Learn to Play

Dr. Tom Murphy from CMU once wrote an AI program capable of playing the game Super Mario Bros. The `algorithm he used is a type of machine learning. You can search for and find the full version of the video on various video platforms. If you’re interested in seeing how AI can learn to play classic games through machine learning techniques, checking out Dr. Murphy’s work would be quite enlightening. His project showcases the potential of AI in understanding and mastering complex environments, similar to those found in video games.

After watching his video, you may keenly observe that the AI’s gameplay process mirrors our own initial experiences. Just as we would when first entering a game world, we tried pressing every button on the controller, observing the corresponding reactions from the game, by experimenting with actions and observing the outcomes, which helps to build a mental model of the game’s rules and possibilities.

The interaction between player and game is the core component of the video game, just like a dialogue between designer and player across the void. The designer must instruct the player on how to engage with the game, while the player, guided by these cues, must think, decide, and act, working towards achieving the game’s objectives.

This communication is essential for creating an engaging and immersive experience, and the player feels both challenged and supported, learning the game’s rules and mechanics through active participation.

Interestingly, this communication is not through words, but through the feedback of the game system. That’s why we should look into behaviorist psychology and how it works in games. Let’s get started.