Skip to content
Thomson Reuters Labs™

Toward a more peaceful world: Using technology to aid nonproliferation

Brian Ulicny  Vice President, Thomson Reuters Labs

Brian Ulicny  Vice President, Thomson Reuters Labs

On the heels of United States President Donald Trump’s historic de-nuclearization summit with North Korean leader Kim Jong-un, non-proliferation is once again a timely topic. How can advancements in machine learning, data analysis and artificial intelligence aid nonproliferation?

Since the dawn of the nuclear age, keeping tabs on who has military-grade nuclear capabilities and materials has been a vital – and difficult – task. Thankfully, it’s also one that may be getting easier, thanks to leaps forward in fields like data analysis, machine learning and artificial intelligence.

Last month, Thomson Reuters Labs was invited to present at a workshop called “Applications of Innovative Tools and Technologies for Nonproliferation and Disarmament” held in Krems, Austria, for diplomats representing their countries at the International Atomic Energy Agency (IAEA) and other international organizations.  The diplomatic workshop was preceded by a day-long session for technical participants at the Vienna Center for Disarmament and Non-Proliferation.

The purpose of the workshop was to help diplomats who represent their countries at the IAEA and other United Nations organizations involved with non-proliferation understand the state of technology for monitoring proliferation in the world.  One session focused on the new capabilities for satellite imaging that are becoming available even as open-source intelligence.  Others focused on machine learning and AI, etc.

In all cases, diplomats are extremely concerned about preserving the credibility of the international mechanisms for maintaining nonproliferation.  Any promising technology’s limitations and potential downsides have to be carefully investigated; mistakes in this area are simply too costly.  Threats to member nation sovereignty are also extremely concerning.

International Atomic Energy Agency (IAEA) headquarters is pictured in Vienna, Austria September 26, 2017. REUTERS/Leonhard Foeger
International Atomic Energy Agency (IAEA) headquarters is pictured in Vienna, Austria September 26, 2017. REUTERS/Leonhard Foeger

The panel in which we participated, “Mining Data, Machine Learning and Data Analysis,” focused on applying cutting-edge technology to nonproliferation intelligence analysis. Our focus was on the scale of training-set data that can be leveraged for different types of AI- driven intelligence or analysis workflows relevant to nonproliferation.  Thomson Reuters analysts perform a great many similar tasks to those that non-proliferation analysts do routinely.   Many of these are or can be enhanced with machine learning.

Machine learning is learning by example. A computer is presented with many, many examples of the phenomenon it is being trained to discern, and it “learns” the statistical regularities that distinguish cases that fall under that label from those that don’t.

Analysis tasks are an ideal use case for developing machine learning capabilities because the historical record of past analyses can be leveraged to train the machine to perform similar tasks, at least in routine cases.  For example, machines can be trained to categorize images by the type of vehicle depicted, identify subtle changes in scenes, and other analytic tasks.   In order to train the machine, increasing amounts of training data are required.

With training data on the order of thousands of examples, it is possible to train a computer to distinguish entities of certain types and to categorize text by topic. With training data levels closer to the millions, we can train the machine to not only identify and categorize elements within text, but also learn to specify the “So what?”: What is this text about and why is it relevant?  While automatic summarization has been around for years, prior techniques have largely been extractive. Deep learning algorithms, along with millions of examples of prior abstracts of texts, allow the machine to learn to produce similar abstracts, where the text of the abstract is not found within the summarized text itself.  The machine learns what sort of summary is called for by the source text by analyzing millions of pairs of example source texts and human summaries.  It can then produce similar results.  Astonishingly, our evaluations have shown that the machine-generated texts are more grammatical in the aggregate than the human-generated ones.

The data and technology that Thomson Reuters provides can play an important role in making analysts more efficient, including within this domain of nonproliferation on which the world is currently focused.  With machines augmenting human analysts, we have a better chance of analyzing all the relevant information and connecting all the dots between bad actors before it is too late.

An Egyptian gestures at Tahrir Square, the focal point of the Egyptian uprising, in Cairo January 1, 2012. REUTERS/Amr Abdallah Dalsh
An Egyptian gestures at Tahrir Square, the focal point of the Egyptian uprising, in Cairo January 1, 2012. REUTERS/Amr Abdallah Dalsh

Learn more

Thomson Reuters supports nonproliferation compliance among its customers in two ways: Because financial institutions are required to avoid sanctioned organizations and individuals, our World-Check database tracks organizations and individuals who are on sanctions lists for weapons proliferation provided by various sanctioning entities such as the U.S. Office of Foreign Asset Control (OFAC) and the UN.

Secondly, our OneSource Global Trade Management solution helps companies comply with export control lists for arms control and dual use goods by categorizing exports and helping them to maintain their export licenses.