A game AI patent filed by Sony Interactive Entertainment was published on October 1st, proposing that non-player characters adjust their in-game performance based on player interaction data. The changes envisioned by the patent could involve character behavior, dialogue, and appearance, but do not point to any announced game features yet.
According to the patent description, the system collects player and game data, especially information generated during encounters between players and non-player characters, which is then processed by a machine learning model to generate new character content and modify the character's in-game performance accordingly. The patent does not specify the particular model architecture.
Regarding this patent, an editor from foreign media outlet wccftech used action games represented by the Dark Souls series as a case study for speculation: these games require players to familiarize themselves with enemy moves, countermeasures, positioning, and timing. If enemies could adjust their strategies based on the player's offensive or defensive tendencies, boss battles might become less reliant on memorizing fixed patterns.
The editor also assumes that in multi-phase boss battles, players who maintain constant pressure might prompt enemies to enter their stronger next phase earlier. However, there is currently no conclusion on whether such systems can be applied to actual games or how difficulty would be controlled.