Game theory guarantees that for this deterministic domain a perfect strategy must exist that will prevail over all others. This does not mean that such a solution can be computed by any general method.
Artificial Intelligence researchers have been trying for years to make programs build plans for their actions. A plan would predict all the situations that an agent will face and give each a solution. If not the best theoretical solution, this plan would constitute a partial one.
Plan building is questioned today as impractical for both computers and people. Intelligence seems to work reactively by reacting to different situations in specific ways while improvising -and improving- on the way.
To attempt learning in our domain, the search space has been drastically reduced: From the universe of all computer programs with 232 inputs and one output to a matrix of 32 values. This values code the reactions that a player will make to its sensory input. No attempt has been made to build plans. The perfect player will not be found, but good enough is possible.