Microsoft files patent for machine learning system to show ads when players are least annoyed

According to documents published by the United States Patent & Trademark Office, the system is designed to find moments where the likelihood of adverse occurrences is low. This includes scenarios such as being in a safe location, opening a save screen, pausing the game, or completing a round. The primary goal is to avoid a "negative use experience" by making ad delivery feel more natural within the gameplay flow.

The patent outlines several potential features of the system:

• Game State Analysis: The AI could monitor changes in model textures (such as taking damage), equipping new gear, or completing specific tasks like defeating an NPC, finishing a level, or completing three online matches.
• Contextual Adaptation: The system might distinguish between different art styles and avoid interrupting "rare events"—moments of excitement or curiosity—such as a legendary loot drop, a parry in a fighting game, or a significant play in a sports game.
• Gamified Ad Consumption: Players might earn visible or invisible "credits" for viewing promotions, which can be used to delay the next advertisement. The system could also inform players whether it is a good time to watch an ad based on their current in-game situation.

To function, the machine learning model would need to detect and identify events within video information, audio, or game state data. This would allow the system to "learn" correlations, such as associating a growing health bar with the use of a health pack, to refine the timing of UI notifications.

This development follows recent trials of ad-supported game streaming on Xbox. While Microsoft has stated that ads should not interrupt gameplay flow and must meet high quality standards, this patent suggests a more granular approach to automated monetization through real-time gameplay analysis.