Automated Assistant Agent Selection With Intent Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated assistants struggle to select an appropriate agent for user intents when the user's input fails to specify a particular agent, leading to inefficient resource consumption and increased dialog turns.
Innovation Solution
A machine learning model is trained to predict the probability of available agents handling user intents based on current dialog and contextual values, enabling selection and invocation of a suitable agent without explicit user knowledge of available agents.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the automated assistant transmits invocation requests to multiple available agents to determine which agent can handle the user intent, then the reliability of agent selection is improved, but the computational resource consumption increases
Solution Approach 1:
The system performs preliminary actions by transmitting invocation requests to multiple available agents before final selection. This allows the automated assistant to evaluate which agents can handle the user intent in advance, ensuring reliable agent selection while managing computational resources through structured evaluation processes.
Solution Approach 2:
The system uses feedback from agent responses to determine which agent should be selected. By evaluating the responses from multiple agents and using this feedback to make the final selection, the system ensures reliable agent choice while avoiding unnecessary invocation of all agents, thus balancing reliability with resource consumption.
2Reliability
If the automated assistant invokes multiple agents to handle user intents, then the reliability of intent fulfillment is improved, but the number of dialog turns increases
Solution Approach 1:
The system performs preliminary evaluation by transmitting invocation requests to multiple agents before the actual dialog begins. This preliminary action identifies which agents can handle the user intent, so that only the most suitable agent is invoked for the actual conversation, reducing the number of dialog turns while maintaining reliability.
Solution Approach 2:
The system takes a partial action approach by transmitting invocation requests to multiple agents for evaluation purposes only, without fully invoking all of them. This allows the system to gather information about agent capabilities while minimizing the actual dialog turns by selecting only the most appropriate agent for fulfillment.
3Productivity
If the automated assistant selects an agent without evaluating multiple available agents, then the computational resource consumption is reduced, but the accuracy of agent selection deteriorates
Solution Approach 1:
The system applies partial action by transmitting invocation requests to multiple agents for evaluation without fully invoking all of them. This allows the system to maintain high agent selection accuracy by evaluating multiple candidates while preserving processing efficiency by selecting only the most suitable agent for actual execution.
Solution Approach 2:
The system performs preliminary evaluation of multiple agents by transmitting invocation requests before final selection. This preliminary action ensures accurate agent selection by comparing multiple candidates while maintaining productivity by avoiding full invocation of all evaluated agents, selecting only the best match.
Data Source
AI summary
Invoking an agent during a dialog between a user and an automated assistant. Some implementations are directed to receiving, during a human-to-automated assistant dialog, natural language input of the user that indicates a desire to engage an agent, but that fails to indicate a particular agent to be engaged. Those implementations are further directed to selecting a particular agent from a plurality of available agents, and transmitting an invocation request to the selected particular agent. In some implementations an agent selection model can be utilized in selecting the particular agent, such as a machine learning model. The machine learning model can be trained to enable generation of output that indicates, for each of a plurality of available agents (and optionally intent(s) for those agents), a probability that the available agent (and optionally intent) will generate appropriate responsive content.