Understanding Machine Intelligence (UMI)
Abstract
The Understanding Machine Intelligence (UMI) program will develop techniques that enable artificial intelligence (AI) systems to better support users through transparent operation. If current trends continue, future U.S. military autonomous systems will need to perform increasingly complex and sensitive missions. AI will be critical to such autonomous systems, but in order for developers, users, and senior leaders to feel confident enough to deploy and use AI-enabled systems, they must operate with high degrees of transparency, reliability, predictability, and safety. UMI will develop AI technologies that support transparency by providing supporting rationale and logic sequences that establish the basis for and reliability of outputs. In addition, efforts will be made to develop a mathematically rigorous virtual stability theory for AI-enabled systems analogous to the (conventional) stability theory developed for dynamical systems (solutions to systems of differential equations). Such a virtual stability theory will enable the creation of feedback mechanisms that flag, interrupt, and modify anomalous outputs and behaviors to ensure safe, predictable operation. UMI implementations will be developed and demonstrated in next-generation decision-support and autonomous systems. This program was previously funded in PE 0602702E, Project TT-13.
Document Details
- Document Type
- Accomplishment
- Publication Date
- Oct 01, 2017
- Source ID
- c4c3d96e31d660c3de0d5c314f0bf54e