Meta genetic behaviour emergence in evolutionary simulation
Guido Tascini
[Università Politecnica delle Marche]
ABSTRACT
How to study phenomena that involve great time scale, like the living being evolution? Computer Science gives us the opportunity of simulating by computer processes and arranging virtual experiments: these imitate natural evolution, and allow investigating processes that, in nature, take place over millions of years. In this case clearly defining the basic laws of transformations and/or evolutions of phenomena to study, appears fundamental. A well-established theory that gathers today the near total biologists is the current dominant explanation of Darwinism, the S.T.E. (Synthetic Theory of the Evolution). This theory is actually the base of research on genetic transformation of human beings, but it is also the reference for some studies on psyche sphere and on social evolution. There are two fundamental elements of the theory: 1) random genetic mutations, 2) sorting, by the natural selection, among those which are favourable to gene or the species (Dawkins 1990). The natural selection and adaptation involves the phenotype, that is the inventory of inherited tracts, and is an adaptation to demand of ecological situation. In practice adaptation stands for process in which the environmental variables select, among individuals in a population, those whose inheritable properties are the best-suited for survival and reproduction.
This theory, that uses a random selection, constitutes the Hard Darwinism, and has received critiques [Beerling 2007, Fodor 2007, Kirshener 2005, Margulis 2002, Mayr 1997, etc.], the most centred on its substantial finalism. The genes, the individuals or the species most suited for survival, depend on the favourable variations in natural selection. But even if it is a random genetic variation result, the mechanism may be considered a utilitarian and finalist design. Modifications are proposed, but the situation is still evolving. Therefore we are wondering if the scientific community is going toward a Week Evolution Theory. The Darwinism was recently integrated by several contributions: molecular biology (Monod 1970); molecular genetics; population genetics (Provine 1971); punctuated balances (Elredge, Gould 1972); neutral theory (Kimura 1990); genetic drifts; etc..The criticisms addressed to the theory may be summarized in the following observation: if it can account for the microevolution, either by phyletic gradualism, or by punctuated balances, it does not explain the macro one and the mega evolution. In practice the independence, outlined by the theory, between the genome and the cytoplasm are not guaranteed. In fact the cellular core permanently interacts with the cytoplasm. In the cell they are permanent exchanges of matter, energy and information, like it is shown by the Molecular (Genevès 1988) and cellular biology (Fain-Maurel 1991); in these exchanges take part all the cellular organoids, nuclear and cytoplasmatic: chromatin, mitochondries, nucleoles, Golgi apparatus, etc.. Moreover the fundamental concept of strictly random mutations is negated by several molecular genetics experiment and observations. For instance the colon bacilli may have an abnormally high mutations able to metabolize lactose in a stock unable to be nourished (Cairns, Overbaugh, Miller 1988); similar experiment is realised on the bacteria Escherichia coli with respect to salicin (Barry Hall, Rochester); mitochondrial D.N.A. and mitochondrial mutations existence was observed, for which it is hypothesized interactions between D.N.A. mitochondrial and nuclear D.N.A. at the mitosis final stage (telophase) (Allorge-Boiteau 1991); the transcriptase opposite transforms the R.N.A. of certain viruses in D.N.A.; etc.. There are some suggestion to introduce probability in the interaction [Borensztejn 2005, Rhodes 2005, Sanguinetti 2006, etc.] between the environmental evolution and the evolution of the organisms. The environment parameters may be various: chemical stimuli, like C, N, H, P, S, etc.; physical stimuli, like electromagnetic waves, sound and vibrations, temperatures, pheromones, etc.; ecological stimuli; pray-pray relations in the beasts; etc... The organism’s reaction to the environmental factors influence is complex, being the biosphere very complex, and they are located at the genome level, as well as at levels of molecular biology, embryogenesis and anatomical structures. The relation between environment parameters and organisms is of probabilistic type and integrates collective phenomena affecting simultaneously distant classes and phyla (Invertebrates and Vertebrates).
From this analysis the basic theory that we hypothesise for our simulation model is a Week Evolution Theory, in which an organism interacts with the environment in a complex way , and reacts to many stimuli of various nature: actually known and still to analyse or to discover. Now it is clear that the environment and its interaction with organisms is the key of the theory. The interaction may be probabilistic and the selection is not more blind, but depends on environment conditioning; and the environment conditioning and finalization is still largely to investigate.
A set of individuals may be viewed as complex system and then we can take care of emergences. Many individuals that evolve may give rise to unpredictable behaviours, that we call emergent. And it is clear that the emergence postulates an observer that sees the emergent behaviour visible at a higher level. In our case we have to hypothesize a level higher then genetic one were the mutations happen. Then if we hypothesize a stratification of the formal theory levels we can localize this emergence observation at a meta-genetic level.
In a context so outlined our model will need fitness function that drive the adaptation to the environment and that take care of interactions previous defined. In our experiments we will also hypothesize that the fitness function will take care of an environment feed-back on the list of individual to select for the survival.
Simulation of evolutionary process
A computer simulation may speed up the evolution process if the goals change continuously [ Kashtan 2007] Computer simulations that mimic natural evolution, allow to investigate processes that, in nature, take place over millions of years. We can simulate a population of digital genomes that evolves over time towards a given goal: to maximize fitness under certain conditions. Like living organisms, genomes that are better adapted to their environment may survive to the next generation or reproduce more prolifically. The work of Nadav Kashtan, Elad Noor, and Uri Alon suggests that varying environments might significantly contribute to the speed of natural and/or artificial evolution. Although the computer simulation is useful for studying theoretical questions of evolution, it may have some practical implications in engineering fields for systems design, and in computer science, for accelerating the optimization algorithms.
The computational model adopted in our experiment is inspired to Holland Model, including a feed-back of the environment on the individual choice. The simulation plans N strings that are random generated. Each string (genotype) is the binary code of a candidate solution (phenotype). At each genotype gi of initial population Pop(t=0 at time t=0, is associated a value of the fitness function ƒ i= ƒ (gi), that represents the ability of the individual to adapt itself to the environment. For detecting the adaptation value the fitness function receives in input a genotype, decodes it in the corresponding phenotype and checks it on the environment. After completion of the evaluation phase of individuals of the population Pop (t), at the time t, a new population Pop (t+1), at the time t+1, of N new candidate solutions is generated; this standard algorithm evolves neural network and structural model of RNA. The population of N individuals is initialized to random binary genomes of length B bits (random nucleotide sequences of length bases). They are several generations: in each generation, the S individuals with highest fitness (the elite) remain unchanged for the next generation. The individuals with least fit are replaced by a new copy of the elite individuals. As the non-elite individuals, pairs of genomes are recombined (with crossover probability Pc), and each genome is randomly mutated (with probability Pm per genome). A simulation runs until max. of fitness function ƒ i is achived for the goal.
Fair-unselfish modelIn general, systems that replicate need resources (energy, space,) for building copies of them. Resources are normally limited and, since each replicator tries to produce a maximum of copies, it will attempt to use resources to the limit. Then when several replicators are using the same resources, there will be competition or conflict. The more efficient replicator will gradually succeed in using more and more of the resources, and the less efficient one will succumb. In the long term, nothing will be leaved for the less fit one, and only the fittest will survive. The concept of ‘altruism’ is present in literature [ Heylighen 1992], and means that a system performs actions for increasing the fitness of another system using the same resources. On the other side selfishness characterizes a system performing actions that increase its own fitness. There is the opinion that, in natural selection, a system not only is selfish, since it try to optimize its own fitness, but it also tends to avoid altruism. Anyway by using the interaction with the environment a schema that brings the system to the solidarity may be the following one. Let a generic system constituted by individuals that may enter a series of resources Ri. Each individual can take the number of resources he wants, necessary for the survival. Being each resource organized in more sections (t), the individual may acquire, for each resource, a number of section m ≤ t to his liking. Since the environment interacts through the fitness function, this may penalize the individuals: this happens if, in section acquisition, they surpass a given number q
The individual selection could be realised, by following the previous policy, with the following formula:
ƒ = max { ∑i ( Ri[q]-Ri[x- q] ) }
Where Ri[j] represents the score for j sections of the Ri resource.
The environment policy, other then punish the not-fair acquisitions, rewards the solidarity: it adds a score for each group of Y sections leaved at other’s disposal, and this for each resource. If Di [y] is this reward for each resource Ri, the selection take place according to the following formula:
ƒ = max { ∑i ( Ri [q]-Ri [x- q] + Di [y] ) }.
This formula is iterated as many times as they are individuals to select for the survival on a group.
Figure 1 shows the results of our simulation. This operates on the interaction of two robots whose behaviour emerge from a simulation of various epochs, and many generations, on many (100) individuals, several (20) parents, and some (5) offspring; using as selection method ‘Elite’, and 1000 simulation steps for each trajectory. The simulation involves a neural structure that is the control of the robot, and runs as just described, including the fair-unselfish policy. The goal of the evolution is the adaptation to the environment: each robot attempts to explore the space, in search of resources for its nutrition, and so, after various epochs of simulation, converging toward a stationary path. The competition is realised with the presence in the same environment of two robots: the application of the fair-unselfish policy allow both the robots to converge toward a cooperative stationary paths.
Anyway the memory of individuals to select remains an important aspect to discover . This may contain inside logics like those just analysed. This fact leaves open a question: from where the stored information comes? But this constitutes object of further investigation. On the other side the analysis of the environment remains a serious aspect, and the questions open are: what the environment is? Can we consider the extension of the environment to the psyche? And other fundamental related questions. Finally the typologies of interaction between environment and genetic individuals have to be clarified.
