Sunday, October 19, 2008

Algorithm Optimization with Open Ended Evolution


I've completed a simple machine capable of open ended evolution. The system produces great complexity and functionality from initially non functional sub units. The code, analogous to information stored in DNA in biological lifeforms, fully defines the construction of the simple, Turing machine like assembler, which in turn determines the behavior of the automaton. A sufficient dose of random mutation per generation has the potential to drive an artificially induced "natural selection" process which when guided with purpose, causes the machine to adapt. These adaptations can either be an increase or decrease in overall complexity but nonetheless cause the system to be more suited for its environment. In a very abstract version of the experiment, I have "guided" the system to evolve from a machine which outputs meaningless numbers, into a machine optimized for generating sequential lists of prime numbers. My current future intentions are to expand this simple adaptive concept and combine it with simple pattern recognition algorithms (which themselves will be capable of limitless evolution) to create a demonstration for a high reliability, high accuracy path-finding algorithm which would have the potential to be used in applications such as automated automobiles and robotics.

No comments: