Foundational Artificial Intelligence (AI) Science
Abstract
The Foundational Artificial Intelligence (AI) Science thrust will develop a fundamental scientific basis for understanding and quantifying performance expectations and limits of AI technologies. Current AI technologies are challenged in handling uncertainty and incompleteness of training protocols and data. This has prevented the successful integration of AI technology into many transformative DoD applications. To address these limitations the Foundational AI science thrust will focus on the development of new learning architectures that enhance AI systems' ability to handle uncertainty, reduce vulnerabilities, and improve robustness for DoD AI systems. One focus area of this thrust is the ability to embed known physics, mathematics, and other prior knowledge to improve performance of AI systems, particularly for problem sets involving incomplete, sparse and noisy data. Another focus area is the development of a model framework for quantifying performance expectations and limits of AI systems as trusted human partners and collaborators. A third focus area is the development of new tools and methodologies that enable AI approaches for accelerated scientific discovery. The technology advances achieved under the Foundational AI Science thrust will ultimately remove technical barriers to exploiting AI technologies for scientific discovery, human-AI collaboration, and other DoD relevant applications.
Document Details
- Document Type
- Accomplishment
- Publication Date
- Oct 01, 2021
- Source ID
- 1341991792b01b4c6bee28445b79ac98