Generating Actionable Understanding of Real-World Phenomena with AI

Estimated reading time: 2 minutes

Rapid comprehension of world events is critical to informing national security efforts. These noteworthy changes in the natural world or human society can create significant impact on their own, or may form part of a causal chain that produces broader impact. Many events are not simple occurrences but complex phenomena composed of a web of numerous subsidiary elements—from actors to timelines. The growing volume of unstructured, multimedia information available, however, hampers uncovering and understanding these events and their underlying elements.

“The process of uncovering relevant connections across mountains of information and the static elements that they underlie requires temporal information and event patterns, which can be difficult to capture at scale with currently available tools and systems,” said Dr. Boyan Onyshkevych, a program manager in DARPA’s Information Innovation Office (I2O).

The use of schemas to help draw correlations across information isn’t a new concept. First defined by cognitive scientist Jean Piaget in 1923, schemas are units of knowledge that humans reference to make sense of events by organizing them into commonly occurring narrative structures. For example, a trip to the grocery store typically involves a purchase transaction schema, which is defined by a set of actions (payment), roles (buyer, seller), and temporal constraints (items are scanned and then payment is exchanged).

To help uncover complex events found in multimedia information and bring them to the attention of system users, DARPA created the Knowledge-directed Artificial Intelligence Reasoning Over Schemas (KAIROS) program. KAIROS seeks to create a schema-based AI capability to enable contextual and temporal reasoning about complex real-world events in order to generate actionable understanding of these events and predict how they will unfold. The program aims to develop a semi-automated system capable of identifying and drawing correlations between seemingly unrelated events or data, helping to inform or create broad narratives about the world around us.

KAIROS’ research objectives will be approached in two stages. The first stage will focus on creating schemas from large volumes of data by detecting, classifying and clustering sub-events based on linguistic inference and common sense reasoning. Researchers taking on this challenge will apply generalization, composition and specialization processes to help generate schemas that describe both simple and complex events, sequence multiple schemas together to understand key contextual elements like roles and timelines, and apply domain-specific knowledge to tailor the analysis for a particular need.

The second stage of the program will focus on applying the library of schemas created during stage one to multimedia, multi-lingual information to uncover and extract complex events. This stage will require identifying events and entities, as well as relationships among them to help construct and extend a knowledge base.

DARPA will hold a Proposers Day on January 9, 2019 from 10:00am to 2:30pm (EST) at the Holiday Inn at Ballston, 4610 N. Fairfax Drive, Arlington, Virginia 22203 to provide more information about KAIROS and answer questions from potential proposers.

This image outlines the two stages of the KAIROS program. The first stage will focus on creating a library of schemas from large volumes of data by detecting, classifying and clustering sub-events based on linguistic inference and common sense reasoning. The second stage will apply those schemas to new information to uncover and extract complex events, as well as relationships among them, to help construct and extend a knowledge base.