Semiosis in Dynamic Cognition of Environment

To understand the behavior of a living organism, it is essential to elucidate the organic behavior of aggregated as well as single elements of cells. In a life system, there is inherent diversity at the individual level, but at the group level, there is also an inherent mechanism that becomes increasingly more stable and deterministic. It is an extremely important challenge to elucidate the dynamics governing the process in which a macroscopic order emerges and disintegrates controlling the degrees of freedom in such an organization or group consisting of multiple elements. Such knowledge may be utilized for clarifying a process of semiosis with respect to bio-mechanical systems and human-robot collaborative systems.
The group of Tomita (Kyoto University) studies medical engineering focusing dynamics of organisms and their environments in terms of hierarchy in a biological system in the context of interactions among cells and between cells and tissues. In an initial experiment to observe the structure and function of cartilage cells and tissues without imposing any dynamic environment conditions, ES cells did not differentiate into cartilage cells; however, the group aims to confirm that, when a dynamic state of the environment is altered, morphology and function are enhanced to organize and adapt to the environmental changes. To investigate the dynamics of this system further, specifically the functional and structural adaptations that constituent units undergo, the group has developed an ES cell-cartilage regeneration simulation model using cellular automata. Based on the experimental analysis of those bio-living cells, this group presents a novel idea of “Bio-Environmental Designing” combining two different methodologies of the environmental modification and function designing.
The semiotic study of living systems is very important to clarify the essential mechanism of emergence of high adaptability under the changing environment. Among various neuronal activity patterns, a synchronous activity associated with behavior and cognition has been observed in many neuronal systems. The group of Aoyagi (Kyoto University) has demonstrated that a network organized under spike timing dependent plasticity (STDP) is capable of encoding a systematic transition behavior among the memorized patterns of the external stimulus, and thus a kind of temporal signs on causal relationships among experienced events can be formed within the actor using this neuronal model. This research is now extended to the developments of brain machine interfaces and artificially cultured neural systems, to both of which the issues of symbol coding within the natural systems and of its usage are fundamentally important.
Apart from the above living cells and/or neurons, to many of the biological beings, complexities of the environments they encounter are quite varied depending upon what embodied interactions are allowed and upon the quality of coordination they need for their survival. For instance, the significance environment to insects are quite simple as compared to human, since it is genetically determined what external stimuli to perceive and how to react to that [3]. On the other hand, more social primate beings have to recognize the other’s intentions reflected within the environment in order to make efficient cooperation with their partners; they learn a variety of powerful social rules which minimize interference and maximize their benefit.
What distinguishes a human ability from other lower biological beings is that a human can build up structures by interacting with the unstructured. Through interactions with the surrounding environment he/she can selectively find some cues that are meaningful to their survivals and structurize them into some ordered internal constructs. This is realized by adaptively rebuilding their internal constructs (i.e., representations) on the objects in use as they accumulate the experiences. Wherein, the objects that a human has to use are becoming more complex and dynamical, thus their behaviors are getting less predictive especially for the novice users.
According to this perspective, the group of Taniguchi (Ritsumeikan University) and Sawaragi (Kyoto University) is focusing on the issue of human-automation collaboration, where a human driver has to correctly recognize a working status of modes that are embedded within an automated vehicle (i.e., design of signs). A computational model of a process of recognizing multiple modes is built up based upon the findings obtained in the fields of brain neurology. Then, the methodology for analyzing the safe usability of the designed artifact is presented that can detect the error-prone logics of the automation inducing the human user’s wrong mode-recognition in use (i.e., performance of signs). The final goal of this work is to find out the fundamental design principles of artifacts that can socially behave and collaborate with a human.
In a general multi-agent environment, an agent should adapt to the diversities of dynamics that are perceived as changes in physical properties of the task environment, from which an agent should be able to recognize a partner’s status and shifts of intentions. Not only to recognize the shifts of the partner’s intentions, a social agent should be able to explore how to collaborate with the partner’s shifts of intentions adaptively. Herein, we do not assume any explicit communication such as verbal commands can be exchanged; they have to interpret the perceived physical cues and have to infer and identify the invisible intentions of the other. None of such a common code table is shared among them as Shannon-Weaver’s classical communication model. Actually for the social robots that perform social interactions with people using body motions, motion learning is an important technical issue for a robot to enhance its autonomy by adaptively organizing its pre-existing internal structures and to elicit human responses. However, true social behavior in robots is probably not possible, given the limitations on abilities to construct and use an objective external environment model to forecast accurately the behavior of other people. Through its interaction with others and its internalization, robots define a new reality, then constantly change and optimize their behavior. The group of Taniguchi (Ritsumeikan University) and Sawaragi (Kyoto University) has proposed a communication model among the two agents who seeks for a common goal and collaborate without any explicit communication but only though their enclosed proactive interpretive efforts on the change of the observed environment.
Generally in a multi-agent environment, an agent should adapt to the diversities of dynamics that are caused by the changes in physical properties of the task environment and in social situations concerning how the other agents behave and change their way of behaviors. Owing to the socially dynamical property of multi-agent system, it is difficult to achieve multi-agent reinforcement learning tasks. The group of Taniguchi (Ritsumeikan University) and Sawaragi (Kyoto University) presents a method for multi-agent reinforcement  learning to generate individual rewards,

which are provided to worker agents from a manager agent. The explicit generation of rational divisions of the roles can change a POMDP (partially observable Markov decision problem) task for each agent into a simple MDP task under certain conditions, thus the role division in the organization is related to the role-differentiation process.