The Memetic Automaton

Abstract

Real-world problems of interest seldom exist in isolation. Thus, we humans routinely resort to exploiting pre-existing ideas, either of our own, and/or those gleaned from others, whenever faced with a never before seen challenge or task. It is these building-blocks of knowledge, that reside in our brains, that were first referred to as “memes” by Richard Dawkins in his 1976 book The Selfish Gene. Incidentally, in the present-day, a perennial source of rich and diverse memes, infiltrating all aspects of human and industrial activity, happens to be the internet. Despite the growing ubiquity of this technology, and its known association with the memetics concept (as evidenced by the spread of so-called “internet memes”), it is striking that most computational systems, including optimization engines, continue to adhere to a tabula rasa-style approach of tackling problems from scratch. In contrast to humans, their capabilities do not grow with experience. This holds true even for the (admittedly limited) algorithmic realizations of memetics in earlier chapters of the book, where discussions were focused on hybrid optimizers in which memes merely served a complementary role in the “lifetime learning” phase of an evolutionary cycle. What is more, even the simultaneous problem learning and optimization strategies in Chap. 3 offered only a partial glimpse of what comprehensive memetic computation (MC) can achieve in practice, as the learning was restricted to datasets originating from a single problem at a time; with little scope for information transfers across distinct optimization exercises. Thus, in order to bring MC closer to human-like problem-solving prowess, in this chapter, we put forward the novel concept of memetic automatons.