Not so long ago we reported on a computer program which can "learn" how to play Game Boy games, but now someone has gone one better.
A team at the Georgia Institute of Technology is currently working on a program which can build its own Super Mario levels - without the use of Super Mario Maker, we might add - but that's not the most amazing thing about the software.
The program learns what makes a good level not by the information fed into it by the research team, but by watching human players and how they move through pre-existing stages.
The program looks at YouTube gameplay videos to see how players perform and to spot trends or design rules. Once it has complied this information it puts it to use in its own original creations.
Here's Mark Riedl, associate professor of Interactive Computing at Georgia Tech, to explain a little more:
Our system creates a model or template, and it's able to produce level sections that have never been seen before, do not appear random and can be traversed by the player. One could say that the system 'studies' the design of Mario levels until it is able to create new playable areas.
This is a pretty unique development as it's the first time that gameplay videos have been used by a computer to build new levels, and it could change the way in which such games are made in the future. If we have a future, that is - we imagine that the program will get bored of endlessly replicating the Mushroom Kingdom and will instead engage in something more challenging, like wiping out our pathetic species.