Roblox’s Chief Scientist Morgan McGuire talks about the AI tools that Roblox uses, focusing on pathfinding and moderation filters. The pathfinding system is designed to scale to novice users as well as experts, allowing the character to navigate around the game world without the need for handcrafting. The moderation filters use a deep neural net based on a transformer model to replace traditional keyword-based filtering. This allows for a high level of safety for text-based communication in a cross-language environment, preventing inappropriate language while eliminating false positives.
McGuire also discusses how AI is used for localization, a task that traditionally requires translation by a team of people. By using AI for localization, developers can put up signs in their games and the system will automatically translate the text in the game into the appropriate language. This helps to reduce the burden on small development teams and makes the game more accessible to players around the world.
Overall, he emphasizes that the focus of using AI tools in Roblox is not just about using buzzwords but about providing new tools that can enhance the community’s experience. The goal is to use AI in a way that helps developers and players alike, making it easier to create and enjoy games in a safe and inclusive environment.