Leveraging Machine Learning and Artificial Intelligence in the Competition Era
GySgt Jeremy Kofsky
In the summer of 2001, the lead singer of the Nickelback, Chad Kroeger, came up with a plan. He started “studying every piece, everything sonically, everything lyrically, everything musically, chord structure. I would dissect every single song that I would hear on the radio or every song that had ever done well on a chart and I would say, ‘Why did this do well?'”[i] The information Mr. Kroeger obtained during this research period allowed the band’s next five studio albums to be certified Platinum, signifying the musical computation Nickelback created led to sustained commercial, if not necessarily critical, success.
Nearly two decades on, Kroeger’s analytic efforts are familiar to anyone who has heard the terms artificial Intelligence (AI) or machine learning (ML). Modern applications for AI and ML abound, from human resources to medical research. Any field where interpreting large volumes of data might be useful, AI and machine learning have entered the conversation. Military intelligence operations are no exception. Indeed, the U.S. Marine Corps has rapidly and enthusiastically embraced AI and machine learning applications[ii], but the service would do well to remember the lessons Nickelback’s ascendance offers: all the data in the world requires the “artists touch.” For intelligence operations that means having analysts on the loop to match the science of the song to the art of analysis.
Who is the Artistic Analyst?
The Marine Corps primarily uses its Intelligence Specialists (MOS 0231) and Intelligence Analysts (MOS 0239) during the Production Phase of the intelligence cycle.[iii] The production phase is further broken down into four stages: Filtering, Recording, Evaluating, and Product Preparation. During all four of these stages the artistic analyst can leverage the various types of AI and machine learning.
To understand how, it is critical to understand a few key definitions, namely, what is AI and machine learning? According to researchers at Northeastern, “On a basic level, artificial intelligence is where a machine seems human-like and can imitate human behavior.”[iv] Meanwhile Machine Learning (ML) “is where machines are taking in data and learning things about the world that would be difficult for humans to do. ML can go beyond human intelligence.”[v]
There are two main types of AI classification, based on their ability to perform functions replicative to those a human brain can perform along this large delta is Functionality AI, or that can that replicate the mechanisms of thought and Capability based AI, or that which can replicate actual intelligence and learning. Functionality Based AI consists of Reactive Machines (the most basic form of Artificial Intelligence), Limited Theory (current level of AI wherein machines can use past experiences to inform calculations), and Theory of Mind and Self-Aware AI, (still in the theoretical stage of development).[vi] The artistic analyst’s current focus will be based on Functionality vice the more theoretical Capability Based Systems, outside of Artificial Narrow Intelligence (ANI), which encompasses all of the AI Applications currently working. Artificial General Intelligence is the next evolution on the Capability classification wherein a system learns, understands, and performs function as a normal human would. Artificial Super Intelligence is the pinnacle of AI development and would start to replace humans as decision makers due to superior processing.
Though nuanced, understanding the definitional differences in AI and machine learning provide concrete understanding the artistic analyst will need to probably understand how to use the various tools available to them in the Future Intelligence Operating Environment. The ability to link the various systems, logics, and programmable vice ANI level applications will allow for a proper mixing and remixing of the five step Production Process to create better products that meet the success metric of properly informing the Commander and keeping Marines safe.
Marine Corps Warfighting Publication 1-4 (MCWP 1-4) ‘Competing’ , the doctrinal cornerstone of the Marine Corps current force transformation, specifically speaks to a new world of information overload, allowing adversary forces to create a ‘salami-slicing’ effect wherein they make almost imperceptible changes in the status quo so as to not arouse suspicion nor reaction; unless an adversary can transform that reaction into a propaganda victory by framing it as excessive[vii] [viii] The proper application of AI and ML in analysis supports the analyst’s identifying, cataloguing, and analyzing these salami-slices, to create overlap and a longer lasting utilization of these tools within the artistic Analyst’s realm to use as seen fit.
Producing the (Artificial) Intelligence Machine (Learning)
Filtering. Filtering allows the Intelligence Specialist to discard irrelevant or repetitive information before it enters the production process.[ix] This is the best time to use of Limited Theory (LT) AI Applications as they have memory capabilities and can understand what the Intelligence Specialist is looking for based on data inputs. These machines’ memory storage enables their learning mechanisms and can be leveraged to ‘learn’ what are indications and warnings (I&W) for Competition level events. By understanding and layering various ‘maps’ of data, theory, actual maps, and events, the LT AI can understand that a report from NASA’s Fire Information for Resource Management System (FIRMS) map application detailing a suspected fire[x] in a random village in Southern Central Africa could be a precursor to a genocide-type event, similar to the initial events during the Rwandan Hutu-Tutsi Civil War of 1994.[xi] Similarly, the recent Russian invasion of Ukraine has demonstrated several applications of indications of Publicly Available Information (PAI), which has been cross-referenced with known Russian Orders of Battle and other PAI to create a more succinct picture of the activities on the ground, leading to strikes and desired actual military effects. [xii].
More data and better ‘curation’ of the various AI applications’ ingested data will yield better results capable of better predictions of potential issues unseen to the human mind. The nebulous characterization of threats in the Competition Domain portend humans dismissing incidents AI can identify as important. This can be caused by the incident simply not being at the level of reporting necessary to warrant human interest. Interactive AI therefore acts as the proverbial ‘canary in the coal mine’ to warn of potential issues to trigger more in-depth analysis, therefore allowing the best way to understand and potentially negate the ‘salami slicing’ efforts of adversaries. While a human is only a quick as their brain’s ability to ingest, and more importantly, understand information, to the LT AI, it is simply another line of code iterating over known data sets and understood theories to measure against for I&W of those ‘salami slicing’ events. It also presents a catalogued library through the Recording Process to learn from later inputted data and programming in case it gathered the wrong information or interpreted it differently in hindsight, thereby improving the overall efficacy of the program.
Recording. Once the curated filtering process has occurred, a Reactive Machine (RM) AI program, similar to IBM’s Deep Blue[xiii], will reduce the information to core writing or graphical representations, subsequently arrange that information into groups of related items based on the proper application of Fuzzy Logic, which is a term for an enhanced Boolean Logic[xiv] wherein multiple factors are encapsulated outside of the traditional binary coding of most computer programs.[xv] In addition to streamlining the actual cataloging of information, the RM AI can register, appropriately log with programmable naming conventions, and draft those into folders meeting various Commander Priority Information Requirements (PIRs). Artistic Analysts will serve as overseers and final touch points of emphasis on certain modalities of information display the RM AI will need to present its data. The RM should be able to take one report, understand its keywords, cross-reference those with similar words or reports, understand any unseen connections, and be able to represent those data points in any medium. The analyst will then proceed to have the RM ‘show its work’ in a Structured Analytical Technique, most likely the Contrarian Model, wherein their job will be to prove the RM wrong. By taking a negative approach to the review of Recorded Data, the Analyst can seamlessly transition into the Evaluation stage.
Evaluating. Evaluation allows the Scientific Analyst to overcome traditional logical fallacies of omission and assumption against a backdrop of mitigating the biases of Culture, Organization, Personal, and Cognitive. This stage is intuitively more people than program run and is critical to ensuring the computer is not being fooled by Deception Operations and/or the human is overcoming the Logical Fallacies. Having a computer running an ANI application will allow the Scientific Analyst to have narrowly defined specific tasks handled just like a human. An ANI program would have to be ‘proven’ the Analyst did not use Omission or Assumption in their Analysis or Recording of Information. This can be accomplished by allowing the ANI access to the RM AI’s system and therefore the data they possess. This will be bolstered by overlying the biases of humans and learning the biases of the Analyst themselves through operational history and training events. In the case of any disagreement, the senior analyst can make the end point as they have experience in the nuances of intelligence aggregation, cataloguing, and evaluation and can therefore render informed decisions.
Product Preparation. Thebest information and analysis is fruitless unless the applicable ‘artist touch’ is used to present that information in a manner that is timely, accurate, and tailored to both the unit and its mission.[xvi] The use of a system that can use curated feedback from commanders to their analysts can make information more digestible, be it graphical (charts and graphs), written (paper and studies), or some combination of the two (map overlays & estimates). This feedback will allow programs to standardize products in near-real time, in turn facilitating the hallmarks of rapid preparation, mutual support between intelligence sections, ease of dissemination, and, most importantly, familiarity for the user.[xvii]
Rise of the Artificial Machines?
While one can look at the preceding event cycle and conclude the Marine Corps can simply get rid of the human/error factor and replace better and better coded machines to conduct ‘perfect’ analysis, the truth is more complicated. While computers do make our lives much easier, they are only as good as the programming they have and the data they ingest. They are a great way to ‘sanity check’ and commit to the Three D’s ‘Dirty, Dull, and Dangerous’ work of Intelligence Operations. This does not alleviate the commander from making the best-informed decision to protect their Marines and complete their assigned mission. The best manner in which to this is to create a symbiotic relationship were the Venn diagram of what they do very closely overlaps in most areas but leans on what each does best. In this way, much like how Chad Kroeger was able to bring together, separate, and analyze mass amounts of data, but also had the cultural zeitgeist knowledge to properly execute on the end state, an Analyst can flourish and too become a Rockstar (Analyst).
GySgt Jeremy Kofsky is a 19 year veteran of the Marine Corps with small unit experience over 12 deployments in 5 continents. He is the first enlisted Brute Krulak Scholar, a Corporal Garrett Jones Senior Research Fellow, and the recipient of the 2023 enlisted award for Expeditionary Warfare Excellence.
[i] “Nickelback Biography. https://www.notablebiographies.com/newsmakers2/2007-Li-Pr/Nickelback.html.
[ii] John Grady. ‘Artificial Intelligence at Core of Marine Officers’ ‘Big Ideas’ for Future of Force’. USNI News. https://news.usni.org/2020/06/02/artificial-intelligence-at-core-of-marine-officers-big-ideas-for-future-of-force. 2 June 2020.
[iii] MCTP 2-10B ‘MAGTF Intelligence Production and Analysis’. 2 May 2016.
[iv] Kristin Burnham. “Artificial Intelligence vs. Machine Learning: What’s the Difference?” https://www.northeastern.edu/graduate/blog/artificial-intelligence-vs-machine-learning-whats-the-difference/ 6 May 2020.
[v] Ibid.
[vi] Andrés Páez. “The pragmatic turn in explainable artificial intelligence (XAI).” Minds and Machines 29, no. 3 (2019): 441-459.
[vii] Yanzhong Huang. ‘America’s Political Immune System Is Overreacting to China.’ https://foreignpolicy.com/2020/09/08/america-overreacting-to-china-political-immune-system/ 8 September 2020.
[viii] American Perceptions Of China. https://china.usc.edu/american-perceptions-china/ 3 March 2022.
[ix] MCTP 2-10B.
[x] Daily Kos Staff. “Ukraine update: Something *big* is happening, as the Battle of the Izyum Salient begins” https://www.dailykos.com/stories/2022/5/14/2097948/-Ukraine-update-Something-big-is-happening-as-the-Battle-of-the-Izyum-Salient-begins May 14, 2022.
[xi] Melvern, Linda (2004). Conspiracy to Murder: The Rwandan Genocide. London and New York: Verso. ISBN 978-1-85984-588-2.
[xii] Janes. “Janes Analysis: Ukraine Conflict” https://www.janes.com/defence-news/ukraine-conflict. 23 May 2022.
[xiii] Haesik Kim. “Historical Sketch of Artificial Intelligence.” In Artificial Intelligence for 6G, pp. 3-14. Springer, Cham, 2022.
[xiv] Alex Tserkovny. “Some Considerations about Fuzzy Logic Based Decision Making by Autonomous Intelligent Actor.” Journal of Software Engineering and Applications 15, no. 2 (2022): 19-58.
[xv] Toshendra Sharma. Fuzzy Logic: What It Is and Some Real-Life Applications. https://www.globaltechcouncil.org/artificial-intelligence/fuzzy-logic-what-it-is-and-some-real-life-applications/. 29 May 2020.
[xvi] MCTP 2-10B.
[xvii] Ibid.
The post I Wanna Be a Rockstar (Analyst) appeared first on Key Terrain Cyber.
