AI Programmer: Autonomously Creating Software Programs Using Genetic Algorithms

Abstract

In this paper, we present the first-of-its-kind machine learning (ML) system, called AI Programmer, that can automatically generate full software programs requiring only minimal human guidance. At its core, AI Programmer uses genetic algorithms (GA) coupled with a tightly constrained programming language that minimizes the overhead of its ML search space. Part of AI Programmer's novelty stems from (i) its unique system design, including an embedded, hand-crafted interpreter for efficiency and security and (ii) its augmentation of GAs to include instruction-gene randomization bindings and programming language-specific genome construction and elimination techniques. We provide a detailed examination of AI Programmer's system design, several examples detailing how the system works, and experimental data demonstrating its software generation capabilities and performance using only mainstream CPUs.

Neural Program Synthesis with Priority Queue Training

“…• Genetic algorithm (GA) implemented based on Becker & Gottschlich (2017). 2 See Appendix 6.3 for more details regarding our implementation.…”

“…Our use of BF programming language enables a comparison between our technique and a concurrent work by Becker & Gottschlich (2017) on the use of genetic algorithms for program synthesis in the BF language. However, our method for program synthesis has important benefits over Becker & Gottschlich (2017) including better performance and the potential for transfer learning, which is possible with neural networks (Johnson et al, 2016). We also make the observation that PQT alone is stable and effective, without needing to use PG.…”

“…We then sample new programs from the RNN, update the queue, and iterate. We also compare against a genetic algorithm (GA) baseline which has been shown to generate BF programs Becker & Gottschlich (2017). Surprisingly, we find that the PQT approach significantly outperforms the GA and PG methods.…”

“…• We propose a learning framework for program synthesis where only a reward function is required during training (the ground-truth programs or correct outputs are not needed). Further, we advocate to use a simple and expressive programming language, BF, as a benchmark environment for program synthesis (see also Becker & Gottschlich (2017)).…”

Why We Do Not Evolve Software? Analysis of Evolutionary Algorithms

“… "We will (before long) be able to run genetic algorithms on computers that are sufficiently fast to recreate on a human timescale the same amount of cumulative optimization power that the relevant processes of natural selection instantiated throughout our evolutionary past … " [39]  "As computational devices improve in speed, larger problem spaces can be searched." [40].  "We believe that in about fifty years' time it will be possible to program computers by means of evolution.…”