Dynamic Agent Engagement Control for Human-Machine Dialog
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current virtual assistants experience inefficiencies and unnatural interactions due to the need for frequent wake-up indicators, leading to 'wakeword fatigue' and the King Midas effect, where unintended actions are triggered by agents remaining engaged after initial requests.
Innovation Solution
Implementing engagement policies such as locked engagement, attentive agents, and hybrid policies that allow users to control agent engagement explicitly or autonomously, reducing the need for constant wake-up indicators and minimizing unwanted actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the agent remains engaged after processing a user request, then the agent can respond to subsequent requests immediately, but the agent may trigger unintended actions on requests not addressed to it (King Midas effect)
Solution Approach 1:
The agent's engagement state is made dynamic rather than static. The system transitions between engaged and disengaged states based on contextual analysis of user requests. The agent monitors engagement indicators (such as whether speech is directed to the agent) and adjusts its engagement state accordingly, allowing it to be selectively responsive to different users in the environment.
Solution Approach 2:
The system implements feedback mechanisms to determine whether a user request is intended for the agent. By analyzing engagement indicators and user intent, the system provides feedback about whether the agent should process the request or remain dormant, preventing unintended actions while maintaining responsiveness to legitimate requests.
2Object-affected harmful factors
If the agent disengages after every user request, then unintended actions are prevented, but the user must repeatedly wake up the agent with wake-up indicators (wakeword fatigue)
Solution Approach 1:
The engagement state is dynamically adjusted based on contextual factors rather than following a rigid disengage-after-every-request pattern. The agent can maintain engaged state for sequences of requests when contextual indicators suggest continuous interaction is intended, reducing the frequency of wake-up indicators needed while still preventing unintended actions through contextual analysis.
Solution Approach 2:
The system changes the parameter of engagement duration based on contextual analysis. Instead of a fixed disengage timing, the engagement duration is adjusted according to factors such as whether multiple users are present, the nature of the conversation flow, and engagement indicators, optimizing the balance between responsiveness and preventing unintended actions.
3Reliability
If the agent processes all user requests while engaged, then service continuity is maintained, but the user experience becomes unnatural due to frequent wake-up requirements
Solution Approach 1:
The agent performs self-service by autonomously determining its own engagement state based on contextual analysis. The system monitors engagement indicators and user intent, automatically adjusting its responsiveness without requiring explicit wake-up commands from the user, thereby maintaining service continuity while improving naturalness of interaction.
Solution Approach 2:
The engagement parameter is dynamically changed based on contextual factors to optimize user experience. The system adjusts engagement duration and responsiveness to match natural human interaction patterns, maintaining reliability while reducing the artificial feel caused by frequent wake-up requirements.
Data Source
AI summary
A method of controlling an engagement state of an agent during a human-machine dialog is provided. The method can include receiving a spoken request that is a conditional locking request, wherein the conditional locking request uses a natural language expression to explicitly specify a locking condition, which is a predicate, storing the predicate in a format that can be evaluated when needed by the agent, entering a conditionally locked state in response to the conditional locking request, in the conditionally locked state, receiving a request without a need for a wakeup indicator, and for a request evaluating the predicate upon receiving the request, and processing the request if the predicate is true.