The Naive Solution
First, you're going to need a way to represent the state of the board and a stack to store all of the states. At each step, make a copy of the board, changed to the new state. Compare that state to all states of the board you've encountered so far. If you haven't seen it, push that state on top of the stack and move on to the next move. If you have seen it, try the next move. Each level will have to try all possible 64 moves before popping the state from the stack (backtracking). You will want to use recursion to manage the state of the next move to check.
There are at most 264 possible board configurations, meaning you could potentially go on a very long chain of unique states and still run out of memory. (For reference, 1 GB is 230 bytes and you need a minimum of 8 bytes to store the board configuration) This algorithm is not likely to terminate in the lifetime of the known universe.
You need to do something clever to reduce your search space...
Greedy-First Search
You can do better by searching the states that are closest to the solved configuration first. At each step, sort the possible moves in order from most lights off to least lights off. Iterate in that order. This should work reasonably well but is not guaranteed to get the optimal solution.
Not All Lights-Out Puzzles Are Solvable
No matter what algorithm you use, there may not be a solution, meaning you might search forever (or several trillion years, at least) without finding a solution.
You will want to check the board for solvability (which is a much faster algorithm as it turns out) before wasting any time trying to find a solution.