In August, the Department of Defense (DoD) took a significant step forward in harnessing the power of artificial intelligence (AI) by establishing a Generative AI Task Force. This task force advises the department on leveraging AI across various domains, from warfighting to healthcare and business affairs. One of its key missions is to explore the use of foundation models in supporting Secretary of Defense Lloyd Austin’s growing priority: understanding, predicting, and preventing civilian harm.
The power of foundation models
Foundation models, like ChatGPT, are a breed apart in AI. Unlike specialized models designed for narrow tasks, foundation models are trained on vast datasets encompassing a wide range of information from the internet. They can engage in conversations, generate creative text, translate languages, and more, making them versatile tools. While they may lack human-like subjective experiences and contextual reasoning, they can recognize patterns and make predictions, which can be instrumental in preventing civilian harm.
Addressing analytic shortfalls
One of the significant challenges the DoD faces is its ability to protect civilians during conflicts. A 2021 report by the RAND Corporation identified a key issue: a lack of data and technology for detecting potential risks to civilians and verifying reports of harm. Threats to civilians escalate rapidly during conflict, making it difficult for human cognition alone to grasp the full scope of harm. Foundation models can bridge this gap by collating data from various sources and highlighting patterns that indicate potential risks or past harm to civilians.
Harnessing social media analysis
Foundation models, particularly Large Language Models (LLMs), excel in sifting through vast social media data. In conflict zones like Ukraine, civilians often provide real-time information on emerging threats and harms on platforms like Twitter, Diia, and eVorog. LLMs can analyze this data, spot patterns, and provide valuable insights for predicting and preventing harm. Their ability to decipher messages filled with local nuances and evolving dialects enhances their predictive capabilities.
Enhancing situational awareness
Combining satellite imagery with foundation models like Vision Transformers can transform static data into dynamic and informative insights. These models don’t just analyze isolated parts of an image but understand the interconnectedness of segments. Military planners can use this capability to decode patterns, trace historical trends, and identify potential civilian congregation points, thus gaining a deeper understanding of the ground reality.
Augmenting human decision-making
While human judgment remains invaluable in military decision-making, foundation models can address challenges associated with rapidly processing vast amounts of real-time data and identifying nuanced patterns. Recent incidents like the drone strike in Kabul and civilian casualties during battles underscore the need for improved decision-making. Foundation models serve as a “gut check,” cross-referencing current decisions against historical data, identifying potential biases, and flagging anomalies. This AI-driven layer of verification can significantly improve targeting accuracy and minimize tragic errors.
Unicorn technologies: A paradigm shift
Foundation models represent the gateway to what some call “unicorn technologies.” Imagine a conflict simulator enhanced by these models, capturing tangible events, socio-cultural dynamics, and historical contexts. Moreover, foundation models go beyond mere translation; they bridge cultural understanding gaps, facilitating more nuanced interactions in conflict zones. However, acknowledging their limitations, including potential algorithmic bias and data omission, is essential.
Challenges and ethical considerations
Foundation models are not without their challenges. Vulnerable groups are less likely to be active online, and disinformation remains a significant concern. Without proper screening, these models could amplify false narratives or fall victim to adversarial techniques. Hence, critical decisions, especially those impacting lives, must continue to center on human judgment while leveraging AI as a valuable tool.
The establishment of the Generative AI Task Force by the Department of Defense marks a significant step toward harnessing the power of AI and foundation models in reducing civilian harm. These models offer a new dimension to conflict management, providing rapid analysis, enhancing situational awareness, and improving decision-making. While they hold tremendous potential, they should be used judiciously with human expertise. As we embrace this future, we commit to a vision where technology becomes an ally in warfare and a guardian of humanitarian values.
