The National Security Commission on Artificial Intelligence (NSCAI) released its final report on March 1st. Its findings and recommendations serve as both a window to the inspirational potential of AI and a warning of the great perils that accompany that potential. For Cougaar Software, Inc., many aspects of the report provide validation of CSI’s technological approach and vision, as well as the theoretical underpinnings of CSI’s work.
NSCAI’s final report is strategic in scope but comprehensive and detailed in its assessments and recommendations. Perhaps its most meaningful contribution is highlighting the fact that our nation is in an AI competition that it cannot afford to lose. The national security implications of this competition are wide-ranging and potentially severe. Thus, the rapid and all-encompassing expansion of AI cannot be left to chance. The United States requires a national strategy and national regime that addresses the full range of areas influenced by AI and applies a whole-of-government approach to providing direction and allocating resources.
Paraphrasing Thomas Edison’s comments about electricity at the dawn of the electrical age, the report calls AI a scientific “general purpose technology of unlimited possibilities,” a “field of fields” and asserts that “how this ‘field of fields’ is used—for good and for ill—will reorganize the world.” Like electricity before it, this reorganization will pervade every aspect of our lives and affect every aspect of our national security.
Given the majority of Cougaar Software, Inc.’s work has been in the area of military planning and execution management (Mission Command), this post will focus on the report’s coverage of AI’s impact on defending our nation. As stated above, the report’s assessments and recommendations provide validation of CSI’s decades-long efforts to further the science of cognitive software agents first explored through DARPA’s Cognitive Agent Architecture (COUGAAR) technology. The report also adds the weight of the NSCAI’s authority to CSI’s assertion that the ability of humans and AI to collaborate over planning and coordinate throughout execution is likely the most critical component of human-machine teaming.
The NSCAI report highlights four areas likely to be at the forefront of AI research in the next decade: hybrid approaches to AI, human-machine teaming, enhanced training methods, and explainable AI. Hybrid AI and human-AI teaming have been CSI’s focus for nearly two decades. While the explosion of narrow AI capabilities has changed the world, almost overnight, moving up the AI value chain to support more complex applications demands hybrid approaches, leveraging more than one AI technique or computational method. Human-AI teaming, the synergy of humans and AI/Autonomy for complex problem solving, offers the potential to truly leverage the best of AI cognitive emulation and the best of human cognition while addressing current concerns related to authorities, ethical use, and bias.
Machine learning has contributed greatly to complex pattern solutions and will increase in value as organizations improve data management, but machine learning results are as narrow as they are deep. By combining ML and other capabilities, hybrid AI techniques like those resident in CSI’s decision support applications allow complex problems to be solved effectively.
The cognitive agent approach employed by CSI is inherently hybrid, arming individual agents with whatever AI or traditional computational capability suits their needs. At the heart of CSI’s developmental approach is a “Divide and Conquer” strategy, which decomposes large, complex problems into smaller subproblems. By using the best technique on each subproblem, maybe an ML classifier or maybe another computational technique, you get the best solution for that class of problem. CSI has employed a wide range of techniques, including three classes of ML, AI Planning, Evolutionary Computing/Genetic Algorithms, A* routing, Case-Based Reasoning, Rules, Fuzzy Logic, and Game Theory. The challenge with the “Divide and Conquer” approach is managing the complexity. CSI has addressed that problem through the evolution of the ActiveEdge agent-based architecture, which manages the organization, messaging, control, and stability of the system. The combination of humans, ActiveEdge architecture, and functional agents utilizing appropriate reasoning techniques achieves the combined and symbiotic advantages of each component within a single well-conceived complementary design.
The report’s assessments and recommendations related to human-AI teaming are extensive and cannot begin to be addressed in a single blog post. At their essence, they highlight the need for humans and AI to collaborate seamlessly, with AI improving both the performance of narrow tasks within a complex system and simultaneously improving the performance of the overall complex system itself. The AI must be just as capable a teammate as the human teammates, being like the staff assistants who are so in synch with the commander that they finish each other’s sentences or capture meaning from each other’s non-verbal cues.
NSCAI members stop short of describing details of mission command but highlight the importance of human-AI teaming in support of future warfighting concepts. It is an easy and short step between the NSCAI’s recommendations and the warfighting concepts taking shape in official Multi-Domain Operations (MDO) and Joint All Domain Command and Control (JADC2) circles. By achieving a “collaborative intelligence” or common situational understanding among human and AI teams, which the report describes as a “cohort of agents,” speed and accuracy of decisions may be improved dramatically. These improvements can enable decisive advantage from the tactical to strategic levels.
While not employing the term Observe, Orient, Decide, and Act (OODA) verbatim, the report describes how AI will augment the decision-making and execution process at each of those steps. Through the enhanced performance of these cognitive processes, AI offers what may be its greatest potential. The report’s explanation is consistent with our Three Layer OODA Loop concept, years in the making, and resident in several of our prototypes.
The report calls for “new concepts in teaming” to better achieve data fusion and for all intelligence products to be both human-readable and machine-readable. To achieve the level of seamless teamwork required, there cannot be a gap between what humans and their AI assistants know and understand. Beyond just being able to rapidly read the same intelligence products, humans and their AI must be able to collaborate over planning and manage the execution of complex operations. Here at CSI, we have been working on a digital planning and execution service that does exactly that. The product of human-collaborative planning in a CSI system is the Digital Living Plan, which allows both humans and systems to understand the plan in context, and provides a rich foundation for analysis, assessment, and execution management.
We are thrilled that a commission of unrivaled expertise and experience would come to technical and theoretical conclusions that closely relate to those underlying our longstanding vision and effort. We feel strongly that CSI’s somewhat unconventional path is in many ways validated by the report. We are excited to continue progress along this path and hope this report helps others see the greater picture and greater opportunity resident in the distributed multi-agent systems approach to cognitive emulation, hybrid AI and human-AI teaming.