Sony Interactive Entertainment has patented a technology that will allow the use of artificial intelligence and machine learning to adapt the behavior of non-player characters directly to the player's actions. Potentially, the system could significantly change the traditional approach to creating enemies and other characters in games, especially in genres where studying enemy behavior plays an important role.
The patent, titled “Enhancing game systems using adapted non-player character AI,” was published on October 1. The document describes a system that collects data about the player and in-game interactions with characters, then feeds it to a machine learning model.
According to the patent description, the process consists of several stages. First, the system receives data about the player and the game situation, including information about their interaction with a specific character. This information is then used to form input data for the machine learning model. In the next stage, the AI generates new content for the characters, which can be used to change the character's behavior, their lines and appearance, as well as other game elements.
However, Sony does not limit the technology to a specific AI architecture. Theoretically, this means that the described approach can be implemented using various machine learning models and potentially applied not only by PlayStation's own studios.
This solution looks especially interesting for action games and, in particular, soulslikes. Games inspired by FromSoftware's Dark Souls series largely require the user to have skills that resemble learning fighting games: it is necessary to memorize enemy move sets, understand their counterattacks, choose the right position, and recognize short windows for attack.
However, traditional enemies in single-player games usually act according to pre-programmed scenarios. After several attempts, the player can learn their behavior and begin to anticipate almost every enemy action. As a result, even a difficult boss eventually turns into a sequence of familiar patterns.
AI adaptation can potentially solve this problem. If a character can analyze the actions of a specific player and change their behavior in response, the familiar strategy of “learned the pattern — won” may stop working. For example, if the user constantly dodges to one side, the enemy could theoretically adapt their attacks, and if the player prefers a certain distance or regularly uses the same technique, the enemy could take this into account in future encounters.