True, genetic algorithms are only implied, but those implied gradients are used in the more successful evolutionary strategies. So while they might not look like it (because it's not used in a continuous descent) they still very much work like (although they represent a smoother function than) regular back-prop gradients when aggregated.