2018-09-15

How to Apply Artificial Intelligence in Military Affairs – A Business Analysis Example at Tactical level

The article explains the features of Artificial Intelligence (AI) feasible to military, analyses a process of tactical Command and Control, and considers how Artificial Intelligence may improve tactical-level command and control. 

The approach for this analysis is an evolutionary, i.e. gradual improvement and not revolutionary, i.e. creating a new way of military affairs. The improvement analysis process runs as follows:

  1. Define the military system of systems, environment and areas of Artificial Intelligence under the consideration
  2. Recognise the primary process to be enhanced and its connection to measurable military capabilities
  3. Recognise the questions or problem types where AI may make a difference and link them to a chosen process for improvement
  4. Retrace the AI features back to capabilities and create a link between AI features and expected measurable benefits
  5. Consider the constraints of existing data, systems to embed the AI and cultural readiness to adopt AI featured solutions.



1. Target Definition

The section below defines the military system of systems, its environment and focuses on the area where Artificial Intelligence effects are studied.

1.1 Area of focus in a military system of systems

Military affairs are a part of confrontation and conflict between societies. The culture, religions and technologies the societies are applying also reflect military structures and affairs. Recently, western societies have been transformed by opportunities created by digitalisation and social media. Now, there are concepts like smart governance, social business, industry 4.0 and artificial intelligence providing new opportunities to develop functions.  This study chooses to analyse a military enterprise facing new opportunities created by the Artificial Intelligence. Furthermore, the research focuses on the capabilities of military force utilisation at the tactical level as illustrated in Figure 1.

Figure 1: Contemporary context for military affairs and system of systems

At the tactical level, the military (BLUE) is focusing on gaining a victory over the adversary (RED) in combat. BLUE intends to create a combined effect through all three areas of effect in the RED’s socio-technical system: Moral, Mental and Physical as presented in Figure 2.  In the physical area, the ways of tactics are manoeuvring the troops to create an advantage, massing the effect to produce enough casualties, having the firepower to overcome the means of defence, and applying everything in tempo that the RED cannot recover at the mental level. In the mental area, the ways are a deception to mislead RED’s expectations, surprise to catch RED unprepared, confusion to slow down RED’s sensemaking, and shock to scatter RED’s moral cohesion. In the moral area, the cohesion of RED’s moral is the key to victory. When the moral cohesion starts rumbling down, and there is no time to rebuild it, the BLUE success is granted. 

Figure 2: Tactical essentials in Blue and Red confrontation

Since the success at tactical level combat is significantly dependent on the human decision making, this study tries to find an answer to how artificial intelligence may improve the human command and control process to gain a better advantage in the battlefield.


1.2 Areas of Artificial Intelligence to consider for an improvement

The study focuses on recent developments in AI related technologies and data science. The significantly improved features illustrated in Figure 3 are: 

  1. Automation leading to autonomous systems, robotics and automated processes
  2. Human Machine Interface is connecting Human with machines through Natural Language Processing features more intuitional way than the legacy keyboard and joystick.
  3. Analytics in making sense of large and complex sets of structured or unstructured data
  4. Image recognition in detecting and identifying patterns and meaningful “images” from pictures, videos, radar, sonar, SIGINT, handwriting, etc.


Figure 3: The four recently developed areas in AI related features

The study focuses on AI analytics application in military command and control at the tactical level, especially in the land battle scenario.


2. Essential Tactical Capabilities and Functions Inclined to Enhancement by Features of Artificial Intelligence

The section below recognises the primary decision-making process to be enhanced using artificial intelligence and links the process to tactical capabilities. The OODA -model by John Boyd is a most straightforward description of the human decision-making process in a tactical situation. The functions of Observe, Orient, Decide and Act are linked into essential tactical functions chosen from the model introduced by B.A. Friedman.


2.1 Human process for commanding and controlling

The OODA -model in this study is simplified to a loop rather than the original meta-paradigm and systems model. The simplification is done to keep this study focused on the four functions of the OODA loop  illustrated in Figure 4 are:

1. Observation: By observing and considering new information about our changing environment, our minds become an open system rather than a closed one, and we can gain the knowledge and understanding that’s crucial in forming new mental models.

2. Orientation: “orientation shapes the way we interact with the environment…it shapes the way we observe, the way we decide, the way we act. In this sense, orientation shapes the character of present OODA loops, while the present loop shapes the character of future orientation.”  There is a twofold process included here:
  • Destructive deduction process analyses and pulls apart existing mental concepts into discrete parts. This is to define the constitutive elements.
  • Creative induction uses these constitutive elements to form new mental concepts that more closely align with what we have observed is really happening around us.

3. Decision happens when actors decide among action alternatives generated in the Orientation phase. The decision is essentially moving forward with a best possible hypothesis — best “educated guess” — about which mental model will work.

4.The action includes multiple parallel actions/tests/experiments going on at the same time so that you can quickly discover the best mental model for a situation. In battle, this might mean having multiple attack points that are using different weapons systems. When the commander determines which targets and weapons are providing the best results, he’ll direct his attention to the winning mental model and mass it to the max until it no longer works. Once the commander observes that it is no longer effective, he’ll orient more mental concepts, decide to use one or several of them, and quickly act to test them out until the enemy reacts in an intended manner.

Figure 4: Essential OODA-loop modelling the command and control process of a battle commander

The above-described loop is modelling the human behaviour of a battle tank commander. The study uses the model to present how the command and control functions link into essential tactical functions to create an effect on the adversary. Next section will illuminate these tactical essentials and study their causality in reference to OODA functions.


2.2 Essential tactical functions used in this study

From the list of three tactical areas of effect and together nine ways applied in battle tactics , this study chooses the following to keep the model simple:

  • Deception in combat is the manipulation of the enemy’s understanding of the situation to achieve an advantageous position. The tactician misleads his enemy by false image and hides his real intention to create surprise and shock to his enemy. 
  • Manoeuvre means attacking an enemy force from a position of comparative advantage. Manoeuvre is any asymmetry that provides the spatial or functional advantage that the enemy is not adept at countering. Forms of manoeuvre are for example frontal attack, flanking attack, envelopment, turning movement, infiltration, and swarming.
  • Mass is an advantageous concentration of combat power or effects in space and time. Massing is to achieve local and timely superiority using the combined arms effect.
  • Firepower is the ability to detect, engage and take down the target. Firepower can be mitigated by dispersal, cover, concealment, and armour. In the modern battlefield, the combined arms firepower is essential to overcome the advantages and disadvantages of the enemy’s weapons.
  • Tempo is the ability to control the pace of combat to your advantage and the disadvantage of the enemy. Both sides are affected by the friction of war, and this entropy of troops can be magnified by deception, attrition and the ability to decide faster. Principally, both the quickening and slowing down can be used for tactical advantage.

2.3 Mapping the command and control to tactical combat essentials

Mapping the human functions of the OODA loop to essential tactical functions will provide the points of effect when considering enhancing human performance with AI features. The mapping is done using the question method by Marr  to reveal the need of understanding in combat command and control. The next table 1 provides samples of crucial questions where AI may give a better understanding and a difference in human cognitive performance in a tactical situation.
Table 1: Assessing the points of effect when OODA loop is applied in tactical battle



Observe
Orient
Decide
Act
Deception
What is happening outside of the focus area?
What is RED’s normal doctrinal behaviour?
What may RED do in scenario A?
Fastest reserve available
Manoeuvre
What is moving over time?
Estimated time to approach a point
Which routes give the best approach?
Are BLUE units on time?
Massing of effect
Coordination of time of arrival
Which are the points of effect on RED’s system?
Optimising the effect and firepower
Is the massing of effect getting to RED’s breaking point?
Firepower
Detecting targets
Identifying and prioritising targets
Identifying and taking down targets
Is BLUE taking down targets fast enough?
Tempo
How is the BLUE doing?
How is the RED doing?
What is the average pace of BLUE and RED in this terrain/ situation?
Optimising the time of decision
Changes in the tempo of both sides
Table 1 will be revisited when the AI features linked to OODA functions are retraced to essential tactical capabilities.


3.Problem Types Where Artificial Intelligence Makes Difference

In quest of not diving too deep into AI technologies, this paper uses McKinsey defined problem types  where AI has been observed to make a difference in real-world cases. When robotics and machine vision oriented solutions are deducted, the following seven types of problem types can be supported by features of AI:

1. Searching

  1. Searching for content, routes, answers, or combinations. Searching is an AI-enhanced function that continually evolves in identifying, sorting, and presenting the data that is most likely to meet the needs of users at that specific time, based on a multitude of variables. 
  2. Answering questions like:


  • What does “this” mean? Google RankBrain embeds written language into vectors that can indicate the closeness of phrases. Thus Google search gives also related content not only keyword matches. 
  • Navigational Queries: Who is this agent? Which domain does this content come? Where is this location? How do I get to this location?
  • Informational Queries: What happened in this location at this time? What is ongoing at this location now? What is planned to occur in this location?
  • Intentional Queries: Who is selling/buying this item? Who is heading to this location? Who is planning to deploy to this area?
  • Transactional Queries: Pay a bill! Purchase an item! Order supply! Publish the report!


2. Classification and anomaly detection

  1. Classify new inputs as belonging to one of a set of categories (trained or determined). Based on history, determine whether new inputs are within normal or anomalies.
  2. Answering questions like:


  • Does the received image contain a specific type of object?
  • Are we driving along the lane?
  • Is the traffic light red, orange or green?
  • Which RED transceiver does this signal belong?
  • Does this event fall into the typical area of behaviour or is it an anomaly?
  • Is this new information or update to an old piece of info?

3. Continuous estimation

  1. Predict the time series. Estimate the next numeric value in a sequence.
  2. Answering questions like:


  • How long can the RED continue firing the current way if we know their organic supply? 
  • How long does the fuel will last if BLUE continues driving the current way?
  • When the BLUE needs its resupply if they continue fighting this way?
  • How many sorties can this BLUE fighter fly before the engine needs to be changed?

4. Clustering

  1. The system creates a set of categories, for which individual data instances have a set of standard or similar characteristics.
  2. Answering questions like:


  • What do 60-65 year old people buy for Eid celebration?
  • Do these RED units belong to the assumed official organisation?
  • Do these tracks belong to which RED fighting vehicle?
  • Is the recorded event normal or abnormal?
  • Is the transmitter signal captured new or already appeared in AOO?
  • Would the RED commander prefer A or B option based on anthropological, psychological and doctrinal data?

5. Optimisation

  1. The system generates a set of outputs that optimise outcomes for a specific objective function.
  2. Answering questions like:


  • What is the optimal route from A to B if we need to reach there as soon as possible sustaining 2/3 fuel reserve?
  • If we have only one Recce patrol, which route would be safest but quickest to get them to point B within the next 2 hrs?
  • If RED wants the optimised effect on our troops with their artillery, which areas they may deploy their batteries?

6. Ranking

  1. Results of a query or request need to be ordered by some criterion.
  2. Answering questions like:


  • Which RED manoeuvre is the most probable in this situation?
  • In which order should BLUE use its reserves to create 10% losses to RED troops?
  • What priority should BLUE use fires to cause over 30% losses when facing RED troops?

7. Recommendation

  1. systems are suggesting next step sorting options by relevance, probability, feasibility or availability before presenting the results to the user.
  2. Answering questions like:


  • What may this customer buy next, based on the buying patterns of similar individuals?
  • If RED is manoeuvring his armoured task force this way, what are his next options to engage BLUE troops?
  • If the RED commander is walking like this, how well he can sustain in 24/7 intensive operations?
  • If RED positions their artillery here, how long they can sustain fire to this far?
  • How trusted this source of information is based on past deliverables?

The above list defines the taxonomy for generic problem types that AI is feasible in solving. The slight military flavoured examples help architects to apply the taxonomy in military command and control process improvement in following sections.
Furthermore, when the design is conducted, there is a need to map the military problems into possible AI algorithms plausible for types of general problems. The following Table 2 provides one view for problem types and likely AI sampling techniques.

Table 2: Generic problem types and plausible AI sampling techniques 
Generic problem type
AI sample techniques
Searching
Brute force search, Breath-First search, Depth-First search, Bidirectional search, Iterative Deepening Depth-First search, heuristic searches, Local searches [1]
Classification
CNN, Logistic regression
Anomaly detection
One-class support vector machines, k-nearest neighbours, neural networks
Continuous Estimation
Feed forwards neural networks,  linear regression
Clustering
K-means, affinity propagation
Optimisation
Generic algorithms
Ranking
Ranking support vector machines, neural networks
Recommender
Collaborative filtering




4. Finding AI Features to Improve OODA

The taxonomy of generic problem types helps to map the previously analysed tactically essential OODA questions from Table 1. The questions about improving OODA functions can be linked to AI problem types as presented in Table 3. This linkage provides a view to applying AI algorithms into the command and control process.

Table 3: Mapping AI features to improve OODA loop

Observe
Orient
Decide
Act
Searching
Detecting targets
What is RED’s normal doctrinal behaviour?
What is the average pace of BLUE and RED in this terrain/ situation?

Fastest reserve available
Classification, Anomaly detection
What is happening outside of the focus area?
Identifying and prioritising targets
Identifying and taking down targets.
Changes in the tempo of both sides
Continuous estimation
How is the BLUE doing?
How is the RED doing?
Estimated time to approach a point

Are BLUE units on time?
Is the massing of effect getting to RED’s breaking point?
Is BLUE taking down targets fast enough?
Clustering
What is moving over time?


Changes in the tempo of both sides
Optimisation
Coordination of time of arrival

Optimising the time of decision.
Which routes give the best approach?

Ranking

Identifying and prioritising targets


Recommendation

Which are the points of effect on RED’s system?
What may RED do in scenario A?


The mapping in Table 3 indicates that the AI features may improve the human performance in OODA loop. Since the mapping of AI technical features and functions of command and control process, Table 3 is an essential artefact to keep alignment between business and technical architects.


5. Retracing AI Enhances in OODA Back to Tactical Capabilities

The previous sections explain how to redact the military capabilities to a level where AI features can be introduced. Now follows the reasoning that deducts the possible technical enhancements back to higher tactical level capabilities. Taking the mapping done in Table 3, the architect can retrace the potential impacts of AI back to Table 1 essential capabilities of a tactical battle. Table 4 presents possible AI features that may affect tactical capabilities. The deduction helps the architect to reason the potential enhancements to owners of operational capabilities and possible investment decisions while maintaining essential metrics for implementation planning and user acceptance tests.

Table 4: Example of AI enhanced features and benefits in tactical battle

Observe
Orient
Decide
Act
Deception
What is happening outside of the focus area?
Classification/
Anomaly Detection
What is RED’s normal doctrinal behaviour?
Searching
What may RED do in scenario A?
Recommendation
Fastest reserve available.
Searching
Manoeuvre
What is moving over time?
Clustering
Estimated time to approach a point.
Continuous Estimation
Which routes give the best approach?
Optimisation
Are BLUE units on time?
Continuous Estimation
Massing of effect
Coordination of time of arrival
Optimisation
Which are the points of effect on RED’s system?
Searching,
Recommendation
Optimising the effect and firepower.
Optimisation
Is the massing of effect getting to RED’s breaking point?
Continuous Estimation
Firepower
Detecting targets
Searching
Identifying and prioritising targets
Classification,
Ranking
Identifying and taking down targets.
Classification,
Ranking
Is BLUE taking down targets fast enough?
Continuous Estimation
Tempo
How is the BLUE doing?
How is the RED doing?
Continuous Estimation
What is the average pace of BLUE and RED in this terrain/ situation?
Searching
Optimising the time of decision.
Optimisation
Changes in the tempo of both sides.
Clustering

The AI possible problem types are expressed in italics and mapped as probable causality for enhancement of tactical capabilities in Table 4. Moreover, there is a need to find the data that AI algorithms need for training, the information systems that AI features can be embedded, and the information security to ensure the integrity and availability of the essential data.


6.Constraints and Issues Related to Data and Information Systems

Artificial intelligence requires data, either for searching and making sense of things or learning from events. Data may be structured, time series, images, video, text or audio as long it is unbiased and exact digital description of an event or a piece of information.  Before assessing the feasibility of AI features, the architect needs to study:

  1. where data exist, 
  2. how data are accessible, 
  3. what kind of information systems live that may able to embed AI features, and 
  4. would the information security guarantee the availability and integrity of the required data? 

In seeking possible sources of data, the architect needs first to survey all related information systems, find enterprise-level records and databases, consider acquiring data from open sources, vendors, government agencies or coalition partners as presented in Figure 5.

Figure 5: Possible environments where data can be found for AI enhanced military affairs

Secondly, the architect needs to study the existing and developing information systems that are supporting the OODA-loop for potential primary systems where AI features may be embedded. Table 5 presents an example of a moderately digitised force, and it's possible information sources and systems that may be used to embed the features of AI.

Table 5: Example of possible sources of data and systems available to embed AI features through the OODA -loop

Observe
Orient
Decide
Act
Structured data
Tactical and operational sensors, Coalition, Open Source (OS), ISTAR systems
Battle Management Systems (BMS), Enterprise Resource Planning (ERP), coalition and government libraries, vendor service
BMS, ERP
BMS, ERP
Time series
Moving target indicators, event data, tracking data
Battle log’s, tracking data, Surveillance data
BMS, ERP, Combat ID
BMS, ERP, Combat ID
Images
GEO, SIGN, OS, Unmanned Vehicle Systems (UVS), Image Intelligence (IMGINT)
OS, threat libraries, vendors

UVS, IMGINT
Video
CCTV, UVS
Video libraries, Coalition, Vendors


Text
Military messaging, transactions, documents
Policies, Doctrines, Field Manuals, Standard Operational Procedures, Studies, Documents, OS, Coalition R&D
Military messaging, BMS, ERP, Emails, Documents

Audio
Communications, Acoustic sensors
SIGNINT libraries

Comms channels

Thirdly, the architect should assess the accessibility of data since military enterprises have a habit of stovepipe structures that are isolated from each other. Notably, the connectivity through the whole OODA process creates end-to-end access to data essential for the full loop. Therefore, the interoperability of Intelligence, Surveillance and Reconnaissance (ISR); Battle Management (BMS); and Enterprise Resource Planning (ERP) processes extended to significant partners provide an excellent platform for AI enhanced Command and Control (C2).

Fourthly, the architect should ensure that information security controls and measures are in place to provide data with the integrity and availability required for trusted AI features. One of the most cost-effective ways for the adversary to suppress the AI enhancement capabilities is to attack the trust between soldiers and their AI systems. The trust is lost if RED can manipulate the data content or cut the flow of data stream in the intensive situation.


6. Roadmap to AI Enhanced Tactical Command and Control – a Case Example

One can create a roadmap in Figure 6 for applying AI to enhance tactical command and control processes at the tactical level by composing the elements from the previous analysis. Tactical capabilities from Table 1 define the performance end states for the journey. Each resource year (Anno) establishes the pace of advance. Existing data, system that can embed the AI and vendors ability to do the integration determine the constraints/enablers for realistic milestones. Tactical performance establishes the priority for the possible OODA loop enhancements.

Figure 6: An example of roadmap to improve tactical combat capabilities through enhancing command and control by Artificial Intelligence


7.Summary of the Architecture Logic Applied in this Study

The study explains how an enterprise architect can analyse military affairs and assess the feasibility of using Artificial Intelligence features to improve capabilities in military force utilisation. The case study is focusing on the tactical level battle and enhancing commander’s command and control abilities using AI features in chosen tactical functions. The example provided illustrates the following analysis process of an architect for AI feasibility study:

  1. Define the system of systems and its environment to create a holistic view
  2. Defining the measurable areas of capability that require improvement in the context of the previous comprehensive systems view
  3. Defining the processes where Artificial Intelligence features may make an impact at capability level = first mapping of improved processes and areas of capabilities
  4. Defining specific functions in target process where AI can make a difference = second mapping of functions and AI specific enhancements
  5. Retracing the AI enhancements through processes back to capabilities
  6. Considering the enablers or constraints of data, information systems, connectivity and information security
  7. Creating a roadmap for AI enhancements

The study focused on analyses and expert features of artificial intelligence only. Therefore, there needs to be a similar study of the highlights of Human-Machine Interfaces, Autonomous systems and Image recognition to create a more holistic understanding of the opportunities provided by recent AI developments in military.

No comments:

Post a Comment