Gaming and AI | Montréal

Gaming is by far the largest entertainment industry, reaching over 3 billion players in 2022. Given the huge opportunities in this space, the games industry is continuously looking for new ways to engage players and manage the ever-increasing complexity of creating a AAA title. One of the key opportunities for addressing these challenges is AI innovation. Much of the research at Microsoft Research (MSR) Montreal can be applied to gaming, be it in Reinforcement Learning (RL), Natural Language Processing (NLP) or Fairness, Accountability, Transparency & Ethics (FATE). Various researchers at MSR Montreal take inspiration from and try to contribute to the challenges faced in AI for gaming.

As an example of how gaming can inspire fundamental research, consider the challenge of building a Non-Player Character (NPC) using RL. If we could train an NPC rapidly to exhibit interesting behavior and generalize to new scenarios, it would have a huge impact on game authoring: designers could focus more on narrative and less on the mechanics of writing a new behavior. This task is not well addressed by current RL research: game simulators are often used as a way to demonstrate learning, but the focus is typically either on single-player games (e.g., Atari) where the agent tries to maximize its score, or on challenging multi-player games (e.g., Go or StarCraft) where the agent tries to beat the best human players. Despite how impressive the research related to these challenges is, these are not scenarios that are directly relevant for the Gaming industry: e.g., an unbeatable NPC leads to little long-term player engagement. Instead, it’s much more interesting to build a «buddy» NPC that learns to collaborate with human players and communicate through natural language, or an enemy NPC that behaves in a human-like way and has a skill level that is approximately the same as the human player.