Sony patent describes in-game difficulty that adapts to player

Sony has filed a patent application for a means of machine learning that adjusts the difficulty of a game according to the player’s level of play. The technique would make it possible to make a boss easier or more difficult, depending on the gamer.

The patent application mainly mentions the example of a battle with a boss of a level. For example, Sony is thinking of a situation where players often use a certain pattern to defeat an enemy, such as by using buttons in a certain pattern, or operating the joysticks in the same way over and over. The game could then come up with a way, thanks to machine learning, where those patterns no longer work, to make the level more difficult. For example, enemies could move faster, or get a shield to fend off certain attacks. The adaptive difficulty may also work in multiplayer games, the patent writes.

The game developer and console maker not only wants to make it more difficult for players, but also thinks of ways to make games easier along the way. Machine learning could ensure that if an enemy continuously beats gamers with a certain weapon or pattern, this weapon or pattern is no longer used by that enemy.

Sony seems to want to distribute this adaptive difficulty level via servers, among other things, which can make the game more difficult or easier with updates. Players would then be able to choose this adaptive difficulty setting, or the ‘standard’ difficulty setting that the game originally had via a settings menu.

With the application, Sony is not only focusing on fighting or shooting games, but the company is also looking at racing, sports and RTS games. Players who beat a more difficult opponent can also get a bigger reward. This reward could also depend on how an enemy is killed. If the player shows ‘new’ behavior to defeat an enemy, they could get a bigger reward. Conversely, a player could get a smaller reward if they use a ‘simpler’ behavior to kill an enemy. A changing difficulty level also ensures that players have more motivation to play a game again. Whether Sony actually wants to use this form of machine learning for games is unknown.

Update, Wednesday: This article originally stated that Sony has patented the technique. That is not correct, only an application has been submitted. As Ribo89 points out, it remains to be seen whether the patent will be granted, because the first assessment is not positive about the inventiveness of the described technique. The application does, of course, show that Sony is thinking about and working on this technology.