Some Nonparametric Approaches for Open-world Novelty (SNAP-ON)

Military uses of AI are anticipated to be widespread. Military operations are typically characterized by novel situations, which arise in open worlds. We use the term “novelty” here to refer to situations that violate implicit or explicit assumptions about agents, the environment, or their interactions. As AI increasingly becomes ubiquitous for various aspects of military operations (including in decision support, human-machine collaboration, and autonomy), it will be essential for military AI applications to be aware of novelty in open worlds and capable of acting appropriately and effectively when confronted by novel situations. DARPA’s Science of Artificial Intelligence and Learning for Open-world Novelty (SAIL-ON) program has been addressing the issue of open-world AI. SAIL-ON seeks to develop the underlying scientific principles and general engineering techniques and algorithms needed to create AI systems that act appropriately and effectively in novel situations which occur in open worlds. The objectives of the program are to 1) Develop scientific principles to quantify and characterize novelty in open world domains, 2) create AI systems that act appropriately and effectively in open world domains, and 3) demonstrate and evaluate in DoD domain/IV&V. For the purposes of this topic, we assume that novelty comes from the SAIL-ON developed novelty hierarchy [2][4]. The key technical challenges associated with this SBIR topic include: developing domain-independent technical approaches that can address detection, characterization, and accommodation of novelty from all novelty hierarchy levels; and identifying “snap-on” technologies that can be added to already existing agents in both action and perception-oriented domains.