How might curiosity-driven learning AI/ML warfighting systems help to probe and exploit opportunities in future warfare?

  • Published
  • By Dr. Andrew Stricker

There is progress in the development of AI/ML systems that self-adapt to evolving contingencies using a learning method referred to as 'reinforcement learning’ (RL). The most common learning approach employed by AI/ML-RL systems is to learn based on external or extrinsic rewards. Such systems operate with goals for maximizing long- term reward. This RL approach has proven effective for AI/ML-RL systems to play complex games (in some cases even better than human experts) and for robotic systems learning how to move and perform tasks. In each example, the AI/ML-RL systems are focused on pre-determined problems offering extrinsic rewards for expected optimal actions that have already been previously found to be effective. The extrinsic rewards are defined by designers and developers of AI/ML-RL systems to serve as reinforcers for action when triggered by sensors detecting their presence in the environment. 

Download the Full Brief