Representing a Solution as a Gene

• There are two tricks to making a GA work. The first is coming up with a good evaluation function.
• The second is to come up with a good representation to the problem as a gene.
• Some things are relatively obvious. How would you represent a sudoku puzzle? A series of 81 numbers 1-9s. Is there another way?
• How would you represent an agent in a video game?
• This of course depends on the problem.