AI & Machine Learning

Nvidia received 1 patent describing a neural network assistant that takes text descriptions of game content and produces long-term strategy outputs rather than simple next-move suggestions. The system is oriented toward goal-driven planning, with role-playing games serving as a primary example, where the assistant might steer a character arc toward a storyline outcome the player wants to reach. Rather than reacting to a single snapshot of game state, the assistant works from text inputs that describe multimedia content, repositioning the tool as a planner rather than a prompter.

Acer received 1 patent for a system that automates the creation of gameplay highlight clips by treating in-game event data as text. Event information from multiple moments in a match gets serialized into a string, which a text classification model then evaluates to determine what kind of event occurred. That predicted category drives the selection of a corresponding segment from recorded video footage, removing the need for a player to manually scrub through recordings to find moments worth saving.

Sony received 4 patents covering a wide range of AI-assisted gaming experiences. One trains an AI player on a specific user's inputs and the game states those inputs produce, allowing it to play in that person's style and hand control back to the human at a chosen moment. Another uses a camera and microphone to read a real viewer's appearance and speech, then applies a machine learning model to adjust how virtual humans and pets in a metaverse look, sound, and behave, with those changes optionally visible on other users' displays as well. A third converts recorded gameplay into short, playable fragments by identifying regions of interest through game state and input analysis, recommends them based on the user's available time and goals, and updates the underlying model from how the user plays those fragments. The fourth addresses the experience of returning to a multiplayer match after stepping away, presenting a keyword cloud drawn from friends' chat during the absence, with selectable topics that connect discussion threads to specific in-game events.

EA received 1 patent for a collusion detection system that evaluates pairs of players rather than individual accounts. The system combines social connection data with in-game behavioral signals into a single feature set for each pair of opponents, then applies anomaly detection to identify statistically unusual pairs. By treating collusion as a relational property between 2 players rather than a characteristic of any single account's performance, the approach targets the coordinated nature of the behavior itself.

Autodesk received 1 patent for a reinforcement learning system that translates a story narrative into a spatial game map. An RL agent takes abstract plot locations and places them on biome polygon maps, satisfying spatial constraints that derive from the narrative rather than from gameplay mechanics or topography alone. Prior map generation methods, as described in the patent, tend to focus on mechanical or geographic concerns and leave narrative placement to manual adjustment, which this system addresses by building story layout into the generation process as a dedicated layer.

Truist Bank received 1 patent for a churn prediction system built specifically around online gaming behavior. The model trains on signals such as gameplay duration, session counts, and virtual resource transactions, and combines those with data from a partner application that tracks real-world resource activity, creating a cross-application picture of each player's engagement within a defined look-back window. When the model identifies a player as at risk of leaving, it triggers a targeted message tied to a reward that requires the player to return for at least 2 separate sessions to claim, turning the prediction into a structured re-engagement mechanism.