|Automation in the Intelligence Cycle|
|System Analysis and Studies|
Artificial Intelligence AI, Automation, Intelligence Cycle, Machine Learning ML, Military Intelligence
As mentioned by the retired U.S. intelligence officer David Shedd, former Deputy Director of the Defense Intelligence Agency (DIA), ”Every successful military plan and operation relies on intelligence ” [ 2016 Heritage Index of U.S. Military Strength, https://s3.amazonaws.com/ims-2016/PDF/2016_Index_of_US_Military_Strength_FULL.pdf ]. In recent years, the intelligence process is growingly challenged. The increasing complexity and dynamics of the context of military operations, as well as the unpredictability of the enemy, create a high demand for the effectiveness and the efficiency of the intelligence process to deliver actionable products.
The technological developments of the last years have a major influence on the intelligence process, too. Sensors are becoming smaller, lighter, faster, less costly and more effective, and are therefore being used in increasing numbers. Platforms can operate for longer periods of time, so that areas can be monitored almost 24/7 (persistent surveillance). This all results in a changing pace of military operations, increased capacity and ease to collect vast amounts of data, increased information flows, and demanding systems’ performances and capacities.
The growing availability of data on the one side, and the growing demands for actionable products on the other side, are a huge challenge for the current intelligence process. Currently, a large part of the intelligence process is in fact still conducted manually, e.g. imagery requires skilled expertise to view and interpret the intelligence material. In addition, the collected data and information is highly complex, being prone to ambiguity, subject to uncertainty, and susceptible to countermeasures which might blur its information content. The human workload is challenged, both from the point of view of the quantity of data, as from the point of view of its complexity.
One of the grand challenges for the NATO Essential Operational Capability (EOC) 2: Effective Intelligence by IST-173, identified “opportunities to accelerate the Intel cycle by application of autonomy” as one of the grand challenges. The high-level objective of this activity is to identify opportunities to improve the Intelligence cycle -either by accelerating it, or by increasing other quality aspects to be identified- by application of AI-enabled systems and automation.
Assumptions - For all the listed objectives, we rely on a number of assumptions. Firstly, to simplify the problem of interconnection between the Intelligence Cycle, the Joint Intelligence Surveillance and Recognition (JISR) Cycle, the Operations Cycle, and the Targeting Cycle, we will look at the Intelligence cycle assuming to have a set of collection requirements already made available, either from an intelligence gap analysis or from the Operations Cycle. Additionally, we will focus on supporting the PED (Process, Exploit, Disseminate) capability of JISR Cycle and the Processing capability of the Intelligence Cycle. Secondly, we will look at opportunities to improve the (designated part of the) cycle in all its aspects, not only in speed. Lastly, we assume that the ultimate goal is not to have autonomous/automated systems replace the human intelligence operator, but to act in a man-machine teaming effort.
Summary of objectives. Firstly, we will address the question whether automation can support the intelligence cycle, at which phases / (sub)functions / activities of the cycle it could play a role, and in which form (which combination of technologies is deemed the most promising for each phase and forms an opportunity for improvement). Secondly, we will choose a subset of identified opportunities and qualitatively (and -where possible- quantitatively) assess the benefit of automation in specific cases within national and NATO experiments/exercises.
Technology scope - Note that in the current document and proposal of work, we look at “autonomy” and automation in a broad sense. AI is one of the key enablers of autonomy, but not the only possible enabler. Therefore, our technological scope encompasses :
• general function approximation techniques for machine learning, including neural networks, (semi-)supervised, unsupervised, reinforcement, ensemble and instance-based learning;
• techniques for machine reasoning, such as reasoning from symbolic knowledge representations- e.g., knowledge (hyper)graphs, (fuzzy)logic reasoning and algorithms for swarm intelligence;
• man-machine interaction technologies;
• data science and enabling information processing techniques, often based on a combination of mathematical modelling, statistics, computer science and specific domain knowledge.
Examples of recent domains where a combination of these techniques have been successful are: autonomous driving and planning, gaming, image recognition/classification, natural language processing and text mining.
1. Objective O1 – “Mapping”: conceptual mapping of AI technologies versus (sub)functionalities of the intelligence cycle, first (conceptual) assessment of the benefit they could bring and proposal of a methodology for assessment of the hypothesized benefit.
2. Objective O2 – Field experiments: assessment of the (hypothesized) benefit of autonomy for the intelligence cycle through experiments with intelligence professionals.
3. Extra Objective O3 (and possible follow-on to this TAP) – Future “could-be” intelligence process: design and experiment novel concept(s) of conducting intelligence and experiment with the novel concepts.
4. Overall deliverable: D4 report collecting results of all objectives (D1.1, D2.1, D2.2, D3.1), overall conclusions and future recommendations from all objectives and any other relevant targets of opportunities identified.
As already listed in the section “Technology scope”, our scientific topics encompass:
• Technological domain: (application of) artificial intelligence, intelligence systems, machine learning; machine reasoning, data science, , information fusion techniques, man-machine interaction methods; explainability of AI; presentation and visualization of information. Note that we will not cover all these
topics in detail, but they may all form part of our conceptual mapping;
• Application domain: Intelligence, JISR.
Most of our activities are NATO unclassified; however, we foresee the possibility to share material with higher classification.