4.1. Selection of the Target Domain
The military intelligence services comprise five stages of the information circulation, namely, planning, collection, processing, analysis, and dissemination. We analyze the work characteristics of each stage of military intelligence areas to be applied with the AI technology. Before determining the priorities for applying the AI technology, we consider the characteristics of each information circulation stage.
The planning stage establishes a plan to collect information using surveillance and reconnaissance assets. However, due to the low number of surveillance and reconnaissance assets available in the military, the military intelligence officer coordinates their operation plan using MS Excel. When further surveillance and reconnaissance assets, such as military satellites, are acquired in the future, it would be necessary to establish a collection plan that reflects various conditions, such as missions, types of collection systems and sensors, and weather. This is expected to complicate the collection-planning task, and the development of an information system will be required for a smooth execution of the task.
In the collection stage, information is collected using surveillance and reconnaissance assets. To collect information, there are various surveillance and reconnaissance assets, such as imagery, signal, and open source information. When most surveillance and reconnaissance assets were acquired, the format of the collected information was not standardized. Therefore, commercial satellites produce images in the Geo-TiFF format, HUAV in the NITF format, and MUAV in the JPEG2000 format, and it is difficult to analyze and fuse such non-standardized data.
The processing stage is to read and interpret individual information, such as images, signals, and machinery information, collected using surveillance and reconnaissance assets. In the analysis stage, all-source analysis synthesizes and analyzes the analysis reports that were prepared in the previous processing stage. In the processing stage, the collection troops analyze the information that was collected using a single source and load the reports into the information systems, while the higher-level organization, i.e., Joint Chiefs of Staff, analyzes the information collected using multiple sources or all sources in the analysis stage. In the dissemination stage, the analysis reports are provided to users. The users log into the information system and search for the intelligence reports on bulletin boards on the basis of the information type.
After examining the current status of the intelligence work of planning, collection, processing, analysis, and dissemination, we found out that task managers use Excel and military information systems. However, information systems are being used as simple storage and shared spaces. For example, the information analyst manually analyzes the collected imagery information and then creates and uploads the analysis reports in Hangul [
51] or a PowerPoint file, following which he/she obtains the necessary information by searching for keywords in the title of the report on the information bulletin board.
Although the amount of information collected would rapidly increase with the acquisition of various strategic and tactical surveillance and reconnaissance weapons systems in the future, the number of intelligence analysts might decrease because of the reduction in the number of troops. In addition, we must efficiently perform military intelligence areas by applying the AI technology in accordance with the development of the latest information and communication technology.
4.2. Priority Determination
To apply the AI technology to military intelligence areas, the procedure presented in
Figure 2 was implemented, and the priority was determined. In the first stage, five researchers determined the priorities, and in the second stage, 19 military intelligence officers and IT experts reviewed the determination results during three workshops. The researchers undertook research in the fields of defense informatization and computer science. One researcher had more than 30 years of experience, two had more than 20 years, and two people more than 10 years. The military intelligence officers were in charge of the policy department and the analysis departments of the Joint Chiefs of Staff, the collection and the command and control military units, and the system development organization (the Defense Acquisition Program Administration). On average, they had approximately 20 years of experience.
Priorities are shown in the graph when applying the determination elements of AI technology to the detailed stages of the military intelligence areas.
In Step 1, we evaluate the difficulty of the requirements for each domain in the areas of military intelligence. The clarity of the requirements is a criterion for a successful IT project [
52,
53,
54,
55]. Many IT projects undertaken in the military have been negatively affected by frequent changes in their requirements. In particular, the exact definition of the requirements is important when developing information systems with the aid of AI technology. The requirements are related to the characteristics of the particular AI technology. This is because the AI algorithm that can be applied varies depending on the requirements.
In the graph depicting the difficulty of requirements, the X-axis indicates the clarity of the requirements, which increases when moving toward the right, and the Y-axis represents the complexity of the requirements, and the requirements become simpler toward the vertical direction. The clearer and the simpler the problem is, the easier it is to apply the AI technology. The difficulty of the requirements can be assessed, as depicted in
Figure 3, by considering the surveillance assets currently in operation, amount of information collected, and tasks to be analyzed.
During the planning and collection stages, MS Excel was assessed to be sufficient for the operational planning and collection of currently operational surveillance and reconnaissance assets without the application of any AI technology. However, the clarity of the requirements was observed to be low to apply AI technology to the planning and collection of future surveillance and reconnaissance assets because the characteristics of the surveillance and reconnaissance assets and sensors to be acquired in the future are still unknown. According to the assessment results, the planning stage was in the third quadrant of graph; thus, was allocated 0 points. The collection stage was in the second quadrant of the graph; thus, was allocated 1 point.
The requirements for single-source analysis during the processing stage are clearer and simpler than those for all-source analysis during the analysis stage. Further, the requirements during the dissemination stages exhibit high clarity and low complexity. In addition, and were allocated 2 points because the processing and dissemination stage were in the first quadrant of the graph. received 0 points because the analysis stage was in the third quadrant. In terms of difficulty, the requirements of the processing and dissemination stage were assigned a high priority with 2 points because they were clear and simple.
In Step 2, we assessed the readiness of the data, as depicted in
Figure 4, corresponding to each domain of the military intelligence service. First, this required the detection of potential data to be applied to each stage by the AI technology and the determination of whether the format was machine-readable. In the data readiness graph, the X-axis represents the availability of data, with availability increasing towards the right. The Y-axis represents the machine readability of the data, with the level of machine readability increasing in the vertical direction.
The planning and collection stages are affected the most significantly by the acquisition of future surveillance and reconnaissance assets. The AI algorithm to be applied depends on the characteristics of the additional assets and sensors. Therefore, during the planning and collection stages, the availability of the data is currently low, and the format of the collected data cannot be confirmed; therefore, the data readiness was also assessed as being in the third quadrant of the graph, with and allocated 0 points each.
By contrast, during the processing, analysis, and dissemination stages, military intelligence data accumulated over the past 10 years are retained, and the most of the data are converted into images, Hangul files, and PowerPoint files, thereby satisfying the second level criteria specified in
Table 3. In particular, certain data from the processing stage comply with the third level requirements in
Table 3, and can be converted into the XML format for processing. Thus, the processing stages spanned the first and fourth quadrants of the data readiness graph; thus,
was allocated 1.5 points, which was calculated as (2 + 1)/2. The analysis and dissemination stages were assessed to lie within the fourth quadrant; hence,
and
were allocated 1 point each. In terms of data readiness, the processing stage had the highest priority with 2 points because a large amount of data was suitable for processing by AI technology through machine reading.
The confidentiality of the data depends on the type of the information and not on the stage of information circulation. For example, secrecy increased along the following hierarchy: “open-source intelligence < imagery intelligence < signal/human intelligence”. Notably, the level of confidentiality is proportional to the difficulty of sharing and disclosing data. Thus, the confidentiality of the data is an additional consideration.
In Step 3, AI technology level is evaluated.
Table 4 shows the technologies and maturity levels that can be applied to each domain of military intelligence areas. Machine learning and reasoning techniques can be applied to the planning stage, as the optimal plan should be formulated according to the mission, characteristics of surveillance and reconnaissance assets and sensors, weather conditions, among others. The collection stage can be applied via machine learning, recognition, reasoning, and application technologies to select the information collected using the sensors of surveillance and reconnaissance assets that show signs of threat and transfer information to the processing stage.
In the processing stage, machine learning and cognitive intelligence can be applied to information such as images, signals, and open-source data. In the analysis stage, machine learning, inference, and application technology can be used to analyze the threats due to synthesizing the intelligence processed by all the sources or to perform time-series target analyses from the past to present. Finally, the dissemination stage can apply not only the UI/UX visualization technology, but also machine learning and reasoning that analyzes user-usage patterns and recommends relevant information.
The maturity level of the technology applied to the military intelligence tasks was inspired by the study conducted by Oh Jin-tae [
56]. The aforementioned study conducted a questionnaire survey among AI experts regarding maturity levels in technology hype cycles corresponding to 13 AI sub-technologies. The results of the positioning analysis for the AI sub-technologies have been depicted by mapping them to the AI technology classification presented in
Table 1. Machine learning and inference applied during planning stage were marked to correspond to 3.5, which is a level intermediate to the third (bubbles removal) and fourth stages (re-lighting).
The applications employed in the collection and analysis stages were assigned scores of 2, and machine learning and inference were assigned scores of 3.5. Further, machine learning and cognitive intelligence applied during the processing stage were assigned scores of 3.5. During the dissemination stage, it is appropriate to apply visualization technology, and when AI technology is applied, it is given a score of 3.5 through machine learning and inference. The maturity level of the technology is normalized to allow it to be combined with other assessment criteria.
In Step 4, the effect of technology application is predicted. The effect of the application of the core technologies applicable to intelligence areas was predicted, as presented in
Table 5, by referring to the automation value of the technologies used in Project Maven [
50]. The effect of the technologies applicable to military intelligence areas.
In the processing stage, computer-vision and cognitive-intelligence technologies that can be applied to the images, signals, and open-source information were predicted to serve as an automation index of 50 times.
Figure 5 shows the maturity of the core technologies to be applied to each detailed military intelligence area mentioned in
Table 5 and the result of the technology application. In the graph depicting the technology maturity and application effect, the X-axis represents the maturity of the technology, with the maturity level increasing toward the right. The Y-axis represents the effect of applying the technique, with the effect increasing in the vertical direction.
In terms of the maturity of AI technology, the inference and application technologies applied during the planning and analysis stages were located on the left side of the graph because of their low maturity, and machine learning and cognitive intelligence applied during the processing and collection stages were located on the right side of the graph because of their high maturity levels.
The application effects of the AI technologies were predicted to be similar in most stages, but it was predicted that applying computer vision/AI to image information during the processing stage would result in 50-fold improvement and would be assessed as the highest. In terms of the maturity of the core technologies and the effects of applying the technologies, the processing stage was accorded a high priority and allocated 2 points.
4.3. Results of Priority Determination
Table 6 depicts the results of priority determination obtained by applying the AI technology during the planning, collection, processing, analysis, and dissemination stages, which are the constituent domains in military intelligence.
As a result of applying the priority calculation formula, the processing stage was found to have the highest priority at 6.9 points as shown in Equation (2). In terms of the difficulty of the requirements, the requirements in the processing stage were simpler and clearer because of the execution of a source-specific analysis based on currently collected information. The processing stage was located in the first quadrant of the requirement difficulty graph depicted in
Figure 3, with
allocated 2 points.
In the case of data readiness, the AI technology could be applied immediately as the information collected in the past was readily available (availability). In addition, the information stored in the database corresponded to the second and third stages, and it was machine readable (readability). The processing stage was located across the first and fourth quadrants of the data readiness graph presented in
Figure 4, and
was allocated 1.5 points as the median score.
Furthermore, machine learning and cognitive functions could be applied to information processing, and the maturity of this technology was ascertained to be between the third (bubble-removal) and the fourth stages (re-lighting). The effect of applying cognitive functions was observed to be the most pronounced in terms of technology-application effects. The processing stage was awarded 1.4 points at the technological maturity (
level, and 2 points were allocated, in the first quadrant of the technical application effects (
graph in
Figure 5. The priority of applying AI technology to the military intelligence area was processing > dissemination > collection > analysis > planning.
The above example is used by policy departments for decision-making purposes. When system development departments assess the priorities by placing higher weights on the development elements, the priorities may vary, as indicated in
Table 7.
If three times the weight (
= 3) is assigned to the data readiness level and twice the weight (
) is allocated to the maturity level of the technology, the results presented in
Table 7 can be obtained. From the perspective of the system development department, the priority can be calculated in the order of processing > dissection > analysis > collection > planning.