The prevalence of artificial intelligence (AI) tools that filter the information given to internet users, such as recommender systems and diverse personalizers, may be creating troubling long-term side effects to the obvious short-term conveniences. Many worry that these automated influencers can subtly and unwittingly nudge individuals toward conformity, thereby (somewhat paradoxically) restricting the choices of each agent and/or the population as a whole. In its various guises, this problem has labels such as filter bubble, echo chamber, and personalization polarization. One key danger of diversity reduction is that it plays into the hands of a cadre of self-interested online actors who can leverage conformity to more easily predict and then control users' sentiments and behaviors, often in the direction of increased conformity and even greater ease of control. This emerging positive feedback loop and the compliance that fuels it are the focal points of this article, which presents several simple, abstract, agent-based models of both peer-to-peer and AI-to-user influence. One of these AI systems functions as a collaborative filter, whereas the other represents an actor the influential power of which derives directly from its ability to predict user behavior. Many versions of the model, with assorted parameter settings, display emergent polarization or universal convergence, but collaborative filtering exerts a weaker homogenizing force than expected. In addition, the combination of basic agents and a self-interested AI predictor yields an emergent positive feedback that can drive the agent population to complete conformity.

Neural Baldwinism concerns the Baldwin Effect in the evolution of brains and intelligence. The first phase of the Baldwin Effect (B.E.), wherein plasticity provides a selective advantage, is intuitive and commonplace in simulations of adaptive systems. However, the second (assimilation) phase often poses problems for Baldwinism in general, and this is particularly acute for biological neural networks, where a complex developmental process greatly confounds the mapping from genotype to functional phenotype: a brain whose synapses are tuned to perform particular tasks. Since a strong genotype-phenotype correlation is often viewed as a prerequisite to this second phase, the body’s most plastic organ would appear to defy Baldwinism. However, a detailed examination of 3 key processes of neural adaptation blurs the distinction between classic developmental and learning stages of brain maturation, thus supporting a re-interpretation of Neural Baldwinism’s phase II as a heterochronous shift of the bulk of these three adaptive processes from postnatal to prenatal stages. This article illustrates Heterochronous Neural Baldwinism (HNB) with artificial neural networks that evolve, develop and learn, and in which some degree of synaptic tuning shifts to the prenatal stage.

The Baldwin Effect is a very plausible, but unproven, biological theory concerning the power of learning to accelerate evolution. Simple computational models in the 1980's gave the first constructive proof of its potential existence, and subsequent work in evolutionary computation has shown the practical, computational, advantages of hybrid evolution-learning systems. However, the basic theory, particularly its second phase (involving genetic assimilation of acquired characteristics) is difficult to reconcile in systems controlled by neural networks, particularly those that arise from their genotypes via a complex developmental process. Our research uses new evidence of the blurred distinction between development and learning in natural neural systems as the basis for an abstract model displaying the Baldwin Effect in artificial neural networks that evolve, develop and learn.

Many neuroscientists view prediction as one of the core brain functions. However, there is little consensus as to the exact nature of predictive information and processes, or the neural mechanisms that realise them. This paper reviews a host of neural models believed to underlie the learning and deployment of predictive knowledge in a variety of brain regions: neocortex, hippocampus, thalamus, basal ganglia and cerebellum. These are compared and contrasted in order to codify a few basic aspects of neural circuitry and dynamics that appear to be the heart of prediction.

This work investigates an intermediate abstraction level, that of neural groups, for modelling the development of complex artificial neural networks. Based on Neural Darwinism \cite{edelman:2000}, Displacement Theory \cite{deacon:1998} and The Neuromeric Model \cite{striedter:2005}, our DEACANN system avoids the complexities of axonal and dendritic growth while maintaining key aspects of cell signalling, competition and cooperation that appear to govern the formation of neural topologies in nature. DEACANN also includes a genetic-algorithm for evolving developmental recipes, and the mature networks employ several forms of learning.

While classic artificial intelligence systems still struggle to incorporate commonsense knowledge properly, situated and embodied artificial intelligence (SEAI) aims to build animats that acquire a common-sense understanding of the world via interactions between simulated brains, bodies and environments. Neuroscientists believe that much of this common sense involves predictive models for physical activities, but the transfer of sensorimotor skill knowledge to cognition is non-trivial, indicating that SEAI may meet a daunting challenge of its own. This paper considers the neurological bases for implicit procedural and explicit declarative common sense, and the possibilities for its transfer from the former to the latter. This helps assess the prospects for SEAI eventually to surpass GOFAI (good old-fashioned AI) in the quest for generally intelligent systems.

To increase the evolvability of larger search spaces, several indirect encoding strategies have been proposed. Among these, multicellular developmental systems are believed to offer great potential for the evolution of general, scalable, and self-repairing organisms. We reinforce this view, presenting the results achieved by such a model and comparing it against direct encoding. Extra effort has been made to make this comparison both general and meaningful. Embryonal stages, a generic method showing increased evolvability and applicable to any developmental model, are introduced. Development with embryonal stages implements what we refer to as direct neutral complexification: direct genotype complexification by neutral duplication of expressed genes. The results show that, even for high-complexity evolutionary targets, the developmental model proves more scalable. The model also shows emergent self-repair, which is used to produce highly resilient organisms.

While classic AI systems still struggle to properly incorporate common-sense knowledge, Situated and Embodied Artificial Intelligence (SEAI) aims to build animats that acquire a common-sense understanding of the world via interactions between simulated brains, bodies and environments. Neuroscientists believe that much of this common sense involves predictive models for physical activities, but the transfer of sensorimotor skill knowledge to cognition is non-trivial, indicating that SEAI may meet a daunting challenge of its own. This paper considers the neurological basis for procedural common sense and the possibilities for its transfer to conscious reasoning. This helps assess the prospects for SEAI to eventually surpass classic AI in the quest for generally intelligence systems.

Baldwin's classic hypothesis states that behavioral plasticity can speed evolution by (a) smoothing the fitness landscape and (b) indirect genetic assimilation of acquired characteristics. This latter phase demands a strong correlation between genotype and phenotype space. But the natural world shows signs of this correlation at only a very coarse level, since the intervening developmental process greatly complicates the mapping from genetics to physiology and ethology. Hence, development appears to preclude a strong Baldwin effect. However, by adding a simple developmental mechanism to Hinton and Nowlan's classic model of the Baldwin effect, and by allowing evolution to determine the proper balance between direct and indirect mapping of genome to phenotype, this research reveals several different effects of development on the Baldwin effect, some promoting and others inhibiting. Perhaps the most interesting result is an evolved cooperation between direct blueprints and indirect developmental recipes in searching for unstructured and partially structured target patterns in large, needle-in-the-haystack fitness landscapes.

This work continues investigation into Gaia theory [Lovelock, (1995) The ages of Gaia, Oxford University Press] from an artificial life perspective [Downing, (2000) in Proceedings of the 7th International Conference on Artificial Life, (pp. 90-99) MIT Press], with the aim of assessing the general compatibility of emergent distributed environmental control with conventional natural selection. Our earlier system, GUILD [Downing and Zvirinsky, (1999) Artificial Life, 5, 291-318], displayed emergent regulation of the chemical environment by a population of metabolizing agents, but the chemical model underlying those results was trivial, essentially admitting all possible reactions at a single energy cost. The new model, METAMIC, utilizes abstract chemistries that are both (a) constrained to a small set of legal reactions, and (b) grounded in basic fundamental relationships between energy, entropy, and biomass synthesis/breakdown. To explore the general phenomena of emergent homeostasis, we generate 100 different chemistries and use each as the basis for several METAMIC runs, as part of a Gaia hunt. This search discovers 20 chemistries that support microbial populations capable of regulating a physical environmental factor within their growth-optimal range, despite the extra metabolic cost. Case studies from the Gaia hunt illustrate a few simple mechanisms by which real biota might exploit the underlying chemistry to achieve some control over their physical environment. Although these results shed little light on the question of Gaia on Earth, they support the possibility of emergent environmental control at the microcosmic level.

Gaia theory, which states that organisms both affect and regulate their environment, poses an interesting problem to Neo-Darwinian evolutionary biologists and provides an exciting set of phenomena for artificial-life investigation. The key challenge is to explain the emergence of biotic communities that are capable, via their implicit coordination, of regulating large-scale biogeochemical factors such as the temperature and chemical composition of the biosphere, but to assume no evolutionary mechanisms beyond contemporary natural selection. Along with providing an introduction to Gaia theory, this article presents simulations of Gaian emergence based on an artificial-life model involving genetic algorithms and guilds of simple metabolizing agents. In these simulations, resource competition leads to guild diversity; the ensemble of guilds then manifests life-sustaining nutrient recycling and exerts distributed control over environmental nutrient ratios. These results illustrate that standard individual-based natural selection is sufficient to explain Gaian self-organization, and they help clarify the relationships between two key metrics of Gaian activity: recycling and regulation.

In the spirit of contemporary artificial life research, EUZONE provides a virtual laboratory for the emergence of complex ecosystems from simple primitives. However, whereas most alife systems abstract away many real-world environmental constraints, EUZONE employs detailed physical and chemical models in combination with evolutionary algorithms to support the emergence of carbon-based aquatic ecosystems. With an emphasis on planktonlike organisms, this research focuses upon the self-organization and evolution of (a) lower levels of aquatic food webs, (b) Gaian interactions between primitive organisms and their physical environments, and (c) species interactions governed by varying life-history strategies.