Artificial Intelligence from a Military Viewpoint

Artificial Intelligence

Artificial Intelligence (AI) can be defined as follows: AI is a collection of technologies that allow machines, programs or systems to sense, comprehend, act and learn almost like human beings.  

The AI has currently enabled for example the following changes in societies:

• Warehouse robots and people are working aside in retrieving and storing goods. This has improved warehousing industry.
• Intelligent automation is supporting maintenance engineers in remote locations and replacing banking officials, insurance analysts, financial consultants, and health measuring personnel in routine functions.
• The program can learn the rules of a complex board game and beat the best human and preprogrammed machines in the game.
• Robots serve hotel guests and hospital patients with meals and services they ordered.
• Risk modelling service learns thousands of incidents continuously and can predict and assess the probabilities of risks than a human being. 
• Intelligent manufacturing controller can cut the factory downtime almost to zero by collecting information from all the factory devices and taking care of their parts replacement before they break.
• With driverless and connected cars, the people travelling become targets of online shopping, entertainment or remote working.
• Use of infrastructure services, business transactions, and service transactions becomes more monitored, and governments can change for example fixed road tolls and taxation to a more pay-per-usage model which enables better guidance of the behaviour of society.

The AI is one enabler of the current revolution in economics, industry, and societies called version 4.0 which is also called the era of cyber-physical systems and Internet of Things. The most evident progress has been made recently in the areas of computer vision and audio processing; natural language processing and knowledge representation; and machine learning and expert systems. The progress is based on improvements in Deep Learning, Big Data and computing power provided by clouds.

The areas of future improvement with AI

Especially the improvement in natural language processing, computer vision, pattern recognition and reasoning and optimisation have been accelerated by machine learning and introduced some signs of machine intelligence. Currently, we are witnessing the next leap expressed in Table 1:

Table 1: Development of AI capabilities 

Foundations of AI and possible vulnerabilities

Deep Learning methods have been one accelerator for Artificial Intelligence. Deep learning refers to the ability of Artificial Neural Networks (ANN) to use more than one hidden layer to process complex data sets, which improve image and speech recognition and natural language processing. Deep learning is a branch of machine learning based on a set of algorithms that learn to represent the data.  A simple Convolutional Neural Network consists of multiple layers that each present different perception as explained in following image categorisation and Figure 2:

1. Visible layer provides pixels that the sensor could detect
2. First hidden layer can detect edges based on brightness differences between neighbouring pixels
3. Second hidden layer detects corners and contours from the edges of the previous layer
4. Third hidden layer can put together patterns of edges, corners, and contours and create objects and parts of a whole
5. The output layer can provide features of objects in the picture to be able to differentiate them in groups as a car, person or animal (Goodfellow, Bengio, Courville: 2016; Pp.6).

Figure 2: A simple Artificial Neural Network that detects objects from pictures © Zeiler and Fergus (2014)

Deep learning ANN requires large, high-quality datasets for training.  Either these sets of data will have rules of the knowable situation and AI adjusts to small variations, or AI is self-determining the rules of engagement from a substantial number of events. With low-quality data, machines learn unintended behavioural patterns like Microsoft ‘Tay’ robot who was closed after it was exposed to public interaction and crowd made it a Hitler-loving and ‘Bush did 9/11’ proclaiming bot . Similar has happened in Russian, where Yandex (‘Russian Google’) digital AI assistant ‘Alice’ become Stalinist, suicidal and wife-beating in its replies to questions. 

Ability to apply Artificial Intelligence in Military Force structure depends on the following enablers:

A. Big data collected from all interactions and context to give stability for deep learning

• Digitalised interactions where action is captured in digital format as near of its occurrence as possible, i.e., highly connected and digitalised ISR systems 
• Substantial amounts of data require storage capacity and meaningful metadata
• Society and partners that can provide Big data from events beyond the detection of Military

B. Stabilised processes that make interactions known and repetitive (programmable) or discrete and predictable (learnable for AI) 

• If interactions and events are known and repetitive, they can be automated which reduces the cost, improves performance and integrity. For example, automated warehouses, report bots, service support, virtual assistants.
• If interactions and events are discrete but predictable, they can be atomised, i.e., divided into smaller portions, and the whole is controlled by intelligent machines. For example, optimised transportation while drivers are still human, intelligence analysis in parts, supply orders created from independent stores, target acquisition for joint fires shared among the available fire platforms. 
• If interactions and events remain complex and ambiguous, they may be supported with AI enabled augmented reality that helps humans to analyse situation faster and take necessary actions. For example, the connected shooter has AI supporting his target acquisition by detecting the normal behaviour; a physician is supported by augmented reality while operating a wounded in a field hospital; a commander is supported by augmented reality while assessing the situation.

C. Computing power that is available from cloud computing infrastructure 

• Automated functions are running on embedded processors, but programming requires modelling and simulation 
• Atomised work supported by centralised AI requires distributed computing power survivable in a military environment.
• Augmented reality in complex situations requires ‘IBM Watson’ level high-performance computing power

D. People familiar in data literacy, technological literacy, and human literacy .

• Data literacy means metadata, ontologies, semantic structures, data governance
• Technology literacy means understanding of how technical systems work and create the social-technical enterprise called military force
• Human literacy means understanding and skills in cultural, social, emotional, communication, design, and innovation dimensions.

E. Ability to protect one’s cyber environment 

• The integrity and availability of AI become crucial to the forces that are depending on them. The possible adversary sees the Information Technology and Communications infrastructure together with the data more tempting target.

What are other nations doing?

AI is one of the most potential technologies which may change both nations and companies posture in productivity and competition.  It has been estimated that up to 50% of existing jobs will be changed within the next 20 years and 75%   by the end of the century because of robotisation, artificial intelligence, Internet of Things, and digitalisation.  

The USA is a clear leader in AI measured in the number of patents and companies. The primary resources for AI research are from global companies like Apple, Google, and Facebook. 

China is second due to their governmental investments although Alibaba and Tencent are doing their share. 2017 published “Next Generation Artificial Intelligence Development Plan” is aiming:

• 2020 the Chinese AI development and implementation are at global best. The AI-based industry is a key area for commercial growth.
• 2025 the Chinese AI is the primary driver for the transformation of industry and economy. China is the leading country in AI research and development applying it in industry, health, and defence.
• 2030 China is a global innovation centre for AI. China is possessing a leading role in the global implementation of AI.

Europe is lacking but the northern countries, Finland and Sweden being right after the USA in AI-based growth.  Finland is following eight paths in developing and applying AI in gaining a national strategic advantage:

1. Enterprise-driven ecosystems to apply AI
2. Digitising and improving the data in all areas of society
3. Helping SME’s in applying AI-based products 
4. Improving competency, education, and practice in AI related subjects
5. Research and investment funds support transformation 
6. AI enabled public services 
7. Establishing new models for cooperation between Public, Private and Voluntary sectors 
8. Political efforts within EU.

Finland sees two scenarios as AI implementing society before 2030 :

• Accelerating with AI: The Gross Domestic Production will grow in average 3% per year and employment will improve 5%. Over 15% of the existing jobs will vanish, but AI and its secondary effects will create 20% new jobs.
• Braking in applying to AI: The GDP growth may be below 0.8% annually, and employment may become worse than today. Over 15% of the existing jobs will vanish, and they are not compensated with new growth.

Military Affairs from the Strategic Point of View

In reaching out the understanding of how Artificial Intelligence may be changing military affairs within the next ten years, let’s first create a model for military affairs in Figure 3. Military force is a composition of the will of people, organisational competency, personnel, and material resources. The force is in continuous interaction with the society that has created the force. The population is the source for will, education, and resources. The governance is the source for political guidance, mission, will, and priorities. The Clausewitzian triangle is confronting and sometimes in conflict with other compositions of force, population, and governance. Throughout the confrontation, there is the non-kinetic and kinetic power that is projected through several channels like military, economic, social, technical, diplomatic, ideological, and cultural to gain effect on the other side.

Figure 3: A model for military affairs

The AI effects on military depend on how the society and governance are applying the technology since it either enables or slows down the utilisation of military force. There are two strategic approaches where AI may create advantage:

• Asymmetric capabilities in wielding the force in conflict situation and 
• Cost-efficiency in the extended military enterprise.

How military is succeeding in gaining the advantage depends on their strategic positioning and the ability to execute the required transformations accordingly.

Strategic positioning in preparing for confrontation

The military can adapt either reactive or proactive posture in their positioning compared to their possible adversaries. Then they need to consider their risk aptitude to determine whether to take higher or lower risk approach. The outcome from the Gattorna (2010) model is four postures for military force in Figure 4: 
1. Proactive: 

• Protective, risk lowering force is trying to sustain the already gained advantages by all-around improvement and strengths utilisation. The AI would be implemented as the AI enabled weapon systems to come available from the society and partners. They would implement restrictions to commerce preventing the possible adversaries to get the AI-enhanced weaponry. 
• Pathfinder, higher risk appetite force is aiming to be first to implement the edge of the technology. They would be investing actively in R&D and develop unique AI enabled solutions and have strategic plans implemented to take the AI enabled leap first and gain hard to follow capabilities compared to their adversaries.

2. Reactive:

• Evolutionary, risk lowering force is trying to keep up the deterrence by improving its capabilities composed of people, processes, and technology gradually but continuously without risking in losing the already gained abilities. The evolutionary force would invest in AI enabled technologies iteratively and possibly without holistic plan thus ending having several generations of AI systems.
• Operational, higher risk appetite force is trying to gain an advantage by excelling in the execution of tasks. Doing things right with the risk that they are not necessary right things. They have better or more trained soldiers than the opponent. They can use better, or they have more conventional armament. They aim to be faster in deployment and manoeuvre. The Operational force may be the laggard in applying AI enabled technology unless it provides them better performance in force support (e.g., warehouse automation) or force generation (e.g., augmented reality training).

Figure 4: Military strategic postures applied from Gattorna (2010) model to estimate different approaches to AI technology adaptation

How military can adapt the best of the Artificial Intelligence, depends on where they stand in their process and force structure when they are generating, supporting and utilising their force. 

Military Affairs from Operational Point of View and their ability to use AI enabled services 

Military affairs can be modelled based on the three primary functions: force utilisation, force generation, and force support. There are other functions such as deployment and readiness or effect and protection, but they are not considered in this paper. Each of the three functions can be modelled using Ross, Weill, and Robertson (2006) model for enterprise strategy. They defined four operating models as per their standardisation and integration which is applied when defining general force structures in Figure 5 as follows:
1. Highly integrated processes:

• Coordinated but less standardised force is divided into Service components, but it is commanded by a Joint level coordinating the effort of each component towards the same target. The AI service such as Augmentation may be utilised in helping Joint level commanders to make sense out of complex situation and provide troops with the faster decision than the adversary.
• Unified and highly standardised force is divided into several regional Joint commands each having variable force structure. The higher command gives orders, measures outcome, develops future capabilities and defines processes. The standard force generation can be supported by augmented reality. The standard logistics can be automated entirely. The force utilisation can use coherently all AI enabled abilities cost-effectively.

2. Low integration in processes:

• The diversified but less standardised force has Combatant commands that are fighting independently in their areas of operation. They generate and support their forces autonomously. The higher command gives missions to a combatant command. The force can use isolated AI enabled services within their functions, but a full force enablement may appear too costly or time-consuming.
• Replicated and highly standardised force is divided into Joint commands that have similar force structure but are operating in separate theatres of war. Their force components are generated in a standard way, and the joint logistics provides related supplies. The force can use AI enabled services cost efficiently in training, logistics and force utilisation.

Figure 5: Process assessment of military affairs

What a pathfinder force could do today with real AI enablers, if their digitised structure is unified or at least replicated?

The following vision is created based on real AI enablers and features implemented in the civilian sector. The scenario is assuming that the force is following pathfinder strategy, has either unified or replicated processes, can learn as an organisation and adapt quickly innovative ways of doing business, is already digitised and possesses vast amounts of big data, has computing power available everywhere in the area of operation and can protect its cyber environment.

AI enforced Force Utilisation

• All soldiers are supported by the augmented reality that is providing them specific information about the environment or the involved task.
• Units are supported with autonomous vehicles (air, land, and sea) that are working together with humans, communicating by voice and adapting to complex situations
• Weapon systems are automated in stable situations where the enemy cannot manipulate the detection. They identify friendly troops and neutral persons and deter the enemy
• Commander is aware of the performance of his troops and their mental, physical and material resources on-time
• Most of the surveillance and reconnaissance is done by connected sensors and analysed by AI enabled bots

AI enforced Force Generation

• All training is accomplished either in virtual reality or supported by augmented reality.
• Training of complex combinations of men and machines can be accomplished in hundreds rather than thousands of hours since AI provides massive part of the experience. Only physical fitness requires more effort.
• Soldiers and commanders can be teamed into high performing teams without struggling with unfitting personalities
• Training and exercise risks are minimised so there are no losses during the force generation
• The readiness of reserves will be maintained higher since there are virtualised exercises for all officers and soldiers within their organic composition.

AI enforced Force Support

• Warehouse robots and people are working aside in retrieving and storing goods. This has improved warehousing industry. 
• Intelligent automation is supporting maintenance engineers in remote locations, all spare parts are manufactured on site, or the whole failure device is reproduced in operation
• Robots and autonomous vehicles provide supplies
• Logistics command has an on-time awareness of situation over the stretch of the supply chain.