DARPA Launches KAIROS AI to Predict Global Events by Linking Billions of Digital Signals

• DARPA wants to uncover complex events found in the digital world. 
• Their plan is to create a semi-automated system that can detect and draw correlations between billions of events currently happening all over the world. 

Artificial intelligence (AI) has numerous applications in today’s world. It can efficiently perform a wide range of activities, including remote sensing, electronic trading, medical diagnosis, and robot control.

The blossoming field of AI has made some impressive strides, especially in the last couple of years. Today, machines running on weak AI can perform a variety of tasks far better than humans, from recognizing and sorting images to defeating the world’s best Go player and diagnosing certain types of cancers.

However, DARPA wants to go a step further in this field by building an AI that can find hidden patterns in global chaos. Recently, they announced a program named Knowledge-directed Artificial Intelligence Reasoning Over Schemas (KAIROS) to create a semi-automated system capable of detecting and drawing correlations between billions of events currently happening all over the world.

KAIROS: A Schema-Based AI

This new program aims at building a machine learning system that can analyze a large amount of data representing daily events and find if there is any thread(s) of connection.

Numerous events, however, do not occur in a sequence, instead, they involve complex phenomena composed of several subsidiary components. The ever-growing volume of unstructured data makes this task even more challenging for existing tools and systems.

That’s why the program will use “schemas” to find connections across information. First conceptualized in the early 1920s, schema is a unit of knowledge in which events are organized in commonly occurring narrative structures.

For instance, people go to stores to purchase something. This type of event involves a purchase-transaction schema that is defined by certain roles (seller, buyer), set of actions (product, payment), and temporal constraints (payable via card if the amount is greater than $100).

Reference: KAIROS | DARPA

In order to build actionable understanding of complex real-world events and accurately forecast how they will unfold, the schema-based AI enables temporal and contextual reasoning of these events. Simply put, it generates board narratives of all things happening around us.

How Does It Work?

DARPA’s objective will be approached in two phases. In the first phase, the system will generate schemas from massive amounts of data by identifying, classifying and clustering events based on reasoning and linguistic inference.

Two phases of the KAIROS program | Credit: DARPA

Developers plan to employ generalization, composition and specialization procedures to create schemas that efficiently describe complex events, and implement domain-specific knowledge to modify the analysis for specific needs.

In the second phase, the system will apply a set of schemas (generated during phase one) to complex, real-world data, and try to extract events and narratives. This won’t be as easy as it sounds because constructing and extending a knowledge base requires detecting complex entities and events, as well as their relationships with each other.

Read: DARPA Will Use Laser Light Source To Power Small Aircraft On The Fly

At this point, the whole system seems theoretical, but that’s why the agency is looking into it. Given the simplicity of AI we have today, it’s quite hard to imagine such a sophisticated system in near future.