A living creature can be regarded as a collection of a large number of sensors, where the system obtains information, maintains its basic metabolism, and changes its behavior according to its internal dynamics. We generated an au- tonomous sensor network system and implemented it as an sound art installation at a bookstore in Tokyo; we called the system Sound Bookshelf. The system was installed for two weeks in February 2012 in the open environment of a book- store, where people come and go from 10 am to 10 pm ev- eryday. We thus investigate a long term experiment of an au- tonomous sensor network. The obtained sensory information was translated into sound patterns and played using unidirec- tional ultrasonic speakers. An artificial chemical reaction as- sumed among the sensory information collected was used to change the sensor sampling frequencies. By walking around the store, visitors were able to experience a novel soundscape.
Using the idea of transfer entropy (TE), we study autonomy and information flow on theWeb. Its phenomena include user action patterns and the rich, autonomous network dynamics sustained by human searching/posting behavior. Such dy- namics show radically different behavior depending on the social context. It is widely accepted that Twitter messages - called ”tweets” - and Google search queries react strongly to significant social movements and accidents, producing bursts of patterns. We call this the reactive mode of theWeb. On the other hand, the Web has an intrinsic dynamics without burst- ing patterns. We call this the default mode of the Web. In this paper we study the default mode of the Web system, which we characterize via transfer entropy. The amount of information flow transferred between different sequences of queries as well as keyword frequencies is investigated. The default mode of the Web can then be characterized by the transfer entropy network dynamics amongst keywords. We use this idea of the default mode to install autonomy into generic artificial life systems.
Over the years, neural networks have become the most used models to study the brain. They also became increasingly popular for robotic control for their ease of use and their low computational demand. Despite 50 years of development, unsupervised learning remains a problem in neural networks, and simultaneously its mechanisms are not yet understood in the brain. The current theory is that memory is implemented through synaptic modifications based on the correlation of neuronal events, a process referred to as Hebbian learning. This synaptic plasticity has been observed in the brain but its relation to learning has always remained unclear. Over time, other non-synaptic plastic mechanisms have been uncovered at the neuronal level, acting in parallel with the synaptic plasticity. Those discoveries lead to the questions of what mechanisms are responsible for the learning capabilities of the brain and how the memories are stored in a population of neurons. Those are the questions we are interested in within this project. Our approach to these questions relies on the idea that the learning mechanisms must be studied within an ecological framework where the brain is studied within a body acting in its environment. As such, we use simulated robots controlled by models of the brain known as spiking neural networks and study the different possible mechanisms for learning within known cognitive tasks. We hope to understand how the observed neural mechanisms interact to produce learning and memory but also to propose new hypothesis about how learning can be achieved. For this reason, our study implements different types of plasticity combined with a Darwinian evolution process in order to gain new insights on the inner functioning of the brain.
The study of imitative behavior is currently a central topic in social, developmental, and comparative psychology, as well as in social neuroscience. It is widely accepted that imitation plays a significant role in social learning and enculturation, and that it serves as an inheritance mechanism for human-specific cumulative cultural evolution. One of the major challenges faced by explanations of imitation is the ‘correspondence problem’: How can one match one’s own bodily expression to the observed other’s bodily expression, especially in cases where there is no possibility of external self-observation? Current theories only consider the possibility of an innate or acquired matching mechanism belonging to an isolated individual. In this paper we evaluate an alternative that situates the explanation of imitation in the inter-individual dynamics of the interaction process itself. We implement and analyze a model of two embodied agents, which are engaged in mutual interaction. The agents can neither directly sense the configuration of their own body, nor even the configuration of the other’s body, and yet surprisingly they are nevertheless capable of bodily imitation. Analysis of the results provides a proof of concept that imitation can be enabled by a property of the collective dynamics, namely the relative stability of the interaction process. We then integrate this insight with considerations of the phenomenology of intersubjectivity in order to propose a parsimonious explanation of the apparently uneven distribution of imitation skills across ontogenetic and phylogenetic lines.
We constructed a self-sustaining machine that generates video imagery, consisting of a camera, an internal mental visual feed- back process, and a chaotic neural network with Hopfield structure and synaptic connections modified by Hebbian dynamics. Outside images and the internal visual feedback enter the neural network which in turn determines visual feedback parameters and camera control. The system does not always couple with the outside world and maintains its internal dynamics. Overall, a number of themes were investigated in this work. Self-organization: how an organism can sustain itself. Time- scales: internal structures occur at different time-scales, how are these organized and related? The subjective experience of time: how does it differ from Newtownian time and how does it manifest within an organism? Vision: how can visual phenomena manifest through feedback loops? Living systems organize their own time-scales driven by their memory structures - as a parallel, our system has multiple time-scales: the neural update timescale, the memory accumulation process, the Hebbian update, and the camera coupling rate.
Robots are useful tools for understanding cognitive mechanisms, for they are situated between living organisms and artificial systems. As embodied agents, they are embedded in a complex environment and act with their physical body, most of which are tough to be implemented in computer simulation. On the other hand, they give complete access to their internal states, which is not achieved with living organisms. Therefore, we use robots as a tool for understanding characteristics of natural organisms, especially focusing on robustness and internal simulation. Robustness is a property present in every living system which provides resilience against internal or external perturbations. Our work presented that using multiple kinds of sensors gives higher robustness than using a single sensor. For another topics, we created a robotic model for internal simulation, which is also an important notion to understand intelligence. It is said that, when rats are simulating internally in their head to evaluate several options, they move their head from one option to other. This head movement is called vicarious trial-and-error (VTE; Tolman, 1939), which we implemented in a robotic model to analyze the mechanisms and roles. We show that VTEs are generated from redundant sensory network, and robots which show VTEs exhibits a robust behavior.
The research investigates signal evolution in the context of agent-based simulations with a spatial distribution of food resources and agents varying in time. We look at how the concept of time can be used to adapt group behavior in the context of an agent based simulation with a spatial distribution of food and agents. The artificial agents evolve a signalling system that improves their ability to efficiently use the world’s resources in order to improve their fitness. One idea is to use a minimalist simulation model to demonstrate that learning about the notion of time facilitates efficient group foraging behavior, compared to environments that exclude the possibility to learn about the concept of time. The notion of time is embedded into agent signals, instantiating the communication of distances to resources as well as defining cyclic resource growth periods.