Posts

Showing posts from July, 2018

Zero knowledge AI for Tic-tac-toe and Gomoku (Five in a row) games

Image
Starting from zero When building AI for games, zero knowledge simply means that when the system starts learning the game it only understands what it means to move, win, or lose. In certain situations, it is certainly easier and faster to simply instruct the machine how to play instead of building a system that can infer knowledge from a gameplay: for simple or very fixed problems, like Tic-tac-toe, an algorithmic approach can be an excellent solution. However, as the game complexity increases it is not always easy to program the system how to play - sometimes advanced players can employ different strategies and beat the computer all the time. In these situations, a self-learning system might be a better approach. Artificial neural networks Neural networks are great to recognise complex patterns and build predictions based on these patterns, although it's worth keeping in mind that they aren't the best solution to every problem: as the "knowledge" is stored in