I think there are lots of applications for gradient-free methods, but it's never been clear to me what the advantage is of the "genetic" aspect as compared to other randomized search strategies. However, the ability to dynamically grow a program tree is interesting and somewhat difficult to model in an ANN-style framework. But in gradient free methods, there are lots of interesting ideas like particle swarm optimisation and simulated annealing, or even just random search. It's not really clear to me what the "gene swapping" approach brings to the table, except that it probably encourages exploring the area in the middle between some good minima that are potentially far apart in the search space. But you could formulate this same idea without appealing to evolutionary metaphors.