of it by audio spatialization, in which major/minor chords, consonance/dissonance, and high/low pitches are utilized to represent the emotional meaning of the social relationships. The evaluation experiment shows that an experiencing person can grasp the social relationships and their dynamics in a two-dimensional space, including cooperation, exploitation and a cyclic process of formation and subsequent collapse of clusters, by feeling a bright, dark or unstable atmosphere created by interactions of sounds.

Inspired by the self-organization of growing embryos and coordinated movement of multicellular assemblies such as the slime mold Dictyostelium, where each cell is controlled by the same controller (a DNA-encoded gene regulatory network), we evolve distributed gait control mechanisms for soft-bodied animats. The animats are made of compressible material, with each body region capable of independent actuation, controlled by a cell at its center. Each animat consists of hundreds of cells uniformly distributed throughout the body, each sharing the same artificial gene regulatory network and aware of the state of their local neighborhood. We found that one of the most common actuation patterns that emerged relied on cells synchronizing their oscillations in order to produce a rotating, spiral wave spanning throughout the body. We found this type of mechanism to emerge for a wide range of animat morphologies as well as in very different types of initial conditions. We investigate how the evolved controllers produce the pattern through local feedbacks and evaluate spiral stability when imperfect, noisy cells are used.

Niche construction is a process in which organisms modify the selection pressure on themselves and others through their ecological activities. Various evolutionary models of effects of niche construction on evolution have revealed that they bring about unexpected evolutionary scenarios. However, little is still known about how niche-constructing behaviors of complex physical structures (such as nest-building) can emerge through the course of evolution, even though it is one of the most ubiquitous and significant niche-constructing behaviors. Our purpose is to obtain knowledge of the emergence and evolution of physically-grounded niche construction and the effect of its ecological inheritance on evolution. We construct an evolutionary model in which a virtual organism has to arrive at a goal by constructing a physical niche composed of objects in a physically simulated environment. In particular, we focus on effects of the degree of ecological inheritance, which is represented as a weathering probability of ecologically inherited objects from a parent to its offspring. We show that it has a nonlinear effect on the adaptivity of the population. In the case of no ecological inheritance, adaptive niche-constructing behaviors such as valley-filling or ramp-placing strategies emerged, which created complex structures composed of multiple objects. It also turned out that the stable ecological inheritance of constructed structures could increase the adaptivity of the population by allowing an organism to maintain the inherited and adaptive structures while the unstable ecological inheritance rather decreases the adaptivity of the population by making previously adaptive structures maladaptive obstacles.

People in some societies tend to put a greater value on equality in distribution of resources even if they have to pay expensive court costs to achieve it, while people in some other societies tend to aim at a maximal share (the whole) but withdraw readily if any conflict occurs. Nash demand game (NDG) is a one-shot two-player game and has been widely used for modeling such bargaining situations in computational and game theoretic approaches. Each player simultaneously demands a portion of some good. If the total amount demanded by the players is less or equal than available good, each player obtains the claimed request. Otherwise, neither player gets anything. Whereas the studies using NDG can account for why people favor the equal distribution, it is too simple to deal with various distributive norms. We use the demand-intensity game (DIG) which adds a psychological factor to NDG while maintaining such simplicity that it can be analyzed by the concepts and tools of the game theory. The goal of this study is to clarify the origin and evolutionary dynamics of distributive norms using DIG. Previous studies have shown that population structures tend to promote cooperative behavior by means of cooperative clustering and assortative interactions. We perform the evolutionary simulation focusing on the effect of the population structures on the evolution of distributive norms. We show a surprising result that network structures significantly change the evolutionary scenario. A population distributed over a regular network tends to evolve a strong equality norm. However, as the random links increase in the network, the more we see the scenario in which monopolists occupy the population who ask for the whole but with a moderate intensity. This result might offer significant implications to us living in a world where an increasing number of people are connected to each other through social networking. We also find that network structures with some intermediate randomness create an interesting scenario in which several norms emerge in a cyclic manner.

Doesnt include abstract