John J. Borek served as a strategic intelligence analyst for the U.S. Army and later as a civilian intelligence analyst in the U.S. Intelligence Community. He is currently an adjunct professor at Grand Canyon University where he teaches courses in governance and public policy. Divergent Options’ content does not contain information of an official nature nor does the content represent the official position of any government, any organization, or any group.
Title: Alternative Future: The Perils of Trading Artificial Intelligence for Analysis in the U.S. Intelligence Community
Date Originally Written: June 12, 2020.
Date Originally Published: August 12, 2020.
Author and / or Article Point of View: The article is written from the point of view of a U.S. Congressional inquiry excerpt into an intelligence failure and the loss of Taiwan to China in 2035.
Summary: The growing reliance on Artificial Intelligence (AI) to provide situational awareness and predictive analysis within the U.S. Intelligence Community (IC) resulted in an opportunity for China to execute a deception plan. This successful deception plan resulted in the sudden and complete loss of Taiwan’s independence in 2035.
Text: The U.S. transition away from humans performing intelligence analysis to the use of AI was an inevitable progression as the amount of data collected for analysis reached levels humans could not hope to manage[1] while machine learning and artificial neural networks developed simultaneously to the level they could match, if not outperform, human reasoning[2]. The integration of data scientists with analytic teams, which began in 2020, resulted in the attrition of of both regional and functional analysts and the transformation of the duties of those remaining to that of editor and briefer[3][4].
Initial successes in the transition led to increasing trust and complacency. The “Black Box” program demonstrated its first major success in identifying terrorist networks and forecasting terrorist actions fusing social media, network analysis, and clandestine collection; culminating in the successful preemption of the 2024 Freedom Tower attack. Moving beyond tactical successes, by 2026 Black Box was successfully analyzing climatological data, historical migration trends, and social behavior models to correctly forecast the sub-Saharan African drought and resulting instability, allowing the State Department to build a coalition of concerned nations and respond proactively to the event, mitigating human suffering and unrest.
The cost advantages and successes large and small resulted in the IC transitioning from a community of 17 coordinating analytic centers into a group of user agencies. In 2028, despite the concerns of this Committee, all analysis was centralized at the Office of the Director of National Intelligence under Black Box. Testimony at the time indicated that there was no longer any need for competitive or agency specific analysis, the algorithms of Black Box considered all likely possibilities more thoroughly and efficiently than human analysts could. Beginning that Fiscal Year the data scientists of the different agencies of the IC accessed Black Box for the analysis their decision makers needed. Also that year the coordination process for National intelligence Estimates and Intelligence Community Assessments was eliminated; as the intelligence and analysis was uniform across all agencies of government there was no longer any need for contentious, drawn out analytic sessions which only delayed delivery of the analysis to policy makers.
Regarding the current situation in the Pacific, there was never a doubt that China sought unification under its own terms with Taiwan, and the buildup and modernization of Chinese forces over the last several decades caused concern within both the U.S. and Taiwan governments[5]. This committee could find no fault with the priority that China had been given within the National Intelligence Priorities Framework. The roots of this intelligence failure lie in the IC inability to factor the possibility of deception into the algorithms of the Black Box program[6].
AI relies on machine learning, and it was well known that machines could learn biases based on the data that they were given and their algorithms[7][8]. Given the Chinese lead in AI development and applications, and their experience in using AI it to manage people and their perceptions[9][10], the Committee believes that the IC should have anticipated the potential for the virtual grooming of Black Box. As a result of this intelligence postmortem, we now know that four years before the loss of Taiwan the People’s Republic of China began their deception operation in earnest through the piecemeal release of false plans and strategy through multiple open and clandestine sources. As reported in the National Intelligence Estimate published just 6 months before the attack, China’s military modernization and procurement plan “confirmed” to Black Box that China was preparing to invade and reunify with Taiwan using overwhelming conventional military forces in 2043 to commemorate the 150th anniversary of Mao Zedong’s birth.
What was hidden from Black Box and the IC, was that China was also embarking on a parallel plan of adapting the lessons learned from Russia’s invasions of Georgia and the Ukraine. Using their own AI systems, China rehearsed and perfected a plan to use previously infiltrated special operations forces, airborne and heliborne forces, information warfare, and other asymmetric tactics to overcome Taiwan’s military superiority and geographic advantage. Individual training of these small units went unnoticed and was categorized as unremarkable and routine.
Three months prior to the October 2035 attack we now know that North Korea, at China’s request, began a series of escalating provocations in the Sea of Japan which alerted Black Box to a potential crisis and diverted U.S. military and diplomatic resources. At the same time, biometric tracking and media surveillance of key personalities in Taiwan that were previously identified as being crucial to a defense of the island was stepped up, allowing for their quick elimination by Chinese Special Operations Forces (SOF).
While we can’t determine with certainty when the first Chinese SOF infiltrated Taiwan, we know that by October 20, 2035 their forces were in place and Operation Homecoming received the final go-ahead from the Chinese President. The asymmetric tactics combined with limited precision kinetic strikes and the inability of the U.S. to respond due to their preoccupation 1,300 miles away resulted in a surprisingly quick collapse of Taiwanese resistance. Within five days enough conventional forces had been ferried to the island to secure China’s hold on it and make any attempt to liberate it untenable.
Unlike our 9/11 report which found that human analysts were unable to “connect the dots” of the information they had[11], we find that Black Box connected the dots too well. Deception is successful when it can either increase the “noise,” making it difficult to determine what is happening; or conversely by increasing the confidence in a wrong assessment[12]. Without community coordination or competing analysis provided by seasoned professional analysts, the assessment Black Box presented to policy makers was a perfect example of the latter.
Endnotes:
[1] Barnett, J. (2019, August 21). AI is breathing new life into the intelligence community. Fedscoop. Retrieved from https://www.fedscoop.com/artificial-intelligence-in-the-spying
[2] Silver, D., et al. (2016). Mastering the game of GO with deep neural networks and tree search. Nature, 529, 484-489. Retrieved from https://www.nature.com/articles/nature16961
[3] Gartin. G. W. (2019). The future of analysis. Studies in Intelligence, 63(2). Retrieved from https://www.cia.gov/library/center-for-the-study-of-intelligence/csi-publications/csi-studies/studies/vol-63-no-2/Future-of-Analysis.html
[4] Symon, P. B., & Tarapore, A. (2015, October 1). Defense intelligence in the age of big data. Joint Force Quarterly 79. Retrieved from https://ndupress.ndu.edu/Media/News/Article/621113/defense-intelligence-analysis-in-the-age-of-big-data
[5] Office of the Secretary of Defense. (2019). Annual report to Congress: Military and security developments involving the People’s Republic of China 2019. Retrieved from https://media.defense.gov/2019/May/02/2002127082/-1/-1/1/2019_CHINA_MILITARY_POWER_REPORT.pdf
[6] Knight, W. (2019). Tainted data can teach algorithms the wrong lessons. Wired. Retrieved from https://www.wired.com/story/tainted-data-teach-algorithms-wrong-lessons
[7] Boghani, P. (2019). Artificial intelligence can be biased. Here’s what you should know. PBS / Frontline Retrieved from https://www.pbs.org/wgbh/frontline/article/artificial-intelligence-algorithmic-bias-what-you-should-know
[8] Angwin, J., Larson, J., Mattu, S., & Kirchner, L. (2016). Machine bias. ProPublica. Retrieved from https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing
[9] Fanning, D., & Docherty, N. (2019). In the age of AI. PBS / Frontline. Retrieved from https://www.pbs.org/wgbh/frontline/film/in-the-age-of-ai
[10] Westerheide, F. (2020). China – the first artificial intelligence superpower. Forbes. Retrieved from https://www.forbes.com/sites/cognitiveworld/2020/01/14/china-artificial-intelligence-superpower/#794c7a52f053
[11] National Commission on Terrorist Attacks Upon the United States. (2004). The 9/11 Commission report. Retrieved from https://govinfo.library.unt.edu/911/report/911Report_Exec.htm
[12] Betts, R. K. (1980). Surprise despite warning: Why sudden attacks succeed. Political Science Quarterly 95(4), 551-572. Retrieved from https://www.jstor.org/stable/pdf/2150604.pdf