By Dan Steiner, Orion Mapping; Todd M. Bacastow, Maxar Technologies; Dr. Todd S. Bacastow, Stephen Handwerk, Dr. Gregory Thomas, and Stevie Rocco, Penn State; and Laura Strater, Raytheon
This article explores the use of artificial intelligence (AI) in the teaching and learning of geospatial intelligence (GEOINT) and suggests future applications of AI to improve education in this field. We recommend future steps in the adoption of AI for teaching, learning, student support, and administration. Finally, we conclude with possible future research themes.
GEOINT education is rapidly changing as the discipline evolves, though it’s not changing nearly as fast as the enabling technology. AI is progressing at an accelerated pace with rapidly increasing numbers of sensors, and amounts of data and computing power, which broadly impact the nature of education. Despite high expectations and lots of optimism for the potential benefits of AI, history shows that technology, no matter how well-intended, can have unintended consequences. There is a benefit to stepping back, assessing where we are and in what direction we are going, to consider these consequences. David Thornburg, a futurist and education consultant, contends that the promise of computers in education disappeared due to the loss of creativity associated with the shift from programming your own software to shrink-wrapped software. This article considers the impact of the AI revolution on GEOINT education and suggests a path forward.
Why use AI in the GEOINT educational community? There are many reasons, but all come back to the fundamental issue that gaining GEOINT expertise is exceptionally difficult and time-consuming. This is supported in the emerging recognition that the core of GEOINT is less a profession or industry and more of a broad set of embodied, effective, and organizational practices in which people create, use, and employ sophisticated geographic understandings and knowledge to make complex decisions. As a rule of thumb, it takes 10 years of progressively more difficult individual practice to achieve expertise in GEOINT, in large part because GEOINT is a meta-discipline with limits of know-how that are largely unbounded by typical academic discipline definitions. It is a discipline in which one must achieve mastery in many fields in addition to the context-laden tradecraft that “glues” this knowledge together. Technologies like AI may shorten the path to expertise in several ways, including identifying individual strengths and weaknesses across meta-disciplines, which has the potential to be a game-changer for GEOINT education.
The State of AI-enabled GEOINT Education
These are exciting times for the purveyors of AI. Bringing together 4,045 attendees and 216 vendors, USGIF’s GEOINT 2019 Symposium’s theme was “Human-Machine Teaming & Innovation Yield Mission Success.” The event focused on machine learning, a statistics-driven form of AI, as a tool to assist human decision-making. Attending the conference, one would think that AI has penetrated GEOINT education. But before attempting to “ride the AI wave” it is important to take a careful look at the GEOINT community. While artificial intelligence in education (AIED) has been developing for more than 25 years, upon closer examination, AI’s role in educating the GEOINT analyst appears to be limited today. It seems that the term AI is so inconsistently defined that it has become ambiguous as a description of a technology, let alone how it’s applied in education. The use of the term AI varies so much we risk it simply becoming a vague marketing buzzword. Herewith, then, is a brief summary of AI, machine learning, and deep learning.
MIT professor Marvin Minsky, who is called the father of artificial intelligence, defined AI as “the science of making machines do those things that would be considered intelligent if they were done by people.” Specific to this article, we use the definition that AI is intelligent behavior by machines as compared to the natural intelligence of humans. Machine learning (ML) is a field of computer science that gives computers the ability to learn without being explicitly programmed, typically through statistical-based methods. Deep learning (DL) is part of a broader family of machine learning methods based on learning data representations with convolutional neural networks, as opposed to task-specific algorithms. Learning can be supervised, partially supervised, or unsupervised. But, what does this mean in real-world applications?
- A computer’s human-like ability to recognize patterns, and then integrate similar patterns into the same group. Examples include toll booths that read license plates, facial recognition, or systems that detect specific objects in satellite imagery.
- Natural language processing that identifies a text’s genre, sentence structures, grammar, people mentioned, etc.
- Services like Amazon’s Alexa, Google Assistant, and Apple’s Siri that provide speech recognition from acoustics and predictive patterns.
In education, perhaps the ultimate example of AI is the idea of creating virtual, human-like teachers that could think, act, react, and interact in a natural way, responding to and using both verbal and nonverbal communication to provide a tailored experience to students that adapts and anticipates their learning needs. While we are still very early in achieving this vision, there are examples of AI being applied to education. Captivating Virtual Instruction for Training (CVIT) is a distributed learning strategy that aims to integrate live classroom methods with best-fit virtual technologies. CVIT integration includes virtual facilitators, augmented reality, and intelligent tutors in remote learning and training programs.
What is not AI in education? Machine grading of multiple-choice questions is not AI. Grading multiple-choice questions is a task in which the rules (correct answers) are manually fed into the machine prior to grading. This is not to say that human intelligence was not involved with identifying the question, appropriately putting the question into a text form, selecting the best answer, and providing appropriate written feedback.
The Future of AI-enabled GEOINT Education
AI is already starting to be deployed in education and is expected to grow in the U.S. by 48% from 2018–2022. As we pointed out earlier, becoming an expert in GEOINT is exceptionally difficult and time-consuming. If AI can significantly shorten the purported 10 years it takes to achieve expertise, the lofty promises of this technology are worth pursuing.
The foremost goal of education in the domain is to create a GEOINT professional with a sense of responsibility to govern the employment of the knowledge. This requires learning that brings intrinsic and instrumental value to the individual. It is about developing the intellect, broadening one’s understanding of the field, and infusing a zeal for discovery. We are specifically stating that academia’s role when employing AI should be broader than supporting the mastery of competencies. It should also improve the individual’s ability to learn and discover in the context of the human-machine environment, which has the ability to create new knowledge.
This invokes the fundamental difference between training and education. Training divorces knowledge, the self, and the surroundings. Here, knowledge is usually defined as skills to be put to use in a context determined by someone under a specified set of conditions other than the person being trained. Using AI in education should also support an opposite outcome that entails knowledge, which helps the individual define the situational context for the application of appropriate skills. AI should offer an arena for free expression and the nurturing of new thoughts. It should offer individuals a chance to discover and develop skills, knowledge, behaviors, and talents. Understanding this context offers possible areas that might be explored.
Intelligent Tutoring Systems. AI has already been applied to education, primarily helping to develop skills and testing in Intelligent Tutoring Systems (ITS). Here, AI techniques can capture subject matter expertise through the detection of patterns. This enables the tutoring system to generate problems on the fly, combine patterns and apply rules to solve the problems, assess each learner’s understanding by comparing the software’s reasoning and results with theirs, and demonstrate the software’s solutions to the participant. An example is “Sherlock.” Sherlock helped to instruct mechanics to troubleshoot avionics test equipment for the F-15 aircraft. The system relies on a knowledge base of concepts and rules developed from subject matter experts. The computer is provided, or infers, information about the learner’s level of knowledge and presents problems that are appropriate to that level, and then adds progressive difficulty, specifically for the purpose of pushing the student’s understanding to a higher level. One study stated that approximately 25 hours of Sherlock training was the equivalent of four years of on-the-job training. According to Hoffman et al., ITS for training the Navy’s Information Technology rating raised the performance of newly trained technicians from a 50th percentile to a 98th percentile level.
AI techniques might be applied to capture how an analyst solves a geospatial problem and represent subject matter expertise in rules. This would both expedite the process of educating analysts and accelerate individual proficiency, including a demonstration of transfer to the operational analytic context.
Personalized learning. GEOINT learning can be tailored to the preferences and interests of the learner, and instruction can be adapted and paced to a student’s unique needs. Different from individualized instruction, personalized learning allows the learner to create activities and relies on the individual’s personal and professional interests. This method is vastly different than most traditional GEOINT training. The intent here is to teach the student to take control and manage his or her own learning when professional goals, curriculum, and content can vary.
Differentiated instruction. This involves the teacher adapting various elements of instruction according to the needs of the learner. Differentiated instruction is facilitated by AI and big data. For example, D2L offers Brightspace Insights, a suite of analytical tools for educators. Brightspace captures, aggregates, and analyzes data streamed from several different sources, including learning apps, online resources, publishers, and other learning management systems to build a complete model of individual student learning behaviors and needs. Student data can be stored and analyzed over time to see what material they engage with more successfully and what educational deficits they may have hidden in their past work that might be inhibiting their future potential. By having a fuller understanding of the learner on day one, educators are better positioned to utilize their skills to address students’ individual needs from the start, rather than spending weeks or months identifying problems they would then have little time to address.
Automate administrative tasks. AI can assist in the administrative interactions among students, advisors, and teachers. Using schedule planners. it can help with selecting the best classes to meet the individual goals of the student. It can develop optimal course sequences to allow the student to progress successfully through an academic program while maximizing student opportunities.
Teachers can also use AI to identify gaps in instruction or concepts that are being missed by students in their courses. Instead of waiting for course evaluations or assessments, the systems can proactively alert teachers to areas that might need remediation. Similarly, AI can also assist teachers to identify gaps in their curriculum compared to similar courses and the broader body of knowledge in a field.
While multiple-choice and true/false tests can be graded automatically, by learning the lesson material and questions students raise on the subject matter, systems can create these types of questions and tailor them to the individual. In addition, through character recognition and natural language processing, AI can aid in the grading of written essays and research papers. This will assist teachers in standardizing these subjective types of assessments and might help teachers give students feedback more quickly.
Tutoring and support outside the classroom. Using AI to support accelerated learning requires consideration of a learner’s retention and the decay of GEOINT concepts and technical knowledge. Simulations provide opportunities for practice to avoid decay of fundamental skills and a means to engage students with immersive, real-world scenarios to develop their problem-solving. Consider the value of an AI-enabled simulation that can adapt to a student’s correct or incorrect answers. Rather than receiving a failing grade or having to repeat the entire exercise, the simulation goes back to the point where a student went off track, highlights key concepts, and enables them to learn from mistakes.
Incorporating AI in online education discussion boards impacts student engagement and their ability to communicate. Packback is an online, AI-supported discussion platform that monitors student engagement in course discussion postings, tracks their curiosity, and provides feedback to improve asking open-ended questions. Instructors enter the AI process by selecting the best student posts as features to reinforce effective learning.
Challenges for AI-enabled GEOINT Instruction. For the foreseeable future, we anticipate that AI tools would be a supplement to human instruction, and not a replacement for human instruction. Therefore, it is important that the human instructor understands the pattern of learner activity and outcomes that the AI instructor has detected, and also understands the learner assessments which the AI instructor derived from those patterns. This is critical for AI to support human assessment of learner performance and progress; however, this is one of the challenges of current AI systems. Often, the patterns derived and the AI’s related assessments are opaque, and as AI continues to learn from an increasing number of engagements with multiple learners, the patterns change and evolve. Increasing instructor insight into AI patterns and their meaning will be important for the performance of the entire educational (human and AI) team.
Success in GEOINT stems from combining the utilitarian aspects of technology with a sophisticated understanding of organizational practices within which people create, use, and employ geographic understandings and knowledge to make decisions. AI has proven its effectiveness in learning in areas other than GEOINT. Education is moving away from the traditional, one-size-fits-all classroom learning that doesn’t actually reflect how people learn and their different backgrounds. AI allows for a tailored and customizable learning experience in which the teachers have access to data on how students are progressing with the concepts they’re trying to learn. This allows content management systems and teachers to adjust their lessons accordingly and to help students where needed. As the role of AI continues to evolve in the field of GEOINT, it makes sense to instill this technology in the education process in an effort to accelerate the learning of the GEOINT analyst.
Recommendations for GEOINT Educators and Organizations
- Begin with faculty development as a catalyst for change. Schools teaching GEOINT can develop online courses to teach established faculty the fundamentals of AI and how it applies to GEOINT, as well as offer recommendations about incorporating AI into college courses.
- Develop AI that monitors relevant educational performance metrics such as examinations and, when numerous students incorrectly answer test questions, alert the instructor of potential misunderstandings of key concepts and simultaneously provide hints or nudges to steer students toward the correct answer. The goal of education is, of course, for students to learn.
- Apply AI teaching assistants to continually search the internet for relevant, open-source examples of geospatial analysis, current imagery analysis, and GEOINT methods that teachers can use for case-based instruction.
Future Research Directions
- Cooperative Human-Machine Learning: Investigate the mutually supportive aspects of tradecraft, learning, and AI in the future tasks of the GEOINT analysts. The goal should be to constantly improve both the individual’s and machine’s ability to learn, discover, synthesize, and assess in the context of the human-machine environment over traditional, manual processes.
- Competency Evaluation: Study natural language processing of courses in a GEOINT program. Identify the similarities and differences of terminology, context, and situational case studies. Consider how the courses reinforce overall key learning objectives and which course or lesson designs could be improved to support continuous learning throughout the program.
- Knowledge Retention: Study AI methods to track the student retention of GEOINT concepts within the progression of a school’s GEOINT program and across USGIF-accredited schools. Apply new insights to curricula to improve the preparation of GEOINT analysts to demonstrate their knowledge and foundational skills at the start of their employment.
- Stephen Noonoo. “How Disruptive Technologies Are Leading the Next Great Education Revolution.” The Journal, 2013. https://thejournal.com/articles/2013/01/14/how-disruptive-technologies-are-leading-the-next-great-education-revolution.aspx.
- Joshua Inwood and Derek Alderman. “The Care and Feeding of Power Structures.” Annals of the American Association of Geographers, 2019. doi:10.1080/24694452.2019.1631747.
- Robert Hoffman, Paul Feltovich, Steve Fiore, Gary Klein, Whit Missildine, and Lia DiBello. “Accelerated proficiency and facilitated retention.” Florida Institute for Human and Machine Cognition, 2010. https://apps.dtic.mil/dtic/tr/fulltext/u2/a536308.pdf.
- Ido Roll and Ruth Wylie. “Evolution and Revolution in Artificial Intelligence in Education,” International Journal of Artificial Intelligence in Education, 2016:26:582. https://doi.org/10.1007/s40593-016-0110-3
- Marvin Minsky (1968) quoted by: Blay Whitby. Reflections on Artificial Intelligence. 1996.
- Technavio Business Report. “Artificial Intelligence Market in the US Education Sector.” Business Wire, 2019.
- Hoffman, 2010.
- D2L is a cloud software company. It is the developer of the Brightspace learning management system, which is a cloud-based software used by schools, higher education, and businesses for online and blended classroom learning.