ARLINGTON, Va. – U.S. military researchers are approaching industry to determine how artificial intelligence (AI) and machine learning could help intelligence analysts and commanders gather, distil, and rank information quickly through machine automation.
Officials of the U.S. Defense Advanced Research Projects Agency (DARPA) in Arlington, Va., issued a presolicitation last week (DARPA-EA-23-01-03) for the Collaborative Knowledge Curation (CKC) project.
Knowledge curation involves acquiring information from many sources; triaging it as unimportant, contextual, or actionable; identifying factors and causal links; finding associated datasets; and developing metrics with which to measure objectives.
This process also uses knowledge curation to convert natural language objectives, such as “building soft power,” into measurable causal hypotheses such as “Increasing D will increase X, as measured by I.”
Despite being a vital step in decision-making, knowledge curation usually is done by hand, which means analysts and decision-makers often miss important factors in complicated socioeconomic systems. For example, restrictions on the purchase of Russian oil did not include bans on the sale of ships to Russia, which enabled growth of a shadow fleet to ferry oil in defiance of price caps.
This ARC Opportunity will explore how machines can help analysts and decision-makers curate information faster and more thoroughly. How can we partially automate knowledge curation to help analysts and decision-makers gain and maintain awareness in complicated, interdependent systems?
The CKC project seeks to treat humans and machines as partners in automating knowledge curation. Humans and AI computers could collaborate by enabling humans to act as experts who handle curation goals and provide common sense. Collaborative knowledge curation comes with several technical challenges that involve data, dynamics, and representation.
Data involves knowledge that often must be curated in challenging data environments, such as scarce relevant information, rapidly evolving data, and holes in the narrative.
Dynamics involves knowledge must often be curated when information can be ignored safely, when it represents important context, and when it becomes actionable. Representation, meanwhile, involves sharing information among humans and computers that should not be simplified, and that does not encourage viewers to absorb and expand on it.
The CKC project seeks to develop enabling technologies to curate knowledge that answers questions such as how to evaluate the effects of economic sanctions; how to plan and evaluate the success of climate statecraft; and how to anticipate the responses of other countries to tensions among world powers.
CKC performers may represent curated knowledge as natural language; interactive visualizations; mutually exclusive and testable hypotheses; causal models; uncertainty analyses; and shopping lists for data and metrics.
A DARPA-designed scenario will help judge performers’ collaborative methodologies and technologies, and the CKC project does not require performers to build executable models. Instead, it seeks reusable knowledge curation approaches that extracts causal factors by machine learning and enables human experts to work together with computers; expert-driven approaches such as knowledge engineering are of particular interest. Those submitting promising abstracts will be invited to give oral presentations.
Companies interested should submit unclassified abstracts no later than 30 Nov. 2023 to the DARPA BAA website at https://baa.darpa.mil. Email questions or concerns to DARPA at [email protected]. More information is online at https://sam.gov/opp/7e083cea9c514f268b02af5350b4ed2a/view.
