A chemical genetic algorithm (CGA) in which several types of molecules (information units) react with each other in a cell is proposed. Not only the information in DNA, but also smaller molecules responsible for the transcription and translation of DNA into amino acids, are adaptively changed during evolution, which optimizes the fundamental mapping from binary substrings in DNA (genotype) to real values for a parameter set (phenotype). Through the struggle between cells containing a DNA unit and small molecular units, the codes (DNA) and the interpreter (the small molecular units) coevolve, and a specific output function, from which a cell's fitness is evaluated, is optimized. To demonstrate the effectiveness of the CGA, it is applied to a set of variable-separable and variable-inseparable problems, and it is shown that the CGA can robustly solve a wide range of optimization problems regardless of their fitness characteristics. To ascertain the optimization of the genotypeto-phenotype mapping by the CGA, we also conduct analytical experiments for some problems while observing the basin size of a global optimum solution in the binary genotype space. The results show that the CGA effectively augments the basin size, makes it easier for evolution to find a path to the global optimum solution, and enhances the GA's evolvability during evolution.

In order for an artificial life (Alife) system to evolve complex creatures, an artificial environment prepared by a designer has to satisfy several conditions. To clarify this requirement, we first assume that an artificial environment implemented in the computational medium is composed of an information space in which elementary symbols move around and react with each other according to human-prepared elementary rules. As fundamental properties of these factors (space, symbols, transportation, and reaction), we present ten criteria from a comparison with the biochemical reaction space in the real world. Then, in the latter half of the article, we take several computational Alife systems one by one, and assess them in terms of the proposed criteria. The assessment can be used not only for improving previous Alife systems but also for devising new Alife models in which complex forms of artificial creatures can be expected to evolve.

An evolutionary core memory system named SeMar is revised so that the entire action of the core might proceed by parallel execution of Proteins. The Proteins created by the transcription of DNA units work as operators and accomplish the corresponding instructions written in DNA units. In the experiment, it is shown that after inoculation of a man-made self-reproducing creature into the core, its offspring evolve through mutational modifications and the efficiency of the replication rate of DNA increases by the polyploidization of DNA sequences.

A novel machine language genetic programming system that uses one-dimensional core memories is proposed and simulated. The core is compared to a biochemical reaction space, and in imitation of biological molecules, four types of data words (Membrane, Pure data, Operator, and Instruction) are prepared in the core. A program is represented by a sequence of Instructions. During execution of the core, Instructions are transcribed into corresponding Operators, and Operators modify, create, or transfer Pure data. The core is hierarchically partitioned into sections by the Membrane data, and the data transfer between sections by special channel Operators constitutes a tree data-flow structure among sections in the core. In the experiment, genetic algorithms are used to modify program information. A simple machine learning problem is prepared for the environment data set of the creatures (programs), and the fitness value of a creature is calculated from the Pure data excreted by the creature. Breeding of programs that can output the predefined answer is successfully carried out. Several future plans to extend this system are also discussed.

A novel system composed of multiple von Neumann computers and an appropriate problem environment is proposed and simulated. Each computer has a memory to store the machine instruction program, and when a program is executed, a series of machine codes in the memory is sequentially decoded, leading to register operations in the central processing unit (CPU). By means of these operations, the computer not only can handle its generally used registers but also can read and write the environmental database. Simulation is driven by genetic algorithms (GAs) performed on the population of program memories. Mutation and crossover create program diversity in the memory, and selection facilitates the reproduction of appropriate programs. Through these evolutionary operations, advantageous combinations of machine codes are created and fixed in the population one by one, and the higher function, which enables the computer to calculate an appropriate number from the environment, finally emerges in the program memory. In the latter half of the article, the performance of GAs on this system is studied. Under different sets of parameters, the evolutionary speed, which is determined by the time until the domination of the final program, is examined and the conditions for faster evolution are clarified. At an intermediate mutation rate and at an intermediate population size, crossover helps create novel advantageous sets of machine codes and evidently accelerates optimization by GAs.