Approximate Reading Time: 13 Minutes
Thomas A. Drohan, Ph.D.
Director, International Center for Security and Leadership
JMark Services Inc.
Excerpt: In a quantum age of rapid technological advances and unprecedented information expansion, the critical capability for national security is to maintain strategic advantage in learning. An effective way to develop intelligence for learning in complex environments combines Socratic questioning, predictive analytics, and data visualization. This article illustrates how a cohort of analysts, planners and operators learn to stay ahead of complex problems through proactive inquiry and advanced analysis.
In a world of relentless data generation, human inquiry remains fundamental to learning. We create and control technology by asking questions. Ethical questions help us to clarify contentious values such as individual rights, state obligations, laws of armed conflict, and rules of engagement. Scientific questions enable us to develop testable knowledge such as Newton’s laws of motion, and Einstein’s theory of relativity. Questioning also reduces our vulnerability to mass-produced micro-targeted truths. Awash in the data of an information revolution, we must learn in an arena of hyper-connected ideas. Competition to develop, understand, and apply information is acute. Technological advantages are vital but temporary. The permanent critical capability is to create strategic advantages in learning. Whether it’s growing market revenue or dominating an operational domain, placing data into different contexts and analyzing information from multiple perspectives can link activities to goals. Purposeful questions are the key. Such intelligence for learning provides a critical yet uneasy strategic advantage in a dynamic environment.
For the Department of Defense, intelligence to learn requires no less than leveraging information to achieve enduring strategic outcomes. Technology continues to play a critical role. Advances in processing speed, storage capacity, and machine-learned tasks automate competitive processes such as detection, targeting, guidance, engagement, and assessment. Sequencing these functions into a kill-chain sequence increases the speed and accuracy of destruction. Network-enabled munitions re-order this sequence to provide more options for multi-domain battle. As this combined arms element of security strategy matures into integrated multi-domain operations, we need to create strategic effect-chains not just kill-chains. In order to achieve persistent outcomes, we will need to orchestrate combinations of diplomatic, informational, military, economic, and social effects. This imperative is pointedly applicable to the information environment.
Information operations are non-stop and their effects broadly diverse. They, therefore, require exceptionally agile learning cycles. With strategic effects in mind, the Joint Concept for Operations in the Information Environment calls for moving beyond a transmission-centric approach that emphasizes data processing, storage capacity, and command and control. These tasks are necessary but insufficient for succeeding in the modern information environment. We also need to understand how information is accessed and assigned meaning. Complicating this task is increased uncertainty, despite having more data. Every operation creates information effects whose consequences that are hard to predict due to entangled causes and effects.
Such pervasive uncertainty focuses attention on the present. This current operations myopia feeds a sense of immediate purpose but creates two results that kill effective strategy: (a) insufficient reflection on the past; and (b) reluctance to anticipate the future. An urgent presentism paralyzes risk-averse organizations, a vulnerability that adversaries are keen to exploit. While some leaders push the need to learn and innovate, new capabilities often lag superior operating concepts. The gap has severe implications for quantum age learning.
Quantum-based technology differs from bit (binary digit: 0 or 1)-based technology due to super-positional states that can be simultaneously 0 and 1. Once computer chips contain many qubits (quantum bits) and few errors, we will need to learn new concepts and apply them in new contexts. Repeatedly and rapidly so, because quantum computing is expected to consume, create and process data at exponentially higher speeds. Beginning with more accurate clocks and better encryption/de-encryption, quantum improvements will disrupt legacy practices. How can we manage breakthrough technology?
Human-directed machine learning can play a vital role in preparing us for quantum age data-driven, human-directed operations. Data visualization in particular helps us ask questions to explore beliefs and acquire knowledge.
Questions such as, how should we and can we design neural networks with algorithms that collectively synthesize new information? Imagine an adversary generating and accessing more accurate information more quickly as automated intelligence for targeting. If we can identify data used to train the algorithms, and measure the accuracy of the synthesized information, we could determine conditions in which the machine learning performs better.
Another question is when to insert human decisions into automated processes. This issue is not just a matter of technical efficiency; it’s also a value judgment. Values vary among political systems and cultures. As developments in miniaturized processors demonstrate, speed and accuracy can create advantages in warfare such as precision destruction, timely jamming, and global mobilization of recruits. Whether the advantages are decisive is not determined by technology. Success also depends on crafting, adapting, and sustaining a sound strategy. Morally acceptable strategies differ, too. Systematic controls on social media in authoritarian states are illegitimate in democracies. Yet, democracies lack the will to sacrifice individualistic values in order counter net-wars on its own citizens. So, we need to assess what success and failure look like in different contexts and from different perspectives. Machine learning can provide data and information needed to exercise our human prerogative of intelligence, that of determining meaning.
To prepare for the hyper-speed encrypted data of quantum processing, we offer an example of visualized machine learning. As in joint operations design, all a manager, commander or supervisor needs to begin is the flexibility to organize activities in order to achieve desired outcomes. In the program we describe, desired outcomes are a mix of three types — knowledge, skills, and attitudes. Strategists looking to leverage machine learning for other purposes could replace these with any number of preventive and causative outcomes such as deter, defend, dissuade, coerce, compel, persuade, and so forth. Similarly, “learner” may take the form of “analyst” or “planner” or “operator;” and “instructor” may be “commander.”
Effective teaching employs learner-oriented approaches that engage people to achieve desired outcomes. We are interested in how to integrate human-directed machine learning with traditional methods because we want to confront learners with more data than there is time to read, and more than is available. Dynamic environments present conflicting information for framing problems. Most narratives are constructed from selective or false information to resonate with familiar frames of reference. So, we need to understand the filters, frameworks, algorithms, and concepts that shape information and meaning.
One way to prepare for advanced learning is to blend Socratic questioning with machine learning. The case is an intensive two-week program. The basic method of asking questions is applicable to any effort that can spare 30-minute sessions. We figure this should be enough time to organize a lesson that addresses two key tasks: (1) What do you want students to learn? (2) How do you know that they learned it? These questions orient you, the instructor, on realistic goals and how to assess success before you decide what to do (similar to John Boyd’s briefings on observe, orient, decide, and act).
Answering these two questions in the time available requires paring down course learning outcomes and lesson learning outcomes to what is achievable. To learn how to learn, it’s better to go deep and broad with less learning outcomes. The tendency, however, is just the opposite. For budgetary or bureaucratic reasons, instructors dense-pack a gluttony of outcomes and force feed them to students. On the job, intelligence analysts are prone to feel this pressure of premature closure. This learning environment does not work well unless you want your intelligence of learning to be tedious, temporary, and at the recall level of thinking. Not likely to out-think smart competitors. With too much to teach in the time available, learning outcomes need to be high in skills and attitude, and with just enough knowledge outcomes to enable further development. Why?
In a data-saturated world, we need to evaluate the quality of data, and acquire knowledge that matters most. We need to establish priorities, and deliberately develop desired proficiencies through assessed practices. There are many ways to design a learner-centered course. Whatever the design, we need to assess effectiveness as we teach so we can improve at the speed of relevance. We can make adjustments by adding a third question: (3) What will you do differently next time? Answering this question after each lesson takes only a few minutes; well worth the time for a timely feedback loop.
Our case is that of a joint-certified course, Information Environment Advanced Analysis. The purpose of this program is to equip intelligence, operations, and plans personnel with analytical concepts and operational constructs to characterize, forecast, target, war-game, and assess the information environment. Each course contains a scenario filled with decision maker profiles that enable students to adopt various perspectives.
Instructors are experienced leaders, educators, trainers, and operators. Their teaching approach is to facilitate learning by using personalized interaction. This means asking questions not feeding answers. The typical day consists of morning and afternoon plenary sessions of all students followed by small groups. Teaching methods are a mix of guided discussion, case method, seminar, and mentoring. Plenaries are concept-heavy presentations, while small groups focus on tasks. Mentors and students draw illustrations to characterize concepts. Students create their own depictions of the information environment and how they plan to set new conditions. Our predicament is that there is so much data to process into information, contextualize into intelligence, and acquire knowledge. This is where machine learning comes in.
Machine learning systems provide search and discovery solutions using iterative processes that learn by ingesting data and continuously refining understanding of that data. The advanced analytics program used in this course, Savant X, helps students create predictive hypotheses by visualizing and discovering relationships across disparate, open source data sets.
Students start by decomposing the scenario into nodes and linkages. Linkage analysis is foundational for other types of analysis. Students map diverse connections and search for patterns, trends, and anomalies. They hand-draw connections found in assigned readings, which engages mental and motor-skills and stokes creativity. Mentors do the same. However creative and visually appealing, the artwork is data-limited. With advanced predictive analytics able to display more connections among data, we can examine the highlights of bigger data. Visualized exploitation sparks curiosity, which drives critical thinking skills and an inquiring attitude. These aspects of character are needed just to compete in the information environment.
The following visualizations illustrate how a cohort of analysts, planners and operators query data to understand relationships and ideas in a realistic scenario. Rather than culminating in a capstone exercise, the scenario winds throughout the course as a live practical exercise. Just as imagery and geospatial analysis lead to targeting and weaponeering, human-directed data visualization via advanced analytics leads to discussing target audiences and how to influence them. Seeing spawns thinking. Machine-learning discovers more “unknowns,” beginning with data relationships or linkages.
To baseline the situation, students select individuals from the scenario. These roles may include politicians, social influencers, business leaders, military commanders, and government officials. The first picture presents a mess of nodes and relationships among people, places, things and concepts. Differently sized spheres are connected by strands of varying thickness. The 3-D view can be rotated to experience different perspectives. Students question and decide which nodes to explore, decomposing the information into networked relationships to be actioned. Students discuss what the information means based on factors such as context, assumptions, logic, and evidence.
Noticing nodes in the picture that refer to contested areas and armed groups, students investigate them based on their hypothesis: territories and cyberspace are test arenas for hybrid warfare in other regions populated by variously vulnerable target audiences.
Selecting these nodes reduces the picture to five nodes. Students investigate highlighted passages from the associated database of text. This inquiry reveals evidence of how specific individuals, groups, and concepts are connected. Such detailed information, machine-processed from the data, is possible because each node contains text-based information. The analytics engine provides content from the database related to that node.
Pictures of spheres and strands are based on hyper-dimensional relationship analysis (HYDRA) that displays 5 dimensions of spatial relationships. The first three dimensions are represented by the orange spheres, while the remaining two dimensions are represented by the pink spheres that emerge after selecting an orange sphere (the darker sphere below) to investigate.
Five spatial dimensions of relationships may be difficult to conceptualize, so consider the satire, Flatland, published in 1884 by Edwin A. Abbott in which he describes the difficulty of explaining three-dimensional objects to residents of a two-dimensional world. HYDRA’s visualization of multiple dimensions provides new insights for data and concept analysis, revealing hidden “unknowns” by showing relationships within data sets.
Next, students query connections between two nodes that they selected because the nodes are large and close to each other. As students examine highlighted text, they evaluate evidence of operational planning. They spot a node in the initial complex picture that relates to a plan. Reducing the nodes to the minimum number that retains this node in the picture renders the following picture:
In successive visualizations, students explore connections among actors and concepts from different perspectives (Red team, Blue team, Red’s view of Blue, Blue’s view of Red’s view of Blue, etc.). Analyzing the depictions and their related text, students discern details of a possible plan to conduct hybrid operations that involve security-related goals, conventional and unconventional forces, information operations, secessionist political-military groups, ethnic unrest, and nefarious measures to prevent external interference.
Daily readings develop the scenario with significant activities. Mentors inject inputs based on student responses and current events. Students further decompose the data and question what’s going on as they learn more types of analysis. As they apply anticipatory analysis, students recompose the data by forecasting what could happen. Then they synthesize information they deem relevant into a new whole. This effort requires going beyond summarizing what has happened—students learn to take intellectual risk (which also can be personal and professional) by making assumptions and evaluating evidence. They also practice taking responsibility and exercising judgment. In this case, students construct an operational design of an adversary’s plan to use traditional and social media, military forces, criminal groups, economic incentives and proxy agents to exploit historical issues and ethnic divisions, and to establish de facto territorial control.
The learning power of visualized data exploitation and inquiry is clear. We explore and experience ways to analyze known, previously unknown, and emergent linkages and systems across all domains. With visualized data informing and re-shaping baseline information, we design ways to influence relationships by creating effects to achieve objectives toward the desired end-state. End states are dynamic, and nothing is predetermined. Visualizations serve an on-going process of adaptive and proactive questioning. We test hypotheses, inform collection requirements, and provide recommendations to decision makers. We characterize current and future conditions through hand-drawn and machine-generated visualizations. As machine-human interface technologies improve and blur that very distinction, we continue to evaluate best learning methods.
Visualized analytics are well suited to enhance learning through discovery. The key is to keep asking purposeful questions. While we attempt to reduce the variability of uncertainty, we expect uncertainty to exist. Fundamentally, we direct machine-speed learning by asking ethical and scientific questions. Inquiry also maintains awareness and control of human-speed learning such as emotional reactions, intuitive hunches, reflective thinking, and moral judgments. A looming question is, has artificial intelligence technology attained human-level cognition and in what areas, and if not, when? Thus, the ultimate reason to learn how to create strategic advantages in learning is to ensure human control of an information environment increasingly filled with machine-learned shaping of our perceptions.
Brig Gen (ret) Tom Drohan, PhD, served for 38 years in the US Air Force in combat rescue and tactical airlift operations, and as permanent professor of military and strategic studies at the US Air Force Academy. He has served as Council on Foreign Relations Fellow in Japan, Curriculum Advisor at the National Military Academy of Afghanistan, Visiting Scholar at the Reischauer Center for East Asian Studies, The Johns Hopkins School of International Studies, and as Dean of the United Arab Emirates National Defense College. He is author of the book, A New Strategy for Complex Warfare: Combined Effects in East Asia (Cambria Press, 2016). Currently, he is the Director of the Center for Security and Leadership at JMark Services Inc.
The views expressed are those of the author and do not necessarily reflect the official policy or position of the Department of the Air Force or the U.S. Government.