Shipboard Assistants for Troubleshooting AI Systems
Abstract
The promise of shipboard Artificial Intelligence (AI) is undermined by the need for a highly skilled crew capable of understanding, debugging and modifying its behavior to meet changing conditions and requirements. This three-year research project will develop the foundations for personalized troubleshooter AI (TAI) systems agents that allow non-AI experts to understand, debug and safely modify the behavior of shipboard AI devices. The proposed TAI systems will adapt automatically to the skill level of the user and the AI device that needs to be operated. This approach to the problem would enable crew members to use AI systems as force multipliers. In order to achieve this objective, the proposed project will develop and evaluate methods for querying AI devices in a manner that is independent of the internal design of the device. Techniques for symbolic and probabilistic inference will be utilized to automatically generate such queries and to derive user-interpretable models of AI devices. Methods for designing and categorizing interfaces for AI devices for supporting such query mechanisms will also be developed. The project’s development will be guided through periodic empirical and user-study based evaluations. The outcomes of this research would enable research and development programs focusing on the practical development, optimization, and deployment of TAI systems. These programs would draw upon feedback from Navy personnel regarding the classes of AI devices that may see early deployment and the skill-profiles of sailors who would need to operate these devices. These inputs would help develop higher TRL research programs focusing on development of TAI systems for specific classes of high-priority shipboard AI systems. They would also help develop research programs focused on regulating and coordinating the development of shipboard AI systems using interfaces that support on-board troubleshooting assistants. The outcomes of the proposed research help advance multiple research programs within ONR, particularly within Division 341 and Division 311.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- May 05, 2021
- Source ID
- N000142112045
Entities
People
- Siddharth Srivastava
Organizations
- Arizona State University
- Office of Naval Research
- United States Navy