|
EPSRC Network on Evolvability in Biology & Software SystemsSoftware Evolution and Evolutionary Computation Symposium Abstracts
University of Hertfordshire, Hatfield, U.K.
|
ALEXANDRA PENN
Centre for Computational Neuroscience and Robotics,
University of Sussex, and
Future Technologies Group, British Telecom.
One of the most prominent mechanisms by which organisms have increased in complexity over evolutionary time is transitions in levels of selection [1], that is the construction of a new level of individuality from a collective of pre-existing individual organisms. For example, the origin of multicellularity, insect societies, or eukaryotes. This is clearly a means by which open-ended evolution is achieved in nature, and as such, is a dynamic which we would wish to emulate in artificial evolution. In trying to investigate these phenomena in an artificial context, one very quickly comes across an extremely basic problem to be addressed. What is the nature of the entity upon which selection acts? That is, what constitutes an individual? In simulation it is usually the case that the individual is a predefined, static entity with a prescribed set of parameters which can be varied, but no scope for real innovation. By contrast real biological individuals are dynamic physical processes in a constant state of flux, continually metabolising in order to simply exist. In simulation, the organism/environment boundary tends to be rigid and preset. Real organisms interact with their environments in a far more direct manner, and their nature and extent as a system are not predefined by a programmer, but arise from the physical and chemical interactions of their components with that which surrounds them. There is a potential richness in these interactions which is generally lacking in artificial evolution. Witness the sometimes unexpected results of real hardware evolution, in which the evolving circuits have been known to exploit their surroundings to satisfy the imposed fitness criteria. Oscillator circuits amplifying and outputting the radio signals from nearby PC monitors using their own circuitry as antennae for example [2]. The fact remains that within all this complexity of interaction an effective organism/environment distinction exists, it is possible to pick out a single organism from amidst the clutter, and describe its interactions. We note that such distinctions are context and scale dependant however, we imagine that an invading bacteria would ``perceive" the cells rather than the multicellular organism for instance. All of this means that the nature of the individual organism is non-obvious.
It would seem to be a neccessary first step in emulating the flexible nature of the individual over biological time to define what we mean by an individual, particularly within the constraints of the living system. Various structural criteria, applicable to both organic and inorganic structures, present themselves: The system possesses a reduced, distinct, set of higher-level system variables with respect to the number of degrees of freedom of its components (temperature, dimensions, metabolic set points etc); The system can be described as interacting with these variables, which we shall call characteristic; The variance of the characteristic variables is low compared to that of the lower-level variables. In addition to this, biological individuals are open, dissipative systems far from equlibrium. They maintain a low internal entropy by the processing of low entropy inputs with an autocatalytic metabolism. They are self-producing, defining their own boundaries. Most importantly, they are homeostatic. Their set points, or characteristic variables, are actively maintained over a wide range of variations in both internal and external conditions. The variables which are maintained are not externally imposed, but arise from the systems intrinsic dynamics. It is our proposition that it is the origin of homeostasis amongst a group of previously independant individuals which marks the point at which a transition has occured.
We intend to present the preliminary results of work in progress on measuring the degree of a system's individualilty based on our criteria of homeostatic maintainance of characteristic variables. Measurements of the informational entropy of the principal components of positional variables of members of interacting groups of simulated Kheperas are used as a metric for group coherance and give linear characteristic variables. The results of subjecting such individuals to perturbation and testing their ability to continue