Technical Sciences 157

ARTIFICIAL INTELLIGENCE IN MILITARY

APPLICATION – OPPORTUNITIES
AND CHALLENGES

István SZABADFÖLDI
National University of Public Service, Budapest, Hungary
[email protected]

ABSTRACT
Artificial Intelligence (AI) is playing an increasing role in planning and
supporting military operations and becoming a key tool in intelligence and analysis
of the enemy’s intelligence. Another field of application of AI is the field of
application of autonomous weapon systems and vehicles. The use of AI is expected
to have a greater impact on the military functions of human-machine interfaces
(machine-learning, man-machine teaming). AI promises to get over the
“3V challenge” (volume, variety and velocity) of Big Data, and is also expected to
reduce the risks concerning the other “2V” (veracity, value), and to render data
processing on a controlled level of decision based on AI’s knowledge. The aim of the
article is to provide an overview on the potentials of application of AI in the military
and to highlight the need to identify and define measurable indicators to evaluate
benefits of state-of-the-art technologies and solutions which are expected to improve
quality and performance of operations focusing on key areas as of situational
awareness and decision-making support and also logistic and operational planning
as well as modelling and simulation (M&S).

KEYWORDS: artificial intelligence, big data, analytics, autonomous systems,

machine learning

1. Introduction grown up out of the laboratories and

Since the beginnings of the mid penetrated the economy from industrial
1900s, the scientific field of Artificial design, through manufacturing, as well as
Intelligence (AI) went through periods of to the consumer’s services (NATO
intense development and drawbacks. From Parliamentary Assembly, 2019).
year 2000, however, a significant and The observers and experts forecast
persistent growth in AI research, high global economic impacts associated
development, and application can be with the application, development, and
perceived. AI research began to speed up adoption of AI over the next ten years, and
around 2001, and then commercial products a reasonable estimation put the range of
and services started permeating the market AI’s economic impact to be between $ 1.49
in large numbers especially since early trillion and $ 2.95 trillion in this period
2010s. Within roughly a decade, AI has (Chen et al, 2016).

DOI: 10.2478/raft-2021-0022
 158 Technical Sciences

The main objective of the article is to also called “weak AI”, is a computer system
survey and analyze the course of that can perform a narrowly defined task
development of AI, and provide an more efficiently than a human. That is where
overview of the current status of most of the AI applications extend today.
applicability and expectations in the – General Artificial Intelligence
military domain. The method of research (General AI), sometimes referred to as
was surveying and comparison of “Strong AI”, can outperform human
publications and policy papers of the achievements in any intellectual task.
NATO and the EU, analysis of forecasts For example, we can see robots with this
and reports of AI state-of-the-art type of AI in movies where they act
techniques, and presenting opinions of AI according to their own goals based on
commercial and military solution providers. conscious thoughts.
– Artificial Super Intelligence (ASI)
2. Research Methodology is expected to be able to outperform
The research methodology for the humans in almost all areas, but especially in
topic was based on a review of the relevant scientific creativity, logic and wisdom, and
scientific literature and interviews with also in social skills.
military personnel. It is important to note Many scientists have a strong doubt
and understand that the access to available that this last would ever be achievable at
sources is limited in details ‒ especially all. Furthermore, even if the ASI or close to
concerning the technical and functional that level of AI will be reached, the human
implementation of the required features and contribution to the process of AI application
capabilities ‒ by the nature of the relative at some point remains a must, since ethical,
novelty and confidential sensitivity of the intuitional and especially spiritual motives
technologies. The methodology is there for and capabilities are exclusively human
limited to open access sources and materials. characteristics, even if, to some extent,
machines can emulate them.
3. What Is Artificial Intelligence – A subset of MI is machine learning
Overview and Demystification (ML), which teaches computers with
Artificial Intelligence, although a term mathematical data models with or without
used since the 1950s, has still not a generally direct supervision. Machine learning uses
accepted definition today. AI cannot be algorithms to identify patterns in the data
interpreted as a stand-alone application, but creating a data model to provide predictions
as a technology that supports existing and responses. The advanced level of ML is
functional applications and is eventually Reinforcement Learning when the system is
based on algorithms designed to solve reinforced with positive feedbacks in the
specific problems, collecting, organizing, recognitions.
processing, analyzing, transmitting, and In the case of Deep Learning, the
responding to larger data sets, suitable and machine is taught with complex amounts of
capable of corresponding to the cognitive data for complex tasks with the help of
ability of the human intellect, and neural networks created by the analogy of
operations approaching it. the human brain, in which the neurons
There are basically three types of AI: perform a sub-function or summarize them.
– Narrow Artificial Intelligence Here it is important to mention the so-called
(Narrow AI). Narrow artificial intelligence, Black Box phenomenon, in which the

REVISTA ACADEMIEI FORŢELOR TERESTRE NR. 2 (102)/2021

Technical Sciences 159

human being is no longer able to follow the The democratization of technology

process taking place in the individual raises the security concern because of the
neuronal levels. So, they pose a significant reduced costs and ease of access. There is a
reliability risk. need of higher level of regulation to protect
There are increasing number of the society from technologically advanced
publications and public events on AI, malicious actors who may be capable to
nevertheless or rather because of these, produce highly threatening equipment and
widespread myths and misconceptions solutions based upon easily available
remain about what AI really is and what AI knowledge. Although Chemical, Biological,
systems can do. These misrepresentations Radiological and Nuclear technologies
make it difficult and unproductive to (CBRN) are generally highly regulated,
understand and discuss the opportunities however these regulations may be
and risks of AI in general and in the inadequate if “anyone can set up a bio-
security field in particular (Saalman, 2019). engineering laboratory in a backyard or
The possibilities and limitations of AI basement” (NATO Science & Technology
technology have to be clearly understood Organization, 2020, p. 35).
and considered especially for decision makers
to avoid creating illusions and launching 5. AI Development and Application
projects with hardly achievable targets. in Military
5.1. The Course of Development of AI
4. Emerging and Disrupting AI went through three development
Technologies (Edts) stages since the concept came to existence
Before AI is discussed in more details
somewhat seventy years ago (Figure no. 1).
it is important to put it in a wider context of
In the first stage the solutions focused on
other technologies which are highly
rules-based approaches like decision trees,
interrelated scientific areas and are
Boolean and fuzzy logic and were generally
considered as major strategic influencers
called expert systems. In the second stage
over the next 20-years. Some of these areas
AI developers focused on the development
are momentarily in early stages of
and application of statistical methods which
development but are expected to go through
resulted in the concept and methodology of
rapid development. These EDTs are:
machine learning. These were successfully
Big Data, Artificial Intelligence (AI),
coming up with solutions such as e-mail
Autonomous Vehicles, Space, Hypersonic
spam filtering and internet search engines.
Aerial Vehicles, Quantum technology,
The third stage of development, which is
Biotechnology, Novel Materials (NATO
going on now, introduced the use of
Science & Technology Organization, 2020).
human-like learning methods such as neural
These EDTs are expected to make
networks, defining the concept and
comprehensive societal impacts, infiltrating
technology of deep learning, and proved to
in telecommunications, entertainment, medical
be successful in sensing and perception
sciences, human-computer interactions,
(NATO Science & Technology
financial services, agriculture, transport
Organization, 2020).
industry, industrial manufacturing and
public security and safety as well.

REVISTA ACADEMIEI FORŢELOR TERESTRE NR. 2 (102)/2021

 160 Technical Sciences

Figure no. 1: The course of development of AI

(Source: United States Government Accountability Office (GAO), 2018)

The Gartner Hype Cycle for AI techniques in general, which is typical in

represents the expected course pf every emerging technology adaptation.
development and application of AI

Figure no. 2: Gartner Hype Cycle for AI

(Source: Gartner, 2019)

REVISTA ACADEMIEI FORŢELOR TERESTRE NR. 2 (102)/2021

 Technical Sciences 161

Deep-learning methods are expected Intelligence Center (JAIC) will see its
to be further developed, besides, new budget double to over $ 208 million, with
research areas including neuromorphic significant increases likely in 2021 and
computing are raising, emulating the neural beyond ... The military is currently seeking
structure and operation of the human brain. to integrate AI into weapon systems
Together with these, adversarial machine development, augment human operators
learning techniques are being developed to with AI-driven robotic maneuver on the
understand how to confuse enemy’s AI battlefield and enhance the precision of
systems. So basically, as of today the military fires” (DoD Growth in Artificial
defense of integrity of AI systems is Intelligence: The Frontline of a New Age in
becoming a critical challenge parallel with Defense, 2019).
the proactive application. AI already has deep impacts on
nuclear armament, cyber and information
5.2. Expectations for AI in Military warfare, materials and bio-technologies,
As for the expectations concerning AI aerospace, and considered that these effects
in military application over the next decade, will have the similar result on global world
some of its techniques will define or order that the appearance of nuclear
re-define key advanced military technologies. weapons created. The AI systems are also
AI solutions will be first of all expected to bring about an adversarial AI
intelligent, exploiting integration of AI arms-race.
knowledge-focused analytic capabilities. Machine learning uses mathematical
Then the AI solutions will be algorithms and formulas to extract patterns
interconnected to utilize the network of from masses of data. However, if an
virtual and physical domains, including adversary can see enough of the inputs and
sensors, organizations, individuals and the outputs of our AI system, they can
autonomous agents, also exploiting the deduce what algorithms are operating, just
advantage of blockchain technology for as reverse engineering works. The whole
data integrity. They will be distributed to game may be turning into a battle between
utilize decentralized and ubiquitous large- adversarial groups of mathematicians,
scale sensor networks, storage, and similarly to the codebreaking competition
computation. And last but not least, they struggle of World War II and the Cold War
will be digitally intermingling human, (Artificial Intelligence the Frontline of a
physical and information domains to New Age in Defense, n.d.).
support novel disruptive effects (NATO Let’s examine some specific AI
Science & Technology Organization, military applications in important domains
2020). where the revolutionary effects are
expected and can already be perceived in
5.3. Key Areas to Examine of AI leading military powers.
Application Advances In case of C4ISR (Command, Control,
“The Pentagon is figuring ways to Communications, Computers, Intelligence,
harness artificial intelligence (AI) for Surveillance and Reconnaissance), the
advantages as far-flung as battlespace combatting units use trusted AI-enabled
autonomy, intelligence analysis, record autonomous systems that are capable of
tracking, predictive maintenance and carrying out tasks that are deemed “dull,
military medicine. AI is a key growth dirty, dangerous or dear” and provide
investment area for DoD, with nearly AI-enabled decision support to war-
$1 billion allocated in the 2020 budget. scenarios and AI-recommended courses of
The Defense Department’s Joint Artificial action (COA). AI supported data fusion,

REVISTA ACADEMIEI FORŢELOR TERESTRE NR. 2 (102)/2021

 162 Technical Sciences

categorization and targeting provide the electromagnetic spectrum, the space,

intelligence analysts higher capability of cyberspace, air, land and maritime domains.
TCPED (tasking, collecting, processing, The project made assessment on the
exploiting, disseminating) and retrieving obstacles to implementation/integration of
information. AI integration into C4ISR AA&R capabilities and strategies to
process results better indications and overcome hindrances. The project objective
warning, information and knowledge aimed to provide assessment of present and
management tools, resulting more reliable future military tasks and functions that
intelligence analyses. could benefit from AA&R considering also
Autonomous systems, UxVs paybacks in efficiency and cost-savings
(Unmanned Vehicles like UAV, UGV, (NATO Allied Command Transformation
AUV etc.) can be operating on a much Operational Experimentation, 2020).
higher efficiency and safety level with AI
support. The implementation of deep 6. Interrelation of AI and Big Data
learning systems into unmanned platforms Advanced Analytics (BDAA) in Defense
will significantly extend robotic capabilities Artificial Intelligence is strongly
for navigation. connected and interrelated to Big Data
AI supports long term Capability technology. Data feed for efficient AI
Planning by the development of analytical application is a key success factor and the
solutions, including supporting complex data feed has a wide variety of sources.
decision-making by assessments of Therefore, Big Data technological
complex factors. development has key impact on AI
AI technology improves autonomy to solutions and vice versa.
achieve sensor integration and data fusion Big Data technology accomplishes
to satisfy requirements for fast and reliable data processing and analysis that brings
detection, identification, and monitoring forth significant volume, velocity, variety,
(DIM) of CBRN hazards. veracity and visualization challenges.
Digitalization, spreading of sensor
5.4 Military Uses of Artificial technology, IoT and virtualization of social
Intelligence, Automation, and Robotics media played significant role in the
(MUAAR) Project development of Big Data.
Having realized the importance of
technological lead of AI development and 6.1. Expected Impact of AI and
application, the NATO launched the BDAA in Defense
Military Uses of Artificial Intelligence, Big Data Advanced Analytics
Automation, and Robotics (MUAAR) encompasses analytical methods for
project under the Multinational Capability understanding and visualizing masses of
Development Campaign (MCDC) in 2020. information.
The project’s scope was to develop BDAA has four basic components:
concepts and capabilities to address the – collection (sensors);
challenges of conducting joint coalition – communication;
operations and provide assessments on – analysis;
them. – decision making (NATO Science
The project examined areas of the & Technology Organization, 2020).
logistics, integrated air & missile defense,

REVISTA ACADEMIEI FORŢELOR TERESTRE NR. 2 (102)/2021

Techniccal Sciences 163

Figure no.
n 3: BDAA A Goals
(S
Source: NAT
TO Sciencee & Technollogy Organiization, 20220)

Seensors are key for collecting data

d peerformance together w with lower power,
and theey are com ming from smart
s textilles, prooviding robbust sensinng capabilitties and
over-thee-horizon (OTH) and passive raddar wiireless com
mmunicationn on the baattlefield
systemss, quantum sensing, Digital
D Twwins (N
NATO Allieed Commaand Transfo ormation
(detailedd virtual models ofo a weappon Opperational Experimenta
E ation, 2020)).
system)), Computattional Imagging (CI), and
a Figgure no. 4 illustrates the interrellation of
Microw wave photoonics delivvering highher AII, ML and BDAA.
B

F
Figure no. 4:
4 The Relattionship bettween AI, Machine
M Leaarning and D
Data Sciencce
(Source: Torrres, Hart, & Emery, 2019)

6.2. BDAA Impacts on Speciific opperational domains,

d BDDAA will result
r in
Militaryy Areas mo ore efficiennt ISR annd is expeected to
W the rapiid developm
With ment and coost- proovide higheer level Situuational Aw
wareness
efficiennt access to sensoors of hiigh (SA) by advvanced maapping of mission
sensitivvity and thhe use off autonomoous areeas for booth planningg and prep paration.
systemss, raising thhe ability to detect anda It will have increasing impact in Training
T
classifyy threats acrross physiccal and virttual an
nd Readiness, and w will result in
i more

REVISTA AC
CADEMIEI FORŢELOR TE
ERESTRE NR
R. 2 (102)/2021
 164 Technical Sciences

efficient Enterprise Management of the submitting proposals to be funded by EU

organizations, optimizing business processes, EDF fund (European Commission, 2020).
as well as real-time monitoring of the Further research is necessary in the
results of decisions. BDAA can provide areas of Human-Machine Teaming, since
predictive assessments through advanced human involvement in military operations
modeling and simulation. Logistics will remain no matter how advanced the AI
benefit from BDAA as well because of the techniques are. Machine and deep learning,
increased integration of the sensors adversarial AI, and neuromorphic and
monitoring the weapon systems. Real-time probabilistic computing in order to achieve
inventory monitoring can also be achieved advanced algorithms are also areas
and the use of digital twins increases the requiring further research to provide
effectiveness of the logistic system. effective algorithms.
BDAA has key role in Support of Since AI is a technology that can
Operations via large quantities of sensor raise the efficiency of predictive and
data, and combining with AI, it provides a cognitive data analytics, the need for
comprehensive approach to operational processes to transform both structured and
planning (OPLAN), courses of actions unstructured data into insights for decision-
analysis and targeting. makers is key for applicability. Structured
data can be processed by machine learning
7. Critical Challenges of AI in Areas techniques by neural networks, while
of Research unstructured data is processed by using
In order to achieve and fully exploit deep learning and natural language technics.
the opportunities and capabilities mentioned These tools may be applied to achieve
above, Artificial Intelligence needs further more realistic analysis and personalized
advances in the following areas. training within modeling and simulation,
One of them is solving the Black Box adaptive techniques in electronic warfare
Problem. Black box can be a device or and development of AI agents for defensive
system that allows seeing the input and and offensive cyber operations throughout
output, but does not provide view of the the entire Infospace. The information gained
through AI can be weaponized and hybrid-
internal processes. The AI Black Box
warfare countermeasures can be achieved.
means that in many AI-based tools the
algorithm processing sections and nodes
8. Conclusions
which are sequentially connected to refine
AI is definitely opening new
data and results are not transparent in terms
perspectives in defense technologies. There
of exact input and output. This is especially are high expectations concerning the
true in case of Deep learning utilizing a application of AI techniques in several
large artificial neural network, with many military domains; however there are still
hidden layers, and it ‘learns’ on its own by hindering factors and unsolved issues for
recognizing patterns layer by layer. further research in order to fulfill those
The Explainable AI is expected to be expectations.
the solution for Black Box problem, as an As it was put by Breaking Defense:
AI tool that will provide transparent results “what we call artificial stupidity: the ways
on those nodes of layers (Zednik, 2019). algorithms can misinterpret the world in
That functionality is not yet available ways no human ever would, because they
in each solution, so the Black Box problem interpret data in terms of mathematics and
remains a significant risk to which the EU logic without instinct, intuition, or common
EDIDP called the attention to avoid in sense. It turns out such artificial stupidity is

REVISTA ACADEMIEI FORŢELOR TERESTRE NR. 2 (102)/2021

Technical Sciences 165

something you can artificially induce. of a New Age in Defense, n.d., p. 17).
The most famous example is an experiment It does not mean, however, that we have to
in which strategically applied reflective be sceptic concerning AI widespread
tape caused the AIs used in self-driving applicability, but rather understand that
cars to misclassify a STOP sign as a speed more intellectual effort and money should
limit” (Artificial Intelligence the Frontline be poured into the R&D.

REFERENCES

Artificial Intelligence the Frontline of a New Age in Defense. (n.d.). Breaking Defence,
available at: https://cdn2.hubspot.net/hubfs/2097098/MCM120_BreakingDefense_AI_ebook
R1%20(1).pdf, accessed on 29 March 2021.
Chen, N., Christensen, L., Gallagher, K., Mate, R., & Rafert, G. (2016). Global
Economic Impacts Associated with Artificial Intelligence, 23.
DoD Growth in Artificial Intelligence: The Frontline of a New Age in Defense. (2019).
Breaking Defence, available at: https://breakingdefense.com/2019/09/dod-growth-in-artificial-
intelligence-the-frontline-of-a-new-age-in-defense/, accessed on 09 April 2021.
European Commission. (2020). European Defence Industrial Development Programme
(EDIDP), available at: https://ec.europa.eu/research/participants/data/ref/other_eu_prog/
edidp/wp-call/edidp_call-texts-2020_en.pdf, accessed on 09 April 2021.
Gartner. (2019). Gartner Hype Cycle for Artificial Intelligence.
NATO Parliamentary Assembly. (2019). Artificial Intelligence: Implications for NATO
Armed Forces. NATO PA 149 STCTTS 19 E rev.1 fin, Science and Technology Committee.
NATO Allied Command Transformation Operational Experimentation. (2020). Military
Uses of Artificial Intelligence, Automation, and Robotics (MUAAR), available at:
https://www.act.nato.int/application/files/5515/8257/4725/2020_mcdc-muaar.pdf, accessed
on 19 March 2021.
NATO Science & Technology Organization. (2020). Science & Technology Trends
2020-2040. Brussels, Belgium, available at: https://www.nato.int/nato_static_fl2014/
assets/pdf/2020/4/pdf/190422-ST_Tech_Trends_Report_2020-2040.pdf, accessed on 19
March 2021.
Saalman, L. (2019). The Impact of Artificial Intelligence on Strategic Stability and
Nuclear Risk. Vol. II, Sweden: Stockholm International Peace Research Institute.
Torres, T.M., Hart, G., & Emery, T. (2019). The Dstl Biscuit Book Artificial
Intelligence, Data Science and (mostly) Machine Learning. 1st edition revised, available at:
https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/
file/850129/The_Dstl_Biscuit_Book_WEB.pdf
United States Government Accountability Office (GAO). (2018). Report to the
Committee on Science, Space, and Technology, House of Representatives. GAO-18-142SP,
available at: https://www.gao.gov/assets/gao-18-142sp.pdf
Zednik, C. (2019). Solving the Black Box Problem: A Normative Framework for
Explainable Artificial Intelligence, available at: https://arxiv.org/ftp/arxiv/papers/
1903/1903.04361.pdf, accessed on 06 April 2021.

REVISTA ACADEMIEI FORŢELOR TERESTRE NR. 2 (102)/2021