This artificial intelligence is able to recreate video game engines just by watching two minutes of the game
Ok, maybe artificial intelligence does not replace the posts of video game developers. This is an interesting project that shows us the ability of an AI platform in games recreation tasks, this without even having to look at the code of that game.
It is a project of Georgia Institute of Technology, who developed an artificial intelligence system capable of studying each of the frames of a 2D video game, and with this build a complete replica of the engine of that video game just by looking at two minutes of the gameplay.
The goal: to lighten the workload
This is a research project that seeks to demonstrate how artificial intelligence could be beneficial in certain industries by lightening the workload. For example, within video games, a platform could be trained to show different game styles based on a small sample, with which developers could spend more time on other factors of greater complexity.
In the case of this project, the researchers trained an artificial intelligence with two videos of less than two minutes. One showed a player in "explorer" mode and the other player had a "speedrunner" style, both playing 'Super Mario Bros'.
)
After this, the system was able to create the gameplay software with graphics, physics, and interaction with the player. An own model that was a replica of the original, with original frames, prediction of future events, as well as the path the player would take and even his interaction with the enemies. All this without seeing the original code of the game.
This was with only two minutes of video gameplay, so their managers say that if you train using additional videos, you could create an almost perfect clone of the game. In the case of this experiment, the researchers mentioned that the engine created by artificial intelligence was almost indistinguishable from the original, although there were occasional failures such as the lack of some frames or characters that disappeared momentarily.
This work is the first focused on generating an AI technique to learn the engine of a game and simulate it in a simple way. This would help future developers to create the bases of an engine so that later it would be created by AI, where the developer would only have to make certain adjustments towards the final objective.
)
They also believe that this tool could have a place in educational or training areas, where platforms would be able to make predictions about their environment and thus create different game modes. Right now the project has focused on 'Super Mario Bros', but they have already started working with 'Mega Man' and 'Sonic the Hedgehog'.