A spatial-temporal agent based model with linear, genetically programmed agents competing and reproducing within the model results in implicit, endogenous objective functions and selection algorithms based on "natural selection". This implicit optimization of genetic programs is explored by application to an artificial foraging ecosystem. Limited computational resources of program memory and execution time emulate real-time and concurrent properties of physical and biological systems. Relative fitness of the agents' programs and efficiency of the resultant populations as functions of these computational resources are measured and compared. Surprising solutions for some configurations provide an unique opportunity to experimentally support neutral code bloating hypotheses. This implicit, endogenous, evolutionary optimization of genetically programmed agents is consistent with biological systems and is shown to be effective in both exploring the solution space and discovering fit, efficient, and novel solutions.

翻译：基于空间-时间智能体模型，在线性遗传编程智能体竞争与繁殖过程中，产生了基于“自然选择”的内隐式内源目标函数与选择算法。通过应用于人工觅食生态系统，探究了遗传程序的内隐优化过程。程序内存与执行时间的有限计算资源模拟了物理与生物系统的实时并发特性。测量并比较了智能体程序的相对适应度以及作为这些计算资源函数的结果种群效率。针对某些配置的意外解决方案为实验支持中性代码膨胀假说提供了独特契机。这种遗传编程智能体的内隐式内源进化优化与生物系统一致，并被证明在探索解空间以及发现具有适应性、高效性与新颖性的解决方案方面均有效。