Written by Artificial Intelligence, Evolved by Genetic Algorithms: An Evolutionary Challenge to Solving the Traveling Salesperson Problem with ChatGPT
Tae-Hun Kim, Yong-Hyuk Kim · Proceedings of the Genetic and Evolutionary Computation Conference Companion · 2025
Artificial Intelligence has recently made rapid advancements, blurring the boundaries of intelligence between humans and machines. Its characteristic of generating different answers to the same question often evokes the thought process of humans. This study explores how AI-generated code, with minimal human intervention, can improve performance. There are over 100 algorithms capable of solving an NP-hard problem, the traveling salesperson problem, and it is difficult for one person to know all of them. Therefore, it is nearly impossible for a human to manually combine the advantages of various algorithms into a hybrid code. However, by leveraging AI, even someone who has no knowledge of the algorithms can generate a hybrid code that combines the strengths of different algorithms. In this experiment, AI-generated codes based on multiple algorithms are applied to genetic algorithms. The performance of hybrid codes, after undergoing crossover and mutation, was evaluated. The results showed that the performance of the AI-generated hybrid code surpassed that of the best-performing single algorithm code. This study demonstrates that AI can go beyond the mere application of existing algorithms by autonomously generating code and combining different algorithms to implement hybrid optimization techniques.