Automated assistant training and/or execution of inter-user procedures

A computer implemented method comprising:

• identifying prior performances of a procedure, wherein the prior performances of the procedure includes a user interacting with an automated assistant to perform one or more actions necessary to complete performance of the procedure;
• determining the one or more actions necessary to complete performance of the procedure have been repetitiously performed during the prior performances of the procedure;
• training a model for automating subsequent performance of the procedure, without an explicit request from the user and in response to determining the one or more actions have been repetitiously performed during the prior performances of the procedure, wherein training of the model comprises:
• generating one or more annotation embeddings corresponding to the one or more actions, wherein the annotation embeddings correspond to a first location in latent embedding space;
• generating one or more procedure embeddings based on the annotation embedding, wherein the procedure embeddings correspond to the procedure and correspond to a second location in the latent embedding space;
• identifying, subsequent to training of the model, a user input;
• processing the user input, wherein processing the user input comprises identifying one or more embeddings associated with the user input;
• determining, based on processing the user input and identifying the user input is associated with the one or more annotation embeddings, that the user input is directed to performing the one or more actions;
• and
• initiating, based on determining that the user input is directed to performing the one or more actions, automated performance of the procedure by the automated assistant or another automated assistant in response to the procedure embeddings being located within a threshold distance of the annotation embeddings in the latent embedding space.

The method of 8, wherein initiating automated performance of the procedure by the automated assistant or the other automated assistant is in response to the procedure embeddings being located within a threshold distance of the annotation embeddings in the latent embedding space is in further response to:

• receiving a separate input from a separate user and/or an application;
• processing the separate input using a neural network, wherein processing the separate input comprises identifying one or more separate annotation embeddings, corresponding to one or more separate actions and a separate location in the latent embedding space, associated with the separate user input;
• and
• determining the procedure embeddings are located within a threshold distance of the separate annotation embeddings in the latent embedding space.

A system comprising:

• one or more computers and one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising:
• receiving, by an automated assistant, an inquiry from a user;
• identifying a correlation between the inquiry and a procedure, based on a prior instance of the procedure being performed within a threshold period of time of the inquiry or a similar inquiry being received by one or more users;
• identifying parameters necessary for performing the procedure, wherein the parameters incorporate data derived from a plurality of sources of information during the prior instance of the procedure being performed;
• determining, based on the parameters, one or more applications separate from the automated assistant that are capable of performing the procedure;
• generating, at the automated assistant and based the one or more applications capable of performing the procedure and based on the parameters, an application programming interface (API) output for the one or more applications;
• transmitting the API output from the automated assistant to the one or more applications;
• receiving, from the one or more applications in response to receiving the API output, application data characterizing one or more features of content by the one or more applications based on the parameters;
• processing the application data using a trained machine learning model;
• and
• rendering natural language content for the user based on the processed application data.