Automated Assistant Agent Selection for Fewer Dialog Turns
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated assistants struggle to efficiently select and invoke appropriate agents for user intents without explicit user knowledge of available agents, leading to potential resource wastage and increased dialog turns.
Innovation Solution
Implement a machine learning-based agent selection model that predicts agent probabilities based on natural language input and contextual values, selecting a particular agent to handle user intents without requiring user selection between multiple agents.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple agents are transmitted invocation requests to handle user intents, then the reliability of intent handling is improved, but computational resource consumption increases
Solution Approach 1:
The system performs preliminary actions by selecting and ranking agents based on predicted performance before transmitting invocation requests. The machine learning model predicts which agents are most likely to successfully handle the user intent, allowing the system to prioritize requests to high-probability agents first, thereby reducing overall computational resource consumption while maintaining reliability.
Solution Approach 2:
Instead of transmitting invocation requests to all available agents (excessive action), the system uses partial action by selecting only the top-ranked agents based on machine learning predictions. This partial approach transmits requests to a subset of agents that are most likely to succeed, reducing computational resources while maintaining adequate reliability through the predicted high-performance agents.
2Ease of operation
If the system waits for user selection between multiple agents, then the ease of operation is improved, but dialog turns increase
Solution Approach 1:
The system implements self-service by automatically selecting and invoking the most appropriate agent based on machine learning predictions without requiring user intervention. The machine learning model analyzes the user intent and autonomously determines which agent should handle it, eliminating the need for users to manually select between multiple agents and thereby reducing dialog turns while maintaining ease of operation.
Solution Approach 2:
The system uses feedback from the machine learning model's predictions to automatically guide agent selection. The model continuously learns from user interactions and adjusts its predictions to better match user intentions, allowing the system to accurately predict which agent should handle each intent without requiring explicit user selection, thus reducing the number of dialog turns.
3Ease of operation
If explicit agent invocation is required from users, then the ease of operation is improved, but device complexity increases
Solution Approach 1:
The system introduces an intermediary - the machine learning model - that mediates between the user's intent and the agent selection process. Instead of requiring users to directly specify which agent to invoke, the ML model acts as an intermediary that translates user intents into appropriate agent selections, simplifying the user interface while managing the complexity of agent routing internally.
Solution Approach 2:
The system replaces the mechanical system of explicit user selection with an automated machine learning-based selection mechanism. Instead of requiring users to manually choose agents through complex interfaces, the ML model substitutes this mechanical process with intelligent prediction and automatic invocation, thereby reducing device complexity from the user's perspective while maintaining ease of operation.
Data Source
Figure 1
Figure 2~3
Figure 4
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.