Genetic algorithms are essentially probabilistic.  In contrast, traditional optimization methods are mostly deterministic.  The probabilistic nature of evolutionary computations allows them to explore areas in the search space that appear impractical at first glance.  Bad solutions (individuals) are not thrown out from the population.  Instead, they have some finite probability of mating and of giving future generations some genetic features that could be very useful in creating true elite offspring.  Thus, the genetic algorithm avoids local optima and can find a true global solution of the problem.

The deterministic character of the traditional optimization techniques is the main reason they sometimes settle on a local optimum.  Sometimes, the elite individuals in a genetic algorithm population may also get stuck at the local optimum.  However, the genetic algorithm always has a good chance of escaping this local optimum because of the mutation and/or diversity operators.