Table Of Contents
This article delves into Raven Sentry, an innovative project that harnessed artificial intelligence to provide early warnings of insurgent attacks in Afghanistan. Between 2019 and 2020, the Resolute Support Deputy Chief of Staff for Intelligence (J2) leveraged a culture of innovation, the urgency of the US drawdown, and a talented team supported by commercial sector experts to develop an AI system that successfully predicted attacks. Although the project was cut short by the war’s end, it offers critical insights into AI’s potential and the conditions necessary for successful innovation.
The Genesis of Innovation in Warfare
Historian A. J. P. Taylor famously stated that “war has always been the mother of invention.” While this sentiment often brings to mind the advent of tanks in World War I or the atomic bomb in World War II, it is equally applicable to the 21st-century conflicts in Afghanistan and Iraq. In these modern theaters, soldiers, sailors, airmen, and marines have continuously innovated, notably through the integration of artificial intelligence (AI).
As US and NATO forces began their drawdown in Afghanistan, the Deputy Chief of Staff for Intelligence (J2) faced the daunting task of maintaining situational awareness with diminished resources. The Resolute Support headquarters in Kabul, known for its openness to emerging concepts, assembled a unique group of personnel adept at leveraging AI. Supported by experts from the Department of Defense (DoD) and the commercial sector, this team embarked on developing an AI model named Raven Sentry, designed to predict future attacks using unclassified data sources.
The Challenge
By 2019 and 2020, with the reduction of US and coalition forces in Afghanistan, maintaining a robust human intelligence (HUMINT) network became increasingly challenging. The coalition’s traditional methods of intelligence gathering, which included aircraft-mounted collection platforms and extensive ground-based networks, were becoming less feasible. As touchpoints with the local population decreased and intelligence-gathering aircraft were reassigned, the ability to maintain situational awareness deteriorated. Insurgents capitalized on these gaps, launching attacks that undermined the credibility of the Afghan government.
Turning to AI for Solutions
In late summer and fall of 2019, as the United States approached a withdrawal agreement with the Taliban, intelligence officers at Resolute Support and the Special Operations Joint Task Force-Afghanistan (SOJTF-A) sought innovative solutions. They were informed by the Intelligence Community (IC) that AI-enabled warning models were making strides in enhancing analytical processes. The intelligence team assessed that a well-designed AI model could recognize insurgent patterns and predict future attacks by processing open-source intelligence (OSINT).
Developing Raven Sentry
The development of Raven Sentry required assembling a skilled workforce and securing commercial sector support. In late 2019, the intelligence leadership consolidated efforts, forming an innovation team at the special operations headquarters. This environment was conducive to experimentation, with leaders willing to tolerate early failures. The team, affectionately dubbed the “nerd locker,” comprised analysts who pulled shifts on the operations floor to build trust and understand operational needs.
The team’s first task was creating a detailed event matrix for district and provincial center attacks. Drawing on years of experience and historical data, they identified patterns and environmental factors influencing insurgent behavior. This data was manually curated and formatted for the AI model, which was designed to process unclassified geospatial data, OSINT reporting, and global information systems (GIS) data sets.
Overcoming Challenges
The development of Raven Sentry was not without obstacles. Data curation proved challenging, requiring significant time and effort to format historical events into machine-readable data. The team worked closely with Afghan military personnel to gain cultural context and identify additional warning signatures. With the support of the Defense Innovation Unit in Silicon Valley, they identified an industry partner capable of developing the model.
Despite bureaucratic hurdles and skepticism within the intelligence community, the team persevered. By October 2020, Raven Sentry had achieved approximately 70% accuracy in predicting attacks, providing valuable insights to analysts.
Lessons Learned
Raven Sentry’s development and deployment offer several lessons for future AI initiatives:
- Leadership and Command Culture: Success hinges on a culture committed to innovation and experimentation. Leaders must prioritize resources and tolerate early failures.
- Human-Machine Teaming: Effective AI deployment requires skilled personnel who understand data and machine learning. Analysts must communicate the system’s outputs to operators and leaders.
- Unclassified Data: Commercially produced, unclassified information can yield valuable predictive intelligence, facilitating collaboration with foreign partners and the commercial sector.
- Continuous Maintenance: AI models require ongoing updates and maintenance to remain effective. Dedicated personnel must continually adapt the algorithm and data inputs to evolving environments.
Conclusion
The Raven Sentry project underscores the transformative potential of AI in military intelligence. While the initiative was cut short by the war’s end, its success demonstrates the importance of leadership, collaboration, and innovative thinking. As advancements in AI continue, the Joint Force must educate its leaders, balance technological speed with human intuition, and create ecosystems conducive to innovation.
For further insights, the full article on Raven Sentry can be found in the Summer 2024 issue of Parameters. and the podcast.
