Discover more from Harding Project Substack
Military Writing and Artificial Intelligence
Unlock AI's potential with better archives
In this first guest post, Jason Atwell presents his thoughts on how improving archives can unleash responsible AI. Consider submitting 800-word Substack posts or 1,500+ word articles at firstname.lastname@example.org.
If the war in Ukraine is any indication, machine-learning and artificial intelligence (AI) will play an increasing role in battlefield command and control in the near future. Though AI is in its infancy, Ukraine has shown how to use both public and private data sources to direct drones and manage the information environment. Fortunately, the US Army enjoys lots of quantitative data on weapons, connectivity, and logistics to teach AI. However, it lacks qualitative data on questions of morale, doctrinal considerations, soldier feedback, and leadership.
The Harding Project’s goal of better archives for Army publications will not just unlock our intellectual legacy, but also unlock the data needed to teach responsible AI.
Future responsible AI models will require organized, qualitative data on questions of morale, doctrinal debate, adaptability, and leadership to provide better advice. This is where military writing can perform two critical roles as data. First, understanding history and context can better prepare for moments before and after AI models provide solutions. Second, models can use the body of military writing as a key data source in course of action analysis and decision-making. Machines won’t be able to “learn” unless they can be informed and instructed from resources that are insightful and relevant.
Professional writing: an AI data source
Tailoring existing models like ChatGPT or Bard for military applications will require some work. Right now, these models sometimes produce code with errors, repetition of urban legends instead of facts, and vague or formulaic responses to specific requests. This is primarily because most AI models at scale are based on the entirety of open-source data combined with language models based on common speech, usage, and queries.
However, the eccentricities of military English and unique military data means the AI that will underpin military models requires specific, closed language models and data sets. The military doesn’t talk like everyone else; rather, it assigns specific meanings to each term. Making things worse, much of the information the military relies on to make decisions is classified and not available on the internet. Fortunately, the military can easily integrate structured data on effective ranges, blast radiuses, cruising speeds, and fuel capacities.
What we do not have are the intangibles. Morale, unit history, the commander’s recent jog on Strava, weather’s effect on troop performance in similar circumstances, discussions of the pros and cons of offense and defense, and even, yes, whether the engaged parties have previously published their own thoughts on what to do in a specific scenario all add important information. This type of writing, historical and otherwise, could be critical data. The Army should set about cataloging and parsing it as such, so it can be available to military-centric language models in the near future.
Operationalizing military writing
By improving the archives, the Harding Project has the potential not just to reinvigorate current discourse, but draw on decades of prior thought leadership and scholarship instantly in new, specialized AI models.
Organizing the Army’s professional writing archives should not be hard. As the Army renews professional writing, it should also work with industry to adapt existing cataloging and search systems such as Google Scholar or JSTOR to meet the specific needs of the military by developing proprietary search functions and custom language models. Additionally, the Army must fully digitize the extensive professional journal and doctrinal archives. This will allow the Army to improve professional discourse while developing data repositories which can leverage past experience for future scenarios.
Improving the Army’s archives will return military history and writing to its proper place as a central component of decision-making on the battlefield–with the help of AI.