Modern Applications of AI in Games

Emulation of Old Games

Artificial intelligence is not only made for developing new games, but it can also be used for redesigning and placing older games onto modern systems. This process is called “emulation” and it is used to put older games that are normally inaccessible to the public onto a more modern system. This has become widely popular on PC systems allowing players to revisit their childhood games that wouldn’t otherwise be available. One example of this would be the emulation of the Atari 2600 called Stella. Stella is an emulator used to bring games from old systems such as Atari 2600 and Sega Genesis. The process of emulation is all possible via reinforcement learning. One of the methods used in this learning is the Arcade Learning Environment (Bellemare, 2013). This learning environment was built on the Stella emulator. This environment allowed researchers and others who were interested to add agents and use visual input such as screen pixels to produce an output. The researchers will give the agents specific instructions so the output is a more modern version of the game on modern systems.


Another implementation of AI in game development is the use of Super-Resolution. Super-Resolution is the process of increasing the resolution of an image from low to high. It is commonly used in surveillance such as security cameras for facial recognition and in the medical field to produce high-resolution pictures during medical examinations. Super-resolution itself has a couple of methods for how it increases and decreases the overall resolution. One game that uses super-resolution is God of War, which was released in 2016. In 2022, God of War received an update that boosted the game’s frame rate and graphics. The result of this is from the super-resolution software FediltyFX Super-Resolution 2.0. This software is designed to upscale the game as you play it. If a game engine runs at 1080p, FediltyFX will boost it to 1440p or 4K resolution (Klotz, 2022).

Super-Resolution Convolutional Neural Network (SRCNN) (Tsang, 2018) uses 3 layers: one for patch extraction, one for non-linear mapping, and the last for reconstruction. Before inputting the image data, the researcher must resize the image to what they want it to be in the end. One approach in super-resolution involves using a multi-layered neural network. The initial layer extracts patches from the input and applies filters to represent them. The second layer performs non-linear mapping, preserving the distances between data points while reducing the image’s dimensions. The final layer, the reconstruction layer, restores the image after all the processing is complete. The process involves intricate mathematical calculations, but it is only one of many applications of AI.