Negative, Suppression is Void: AI, Deception, and Fighting Machines
Abstract
This paper examines the implications for the introduction of AI and autonomous weapons systems to the battlefield. An overview of the United States current efforts in AI is compared to those of Russia and China. This paper then explains how machine learning works in order to examine how it can be deceived. This paper then examines the current state of research into deceiving AI via a literature review, focusing on adversarial examples and data poisoning attacks. This paper then postulates a threat model on how an adversary could leverage either type of attack in order to develop a method of comparing the relative strength of the two. These attacks are then conducted against simulated data.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 07, 2021
- Accession Number
- AD1178260
Entities
People
- Brian J. Strom
Organizations
- Marine Corps University