Task Planning With Retrieval-Guided Resolution for Voice Assistants
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing intelligent automated assistants face inefficiencies in processing natural-language speech inputs due to the need for extensive language model interactions, leading to increased power consumption and reduced reliability.
Innovation Solution
Implementing a system that uses a language model to generate task plans and, if necessary, queries an information retrieval service for additional data to resolve the plan, thereby reducing the burden on the language model and enhancing efficiency and reliability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the system relies heavily on the language model to process and resolve task plans, then the system can handle complex natural-language requests, but the power consumption increases and reliability decreases due to language model hallucinations
Solution Approach 1:
The patent introduces an information retrieval service as an intermediary between the language model and task execution. The language model generates initial task plans, and when resolution criteria are not met, the system queries external information retrieval services to obtain accurate data. This mediator approach reduces reliance on the language model's internal knowledge, thereby reducing hallucinations while maintaining the ability to handle complex requests.
Solution Approach 2:
The patent segments the task resolution process into distinct phases: initial plan generation by the language model, evaluation against resolution criteria, and conditional querying of information retrieval services. This segmentation allows the system to use the language model only for creative plan generation while delegating factual verification to more reliable external sources, thus improving reliability without sacrificing adaptability.
2Ease of operation
If the system uses extensive language model interactions to process speech inputs, then the system can understand complex user intents, but the processing time increases and battery life decreases
Solution Approach 1:
The patent applies partial action by having the language model generate only an initial task plan without requiring it to resolve all ambiguities. The system performs selective querying of information retrieval services only when resolution criteria are not met, rather than requiring the language model to provide complete, fully-resolved plans. This reduces the computational burden on the language model and accelerates processing while maintaining understanding of complex intents.
3Reliability
If the system requires the language model to provide complete and accurate task plans, then the reliability of task execution improves, but the number of language model interactions increases leading to higher power consumption
Solution Approach 1:
The system implements a self-service mechanism where the information retrieval service provides accurate data independently when the language model's initial plan is insufficient. Rather than requiring multiple rounds of language model interactions to achieve reliable task plans, the system autonomously queries external sources to resolve uncertainties, reducing power consumption while maintaining execution reliability.
Data Source
AI summary
Techniques for performing tasks are provided. An example method includes receiving, via the one or more input devices, a natural-language speech input including a request to perform a task; providing, at a language model, a plan corresponding to the task; determining whether the plan satisfies a set of resolution criteria; in accordance with a determination that the plan satisfies the set of resolution criteria, initiating performance of the task according to the selected plan; and in accordance with a determination that the plan does not satisfy the set of resolution criteria: providing a query to an information retrieval service requesting a set of resolution data; receiving, from the information retrieval service, the set of resolution data; resolving the plan based on the set of resolution data; and initiating performance of the task according to the resolved plan.