A Neural Network to Recognize Cause and Effect in Game Events
Cole Hille / chille@bellarmine.edu / Faculty
Advisor: Nathan P. Johnson
Neural networks are often implemented for the purpose of
prediction, everything from future movement of stock prices to sunspot
activity; it is less common for a neural network design to extract the cause of
a given event from a domain of possible causes. It is, in fact, considered to
be an extremely difficult, contemporary problem in the field of Artificial
Intelligence where neural networks succeed at this task a relatively small
percentage of the time. This software project is an attempt to determine the
correct cause in a restricted domain of possible causes given a well-defined
effect. A legal gameboard acts as the input to a neural network, which was built
using Google’s TensorFlow libraries and the Python programming language, and
the game input commands that caused the move make up the output of the neural
net.