Ali Hafez/Projects/

For my final project in Computational Intelligence for Games with James Glenn, I used Monte Carlo Tree Search to create an AI to play the popular game 2048. The algorithm depends on how much time (computing time) you give per calculated move, and at 0.1 seconds per move, the AI could get to 2048 (thus beating the game) almost 50% of the time.

Monte Carlo Tree Search is an algorithm that uses markov chains (in the form of decision trees, that branch off per possible move at each game state) to explore possible outcomes following each decision and determine the best one. In the case of 2048, the state was simply the board, the decisions were the directions to move in, and the "best" outcome was determined with heuristics including total score and highest value square. In the project, I had to code up the entire game in C++ as well as the Monte Carlo algorithm and heuristics.