Abstract: Aspects of the present disclosure relate to a personalized agent service that generates and evolves customized agents that can be instantiated in-game to play with users. Machine learning models are trained to control the agent's interactions with the game environment and the user during gameplay. A user may request that a personalized agent join the user's gameplay session. The user device sends a request for the personalized agent to a game platform. The game platform determines whether the user has a license to execute a second instance of the game. When the user has a license to execute a second instance of the game, the second instance of the game may be executed on the user device. Information received from a personalized agent service is used to instantiate a personalized agent in the second instance of the game.

Abstract: Aspects of the present disclosure relate to a personalized agent service that generates and evolves customized agents that can be instantiated in-game to play with users. Machine learning models are trained to control the agent's interactions with the game environment and the user during gameplay. As the user continues to play with the agent, the one or more machine learning models develop gameplay styles for the agent that complement the user's preferred playstyle, incorporate the user's preferred strategies, and is generally customized for interaction with the user. The agent personalization data generated during gameplay is stored by the service. An application programming interface is provided by the personalized agent service. Using the API, games can import agent personalization data in order to customize in-game non-player characters (NPCs), thereby customizing the in-game NPCs in accordance with the user's preferences.

Abstract: In examples, a developer may define a set of computer-controlled agent attributes, which may be processed by a generative multimodal machine learning model in conjunction with background information associated with a virtual environment (e.g., “lore”) and other agent information to generate multimodal model output with which to control the behavior of the computer-controlled agent. Thus, a player may interact with the computer-controlled agent, such that user input from the player is processed using the ML model to generate model output to affect the behavior of the computer-controlled agent, thereby enabling the user and the computer-controlled agent to interact. As compared to manual dialogue authoring, use of agent information to define the behavior of a computer-controlled agent may result in reduced effort on the part of a creator while also offering increased depth and variety for computer-controlled agents of a virtual environment.

Abstract: Aspects of the present disclosure relate to a machine learning-based dialogue authoring environment. In examples, a developer or creator of a virtual environment may use a generative multimodal machine learning (ML) model to create or otherwise update aspects of a dialogue tree for one or more computer-controlled agents and/or players of the virtual environment. For example, the developer may provide an indication of context associated with the dialogue for use by the ML model, such that the ML model may generate a set of candidate interactions accordingly. The developer may select a subset of the candidate interactions for inclusion in the dialogue tree, which may then be used to generate associated nodes within the tree accordingly. Thus, nodes in the dialogue tree may be iteratively defined based on model output of the ML model, thereby assisting the developer with dialogue authoring for the virtual environment.

Abstract: Systems and methods are provided for automatically generating a program based on a natural language utterance using semantic parsing. The semantic parsing includes translating a natural language utterance into instructions in a logical form for execution. The methods use a pre-trained natural language model and generate a canonical utterance as an intermediate form before generating the logical form. The natural language model may be an auto-regressive natural language model with a transformer to paraphrase a sequence of words or tokens in the natural language utterance. The methods generate a prompt including exemplar input/output pairs as a few-shot learning technique for the natural language model to predict words or tokens. The methods further use constrained decoding to determine a canonical utterance, iteratively selecting sequence of words as predicted by the model against rules for canonical utterances. The methods generate a program based on the canonical utterance for execution in an application.