Adversarial Autonomous Agents: Implementation and Application

Fundamentally, the aims of this research are to explore the applications of deliberately misaligned AI. While we are waiting on feedback from QinetiQ and Naimuri as to what specific
deliverables they desire, the following are the deliverables they have expressed interest in:
- An implementation of an Adversarial Autonomous Agent (AAA) trained on propaganda and extremist communications pertaining to a specific ideology e.g. far right accellerationism,
Islamic fundamentalism, etc. This model would have potential use cases for intelligence analysts who seek to:
1. Infiltrate online groups for intelligence gathering purposes
2. Better understand the ideology, praxeology and dialect of these groups more extensively
3. Gain knowledge of coded terms, dog whistles and other subtleties of extremist language
4. Understand ways in which terrorists may use such a tool in future to automate their
radicalisation activities