AI learns how to build Super Mario levels by watching YouTube
So, in some lighter news, there now exists an artificial intelligence that can build it’s own level in Super Mario Bros. Seriously.
Researchers at the Georgia institute of Technology created an AI that learns the rules and gameplay of the game simply by ‘watching’ gameplay videos on Youtube and Twitch, observing and interpreting what it analyses by the players behaviour, the placement of terrain and the rules of the game itself.
The study was an effort to ‘show how rich game design information can be automatically parsed from gameplay videos and represented as a set of generative probabilistic models’.
In a blog post, lead researcher Matthew Guzdial explained –
“The system focuses on the gaming terrain (not the playable character) and the positioning between elements on-screen – be it pipes, blocks, coins or Goombas – and it determines the required relationship or level design rule. For example, pipes in the Mario games tend to stick out of the ground, so the system learns this and prevents any pipes from being flush with grassy surfaces. It also prevents “breaks” by using spatial analysis – e.g. no impossibly long jumps for the hero“.
Eventually, this system led to 151 distinct sections from 17 samples in the original game, increasing to 334 level sections as the constraints were loosened.
Whilst this type of research may not have any immediate impact, it is definitely interesting to vision what sort of changes the advancements of artificial intelligence could have on how video games are designed in the future. And, regardless, it’s pretty cool.
Here is the full paper, if you are interested.