The Impact of Artificial Intelligence on the IC

By Ian Fitzgerald

Ian Fitzgerald, M.A. student in International Security at George Mason University with research interests in Great Power Competition, Cyber Warfare, Emerging Technologies, Russia and China. Pursuing a career as an all-source analyst with the Intelligence Community. Currently assisting a project regarding disinformation in the 2020 Election.

The Cipher Brief’s Academic Incubator partners with national security-focused programs from colleges and Universities across the country share the work of the next generation of national security leaders.

Ian Fitzgerald is an M.A. student in International Security at George Mason University with research interests in Great Power Competition, Cyber Warfare, Emerging Technologies, Russia and China.

ACADEMIC INCUBATOR — The explosion of data available to today’s analysts creates a compelling need to integrate artificial intelligence (AI) into intelligence work. The objective of the Intelligence Community (IC) is to analyze, connect, apply context, infer meaning, and ultimately, make analytical judgments based on that data. The data explosion offers an incredible source of potential information, but it also creates issues for the IC.

Today’s intelligence analysts find themselves working from an information-scarce environment to one with an information surplus. The pace that data is being generated by technical intelligence (TECHINT) collection or through open sources is growing at an exponential rate; web-based alone will be at 3.3 trillion gigabytes by 2021, according to the Office of the Director of National Intelligence. TECHINT disciplines are collecting data at a rate that far exceeds the IC’s ability to process and exploit it into digestible information, while making it harder for analysts to find the information needed to create solid assessments.

AI is beginning to look like a way to help intelligence analysts overcome the challenges of information overload. AI allows analysts to classify data into meaningful information in ways more reliably and accurately than humans and allows for fusing massive amounts of data across large different data sets at scale and in real time. The technology offers anomaly detection by analyzing routine patterns of behavior and then identifying new behavior outside the norm. Such technology would give intelligence analysts the tools to identify connections between data, flag suspicious activity, spot trends and patterns, fuse disparate elements of data, map networks, and make statistical predictions about future behavior based on past history.

Offloading many of the IC’s data-heavy processing and exploitation tasks onto machines – including data cleaning, labeling or pattern recognition – would free up much of the analysts’ time. An all-source analyst with the support of AI-enabled systems, could save as much as 364 hours or more than 45 working days a year. This allows analysts to direct their energy on the needs of the decision maker, draft their analytic judgements, and disseminate their finished assessments.

This benefit has already convinced the IC leadership of the value of AI as a tool to increase the effectiveness of their analysts. Dawn Meyerriecks, the CIA’s Deputy Director for Science and Technology called AI a very powerful asset to the IC’s scarcest resource: really good analysts. Meyerriecks reported in 2017, that the CIA had 137 pilot projects directly related to AI which “include everything from automatically tagging objects in video… to better predicting future events based on big data and correlational evidence”.

The CIA’s Open Source Center is using AI to comb through news articles from around the world, monitor trends, geopolitical developments and potential crises in real-time. The National Geospatial-Intelligence Agency sees AI as a way to automate certain kinds of image analysis to help free up analysts to perform higher-level work. At the National Security Agency, AI is being used to better understand and see patterns in the vast amount of signals intelligence data it collects, screening for anomalies in web traffic patterns or other data that could portend an attack. Finally, this past year, the Office of the Director of National Intelligence published a principles and ethics framework, laying out guidelines regarding the use and implementation of AI for the IC’s mission.
Leadership at the IC has gone on record saying that AI cannot and should not replace the role of the human analyst. Both Former NGA Director Robert Cardillo and former NSA Director Admiral (ret.) Michael Rogers said at the 2017 Intelligence and National Security Summit that as the bedrock of intelligence is credibility and trust, it requires keeping human analysts involved, and that the increased use of AI does not mean IC agencies should cease using human analysts.

An AI-enabled image recognition system could accurately identify an object as a missile but would struggle to tell a coherent story about what was happening. While AI may be able to reach a conclusion that is convincing, it is unable to show how it got the answer, unlike human analysts. In other words, it can only inform, not explain. At one of the CIA’s initiatives, experts found that in many cases, the AI analytics that had the most accurate results were also the ones with the least ability to explain how it got the answer.

The IC must begin acceleration of AI innovation and adoption wide-scale in order to amplify its human analysts and stay ahead of America’s adversaries. First, AI should begin to be integrated with TECHINT agencies, such as the NSA and NGA, that are experiencing data surpluses and then with all-source agencies that depend on the availability of information to make their analytic judgments.

The first step is training the AI in order for it to perform the tasks the IC has in mind. Most machine learning methods require large, high-quality, tagged data sets and the amount and type of data used to train it has great significance. For the IC’s purposes, the training data would likely require the use of classified information. Procedures will have to be set in place in order to protect that classified data as the AI is trained with it. AI training models are easily vulnerable to attacks on the training data where an adversary attempts to poison the system, something which AI researchers have yet to find a workable solution to protect against.

Second, there is the question of who does the training. The IC is lacking a significant degree of manpower in collecting and labeling data for machine learning purposes. While the same challenge is true for the private sector, the inherently secretive intelligence community cannot rely on crowdsourcing machine learning platforms. The IC would have to bring in or find trustworthy data scientists in the private sector who are approved to build and handle the classified training data. They would also need to take steps in mitigating the tendency for bias to sneak into the learning models, something noted in ODNI’s ethics framework.

Lastly, the IC will have to stand up multi-layered cloud computing environments. Cloud technology is the only way to achieve the huge computing power needed to run AI tools at the scale of U.S. intelligence operations. This new environment would cost tens of billions to build on top of current infrastructures.

In a world of growing data surpluses, how fast the IC can make sense of the information it collects will inevitably affect U.S. national security. In order to stay ahead of U.S. adversaries, and assure policymakers of having decision advantage, the IC must begin to fundamentally change the way it performs its duties in processing. IC agencies have a strong appetite for information, but in order to keep up with it, the IC will have to begin integrating AI into its intelligence cycle.

Find out how your University or college national security program can join The Cipher Brief as an Academic Incubator partner.

Read more expert-driven national security insight, analysis and perspective in The Cipher Brief

Related Articles