Genetic programming uses methods which are similar to genetic algorithms (GA), but is based on programs which perform tasks whose results can then be evaluated to deliver a fitness function similar to GAs. Instead of using pools of parameter lists to be evaluated by some evaluation procedure, GP uses pools of programs which are to be run to perform the required task. A technical difference between GAs and GPs is that GAs use list structures, often of fixed size, to store their data, while GPs use tree structures which can vary in size and shape for each program used in the program pools.
The application of a tree representation (and required genetic operators) for using genetic algorithms to generate programs was first described in 1985 by Cramer. Koza, though he did not invent genetic programming, is indisputably the field's most prolific and persuasive author.
So far GPs have successfully solved some toy problems, such as the lawn mower problem, but the method is very computationally intensive, and may not compare favourably where simpler methods, such as genetic algorithms or random optimisation can be used instead. It is possible that some more complex problems may be more amenable to solution using GPs than other optimization methods.
Unfortunately, due to the lack of solid theory regarding the performance of genetic programming vs. traditional search methods (such as hill-climbing), genetic programming remains a sort of pariah amongst the various techniques of search. While genetic programming has achieved results that are as good as and sometimes better than human-generated results, more work needs to be done on the theory in order to bring the technique into more widespread use.