Fiscal Year	2023
Division	Research & Sponsored Programs
Department	Naval Research Program
Investigator(s)	Rowe, Neil C.
Sponsor	NPS Naval Research Program (Navy)
Summary	Trust is an important issue with artificial intelligence (AI) for military applications. Many military AI systems are untrusted by users because they are hard to understand, rarely tested in situations similar to real conflict, use complex mathematics that cannot easily be explained, and used in contexts where deception is prevalent. Untrustworthy technology is likely to get turned off. This study will examine ways to address these problems primarily with better explanations of AI processing, with also study of user-interface concepts that could enhance explanations. A major focus will be on alternative explanations to provide more trust. For instance, while neural networks cannot explain their conclusions well, alternative machine-learning methods such as if-then rules or Bayesian networks than can explain better what they are doing, and we can train them on the same data as a neural network, using insights into the data identified by the neural network to bootstrap the alternative models. We will run tests with a subset of the missile battle-management system developed by NAWC-WD in previous work, a system that used sophisticated neural-network models whose results were difficult to justify to users. We will write a report summarizing our findings and making concrete recommendations for increasing trust in AI systems.
Keywords	artificial intelligence, machine learning, trusted AI, human-machine interactions, explainability, AI system safety
Publications	Publications, theses (not shown) and data repositories will be added to the portal record when information is available in FAIRS and brought back to the portal
Data	Publications, theses (not shown) and data repositories will be added to the portal record when information is available in FAIRS and brought back to the portal