The genetic algorithm (GA) is an ideal tool to consider for design of complex experiments, because it is performing simultaneously the function of exploitation and exploration.
In GA parlance, the individuals are particular experiments, the genome consists of the specification of the experiment by a point 
in the experimental space, and previously described, the population at generation g is a set of P experiments,
.
Exploitation is achieved by making small mutations about the 
in a given generation, and exploration is achieved by making large mutations in the or by making crossovers. Euclidean distance may be used to define large and small.
The genetic algorithm has proven successful [1, and cf. "Vesicle optimization" below]. But the use of the GA proved to have some drawbacks, primarily:
(i) lack of control of mutation for an efficient tradeoff between exploration and exploitation, and
(ii) random variations to choose one generation from the previous prove not to take most efficient advantage of structure present in the previous.
Development of model-based algorithms have led to much more precise tools to exploit the structure present in experiments as they are performed.