DARPA Names Participants in Artificial Intelligence Agent Development Program

The Defense Advanced Research Projects Agency has named five teams from the private and academic sectors to participate in Environment-driven Conceptual Learning, a program to develop artificial intelligence assistants for urgent national security tasks.

DARPA seeks AI agents that can learn from linguistic and visual input and work with humans to analyze multimedia documents. Research on human-AI teaming is part of the agency’s broader goal of ensuring the trustworthiness of autonomous computer systems.

Wil Corvey, program manager of ECOLE, explained that reliable AI automation hinges on the ability to gain conceptual knowledge, which in turn requires new methods of analyzing the properties of objects and activities.

The ECOLE participants are expected to employ complementary techniques in their respective research projects. A team from the University of Illinois Urbana-Champaign plans to create a curriculum learning platform that extracts “symbolic knowledge representations from unconstrained unlabeled multimodal data” without supervision.

Meanwhile, General Electric Research will devise ways to automatically generate curricula for discovering the properties of objects and actions, DARPA said Tuesday.