Let's first build a simple neural network agent to play the game Pac-Man. We will create an agent with a set of random weights and biases. These agents will then try to play the game; we select the agent that is able to play for the longest average duration, assuming theirs to be the best policy.