Engineering Artificial Intelligence Systems Implementations (EAISI)
Abstract
The Engineering Artificial Intelligence Systems Implementations (EAISI) program will create technologies and tools to support the development of viable and trusted systems that include AI and machine learning (ML) capabilities. Modern AI-dependent systems may include multiple AI components, drawing on a diverse set of AI-related techniques, ranging from ML to knowledge representation, search, planning, game theory, and optimization. Current methods for development of such systems remains primarily based on trial-and-error designs, with limited abstractions, architectures, and patterns. These developments can be costly, risky, and demanding of very high levels of expertise. To address this, EAISI will develop abstractions, patterns, architectures, assurance techniques, and iterative processes that facilitate the analysis and synthesis of complex systems that must rely on AI-based components and associated training data. One of the more difficult engineering challenges with AI is evaluation and assurance, since AI-based systems tend to resist traditional approaches to testing, inspection, and analysis. It is not possible to fully test an AI-based system for every situation it will ever encounter, so new techniques are needed for verifying and validating AI-based systems. EAISI aims to create software and systems engineering techniques, tools, and practices to facilitate the development of AI-based systems that are capable, trustworthy, affordable, and timely.
Document Details
- Document Type
- Accomplishment
- Publication Date
- Oct 01, 2022
- Source ID
- 6ead37f36bd58932197fa51152472ba0