Artificial Intelligence Technology Scouting

  • By: JAIC Public Affairs
Artificial Intelligence Technology Scouting

FEBRUARY 2021

This edition will address the below topics:

  • Reinforcement Learning
  • AI for Biosecurity
  • AI for Cybersecurity


Topics

Reinforcement Learning

There are signs that AI systems might learn best when trained on “curriculums” that the systems themselves design. Advances in this area could accelerate the use of so-called “Reinforcement Learning” in defense robotics.

Increasingly, AI researchers are training AI systems using a paradigm called “Autocurricula Generation” in which an AI agent learns to master a task by effectively generating its own sequence of progressively more challenging versions of a given task. These systems use “Reinforcement Learning”, an approach for training AI in which an autonomous or semi-autonomous system seeks to maximize the reward that it receives for completing a specific pre-defined task. The approach has especially accelerated robotics capabilities.

Link: Who needs a teacher? | Science

AI for Biosecurity

In the near future, AI may help biological security researchers identify dangerous virus mutations before they are seen in the wild.

Researchers at MIT used an AI approach that was borrowed from Natural Language Processing (NLP), and applied it to the automatic processing of genetic data in viruses. These researchers utilized an analogy between grammar in language and infectiousness in viruses. Applying methods commonly used in teaching AI grammar rules, the researchers had their system learn the rules to decide whether a virus’s genome will cause it to be infectious.

Link: Can an AI Predict the Language of Viral Mutation? | Wired  
Link: The 5G Ecosystem: Risks & Opportunities for DoD - Defense Innovation Board, 3 April 2019 STUDY | Defense.gov  

AI for Cybersecurity

AI is being deployed aboard UAVs to detect and respond to previously unseen cybersecurity threats, without an internet connection.

A group of tech companies recently announced a partnership to integrate AI-based cybersecurity to aerial drones. These edge-AI implementations classify some payloads as potential cyber-attack vectors in advance, and have, according to Aviation Today been used to successfully catch and prevent advanced zero-day attacks onboard UAVs.

Link: Using Artificial Intelligence to Counter Cybersecurity Threats for Drones | Aviation Today

The following blog is for AI/ML learning and education purposes only. The programs and initiatives mentioned in this blog are not necessarily an endorsement by the U.S. government, the DoD, or the JAIC.