PERCEPTRONICS SOLUTIONS, INC — Department of Defense SBIR Phase I: N202-126

Description

This proposal is for an ARTificial Intelligence, Social networks And Natural language generation (ARTISAN). ARTISAN will provide: (1) Rapid generation of realistic content to facilitate the development of successful simulations of the information environment for military training exercises. (2) Fine tuning using a social network discriminative engine; and (3) Modular organization to facilitate incorporation of today’s continuous improvements in natural language processing algorithms. In today’s high-tech battlefield, information environments are hectic, chaotic, and hard to predict. Advances in artificial intelligence and natural language generation enable adversaries to create synthetic fake news and targeted propaganda that closely mimics the style of real news sufficiently to disrupt civil discourse, to sew discord, and to spread disinformation. To prevail in the information battlefield, warfighters require more realistic training exercises to develop, test and validate tactics, techniques, and procedures for information maneuvers, to blunt information attacks, and to compete effectively in conflict situations. ARTISAN responds to these challenges with a multi-pronged approach comprising: (1) discriminative fine-tuning to rapidly learn network, community, and geographic markers and create an information environment reflecting a realistically nuanced social community; (2) an intuitive, insight-generating interface;  which will allow users to train the hidden neural network language model to produce even higher quality material as they naturally respond to synthetic texts; and (3)  modular organization to allow incorporation not only of current solutions but also of as yet undeveloped optimal solution(s).