That idea is oversimplified. One can write a traveling salesman solver, that will outperform most humans in several weeks of work: any decent RL algo like A2C will do. Even a simple genetic algorithm will probably perform better actually, and that can be written as an exercise during a CS school lesson.