(Abridged version published in SP’s Military Year Book 2017)
Intelligent machines were the focus of research work at many institutes after the WWII. In 1950, Alan Turing argued that if the machine could successfully pretend to be human to a knowledgeable observer then one certainly should consider it intelligent[i]. The credit of coining the phrase ‘Artificial Intelligence’ goes to John McCarthy in 1955. A number of scientists have defined Artificial Intelligence, (AI) in a varying manner; however, there appears to be no single definition, which has been universally accepted. All the definitions of AI are connected with human intelligence in some way, some of them are:
– “The study of mental faculties through the use of computational models”[ii].
-“The art of creating machines that perform functions requiring intelligence when performed by people”[iii].
-“A field of study that seeks to explain and emulate intelligent behavior in terms of computational processes”[iv].
– “The study of how to make computers do things at which, at the moment, people are better”[v].
– “The study of the computations that make it possible to perceive, reason, and act”[vi].
– “The branch of computer science that is concerned with the automation of intelligent behavior”[vii].
Strong AI has been defined as that moment when “humankind is in the presence of an intelligence greater than its own”[viii], and as “strong AI is reached once the computer regarded as such is conscious of its abilities”[ix].
AI imbibes knowledge from different fields like Computer Science, Mathematics, Engineering, Cognitive Science, Philosophy, and Psychology. AI embodies a wide range of intelligent search methods, techniques for obtaining clarity where uncertainties exist in data and knowledge, and various types of machine learning & representation schemes of knowledge. Its various applications include, speech recognition, natural language processing, expert systems, neural networks, intelligent robotics, gaming and 3D vision. There is a need to define machine learning and deep learning before moving on to the military applications of AI.
Machine learning. It has evolved from the study of computational learning theory, pattern recognition, and artificial intelligence. It is a subfield of computer science.[x] It has been defined in 1959 by Arthur Samuel as a “Field of study that gives computers the ability to learn without being explicitly programmed”. Machine learning relies upon utilizing algorithm constructions to perform predictive analysis on data[xi]. Machine learning tasks fall into three basic categories namely[xii]; Supervised learning is one in which the computer is presented with example inputs and their desired outputs, and the goal is to learn a general rule that maps inputs to outputs; Unsupervised learning is one where no labels are given to the learning algorithm, leaving it on its own to find structure in its input; and Reinforcement learning is one where a computer program interacts with a dynamic environment in which it must perform a certain goal.
Deep Learning. Le Deng and Dong Yu of Microsoft have provided the following definitions for Deep Learning[xiii]:
-A class of machine learning techniques that exploit many layers of non-linear information processing for supervised or unsupervised feature extraction and transformation, and for pattern analysis and classification.
-A sub-field within machine learning that is based on algorithms for learning multiple levels of representation in order to model complex relationships among data.
-A sub-field of machine learning that is based on learning several levels of representations, corresponding to a hierarchy of features or factors or concepts, where higher-level concepts are defined from lower-level ones, and the same lower level concepts can help to define many higher-level concepts.
Some of the deep learning architectures built around neural networks are deep belief networks, deep neural networks and recurrent neural networks. The use of deep learning architectures in automatic speech recognition, bioinformatics, natural language processing, and 3D vision etc has resulted in remarkable successes.
As per Jeff Hawkins and Donna Dubinsky of Numenta, building of smart machines has involved three basic approaches. These are the Classic AI, Simple Neural Networks, and Biological Neural Networks.[xiv]
The classic AI approach involved computer programs that were based upon abilities of the human brain to solve simple problems. However, the computers required large amounts of inputs from knowledge experts to lay down the rules based upon their expertise and experience in problem solving. Thus, the classic AI systems were created specific to a problem, while they were very useful in case of problems which had been defined in detail they could not learn on their own and provide solutions to problems. They failed in comparison with general human intelligence.
When the limitations of Classic AI were encountered, scientists looked at the functioning of the human brain at the level of neurons and this resulted in Artificial Neural Networks (ANNs). The ANNs lay emphasis upon unsupervised learning from data provided to them. Thus, the Simple Neural Networks learn from data and do not require experts to lay down the rules. The Simple Neural Network is a mathematical technique that locates patterns in large, static data sets[xv]. The ANNs are a subset of machine learning techniques that processes large amount of data using statistical and mathematical techniques in addition to ANNs to provide results. ANNs have transformed into Deep Learning networks with the advent of humongous data and fast computers. Thus, Simple Neural Networks could provide solutions where Classic AI could not. However, the Simple Neural Networks too have limitations when data is dynamic or when data is limited for training.
In the Biological Neural Approach, emphasis is laid upon studying how a human brain works to cull out the properties that are required for intelligent systems. It is established that, information is represented in the brain using sparse distributed representations or SDRs. Further, it is known that memory is a sequence of patterns, behavior is essential part of learning, and that learning has to be continuous. Therefore, the building blocks of intelligent machines should be SDRs[xvi]. The biological neuron is also not as simple as conceived during the Simple Neural Network approach.
Military applications of AI can be found in almost all aspects of military from decision-making, equipment operations, sensors, weapons systems to unmanned vehicles. The military is adopting AI mainly because it results in much fewer casualties, higher efficiency, and lower costs. Intelligent robotics and unmanned vehicles for army, navy, and air force are bringing in a new revolution in standoff warfare. The war against terrorism is practically being fought with unmanned weaponized aerial vehicles in Afghanistan, Syria and Iraq. Be it air traffic control in a combat zone, which would allow manned and unmanned aircraft, weapons etc. to operate without conflict by automated routing and planning; or military decision making in fog of war; or a radar’s target identification algorithms which look at the shape of possible targets and their Doppler signatures; AI is integral to all these systems. In this article two major categories of military applications are discussed which pertain to cyber defence and military logistics.
Applications of AI in Cyber Defence
In 2009, Conficker[xvii] worm infected civil and defence establishments of many nations, for example, the UK DOD reported large-scale infection of its major computer systems including ships, submarines, and establishments of Royal Navy. The French Naval computer network ‘Intramar’ was infected, the network had to be quarantined, and air operations suspended. The German Army also reported infection of over a hundred of its computers. Conficker sought out flaws in Windows OS software and propagated by forming a botnet, it was very difficult to weed it out because it used a combination of many advanced malware techniques. It became the largest known computer worm infection by afflicting millions of computers in over 190 countries.
It s evident that the amount of data and the speeds at which processing is required in case of cyber defence is not feasible for human beings to carry it out. Conventional algorithms also cannot tackle dynamically changing data during a cyber attack. It appears that cyber defence can only be provided by real time flexible AI systems with learning capability.
The US Defence Science Board report of 2013[xviii] states that “in a perfect world, DOD operational systems would be able to tell a commander when and if they were compromised, whether the system is still usable in full or degraded mode, identify alternatives to aid the commander in completing the mission, and finally provide the ability to restore the system to a known, trusted state. Today’s technology does not allow that level of fidelity and understanding of systems.” The report brings out that, systems such as automated intrusion detection, automated patch management, status data from each network, and regular network audits are currently unavailable. As far as cyber defence is concerned in the US, it is the responsibility of the Cyber Command to “protect, monitor, analyze, detect, and respond to unauthorized activity within DOD information systems and computer networks”[xix]. The offensive cyber operations could involve both military and intelligence agencies since both computer network exploitation and computer network attacks are involved. The commander of Cyber Command is also the Director of National Security Agency, thus enabling the Cyber Command to execute computer exploitations that may result in physical destruction of military or civilian infrastructure of the adversary. Some advance research work in respect of active cyber defence has been demonstrated under various fields of AI, some successfully tested examples are:
Artificial Neural Networks- In 2012, Barman, and Khataniar studied the development of intrusion detection systems, IDSs based on neural network systems. Their experiments showed that the system they proposed has intrusion detection rates similar to other available IDSs, but it was at least ~20 times faster in detection of denial of service, DoS attacks[xx].
Intelligent Agent Applications-In 2013, Ionita et al. proposed a multi intelligent agent based approach for network intrusion detection using data mining[xxi].
Artificial Immune System (AIS) Applications- In 2014, Kumar, and Reddy developed a unique agent based intrusion detection system for wireless networks that collects information from various nodes and uses this information with evolutionary AIS to detect and prevent the intrusion via bypassing or delaying the transmission over the intrusive paths[xxii].
Genetic Algorithm and Fuzzy Sets Applications- In 2014, Padmadas et al. presented a layered genetic algorithm-based intrusion detection system for monitoring activities in a given environment to determine whether they are legitimate or malicious based on the available information resources, system integrity, and confidentiality[xxiii].
Miscellaneous AI Applications- In 2014, Barani proposed genetic algorithm (GA) and artificial immune system (AIS), GAAIS – a dynamic intrusion detection method for Mobile ad hoc Networks based on genetic algorithm and AIS. GAAIS is self-adaptable to network changes[xxiv].
From the above it can be seen that there is rapid progress in design and development of cyber defence systems utilizing AI that have direct military applications.
Applications of AI in Military Logistics
Some of the challenges being faced by militaries in both peace and war include ensuring the adequacy of maintenance and repair of sophisticated equipment, weapons, armament and ammunition; ensuring the supportability of missions with due planning; and guaranteeing the availability of qualified personnel to carry out the assigned tasks. AI and associated technologies have made impressive inroads in civil and military logistics to ease the cumbersome operations and procedures involved. It has now been established that AI has significantly improved the systems and processes in the logistic chain and has led to considerable savings for the military establishments. AI encompasses many innovative technologies that are being used in military; some of these are discussed in succeeding paragraphs.
-Expert systems are software programs that usually serve as intelligent advisors in specific areas of expertise. Expert system technology has percolated to all functional areas of production and logistics of the military. Logistics expert systems in areas of inventory management, transportation, warehousing, acquisition, maintenance, and production are common. Examples include, the Inventory Manager’s Assistant of US Air Force, Dues Management Advisor (DMA) of the US Navy and Logistics Planning and Requirements Simplification (LOGPARS) system of the US Army.
-Natural language systems convert languages into computer language, thus making it feasible to communicate with computers in language of choice obviating the need to master computer languages. Natural language applications are being used to provide user-friendly query capability for large databases pertaining to logistics.
-Speech recognition systems allow user to interact directly with computers thus eliminating the use of keyboards. The voice signal is digitized and compared with stored voice patterns and grammatical rules for computer to understand the voice message. For example, US Air Force Logistics Command (AFLC) is using a speech recognition system in its depot warehouses to interface with the warehouse’s automated storage module (ASM); the US Army is using speech recognition system in association with a diagnostic system for carrying out maintenance of its motor vehicles as well as in its transportation planning[xxv].
-3D vision technology allows a computer to “sense” its environment and classify the various objects in its vicinity. The US Navy is using this in its Rapid Acquisition of Manufactured Parts (RAMP) program and the US Air Force for reverse engineering parts in its maintenance facilities. 3D vision applications are of significant importance in using robotics for logistics.
-Intelligent robots incorporate a host of AI technologies to mimic specific work undertaken by human beings. Mobile robots are being increasingly utilized in activities from patrolling to investigating and neutralizing explosives[xxvi]. Mobile robotic systems are used for carrying out routine inspections of nuclear missiles. They have eliminated the need of human element from going into containment systems. The robot is remotely operated from outside the containment zone. As far as arming of robots (Lethal Autonomous Weapons) is concerned, thousands of scientists and technologists, including, Elon Musk, Stephen Hawking, and Steve Wozniak signed an open letter in 2015 asking for a ban on lethal weapons controlled by artificially intelligent machines[xxvii]. The letter states “Artificial Intelligence (AI) technology has reached a point where the deployment of such systems is—practically if not legally—feasible within years not decades, and the stakes are high: autonomous weapons have been described as the third revolution in warfare, after gunpowder and nuclear arms.”
-Neural networks are designed based upon models of the way a human brain functions. They are capable of associative recall and adaptive learning. Because of the massive processing power associated with such networks, they are being increasingly utilized in logistic applications. Eyeriss is a new microchip fabricated at MIT and funded by DARPA that has the potential to bring deep learning to a smart phone that can be carried by a soldier[xxviii].
DRDO and AI
Centre for Artificial Intelligence and Robotics (CAIR), Bengaluru and Research and Development Establishment (Engineers) R&DE(E), Pune are the main laboratories of Defence Research and Development Organisation (DRDO) in India working in the area of artificial intelligence and robotics. A family of robots that have been developed for various surveillance / reconnaissance applications include[xxix]; RoboSen mobile robot system for patrolling, reconnaissance, and surveillance. It is capable of autonomous navigation with obstacle avoidance capability and continuous video feedback; Miniature Unmanned Ground Vehicle (UGV) is a ruggedized man-portable robotic system for low-intensity conflicts; Walking robots with six and four legs for logistics support; and Wall climbing & flapping wing robots for potential usage in Low Intensity Combat (LIC) operations.
Some projects under development include[xxx]:
-AI Techniques for Net Centric Operations (AINCO) – A suite of technologies for creation of knowledge base, semantic information reception and handling, inference reasoning, and event correlation.
-Knowledge Resources And Intelligent Decision Analysis (KRIDA) – A system that aims to achieve the management of large-scale military moves using extensive knowledge base and data handling.
-INDIGIS 2D/3D – An indigenous Geographic Information System (GIS) kernel that provides platform for development of display, analysis, and decision support involving spatio-temporal data.
-S57 Viewer – for viewing more than one lakh tracks.
-IVP_NCO and IP Lib – A comprehensive suite of image and video processing applications to provide a unified solution to image and video processing in the net-centric operations.
-Indigenous Network Management System (INMS) – An indigenous NMS with resource planning, network planning, and network monitoring tools for IP network management.
Future of Military Artificial Intelligence
The global defence sector has seen unprecedented adoption of unmanned systems and robotics. This has been mainly due to various factors like; reduction in own casualties and feasibility of riskier missions using robots; high precision, minimal collateral damage, longer endurance and range; quicker reaction times with greater flexibility; and finally cost benefits accruing due to reduction in cost of technology with increased percolation. Unmanned aerial systems comprise as much as over 80% of all military robots, in past six years US spending on military UAVs has increased by ten times[xxxi]. Today over 90 countries are operating drones with over 30 armed drone programs. Many programs including, Drone mother ships in air and water; swarm warfare on land, sea and air; high definition real time ISR; wearable electronic packages for soldiers with exoskeletons; and exotic weapon systems are likely to be inducted within the coming decade. The threat of cyber attacks on the AI systems is very real. AI Machines are connected to the human controllers for taking and executing critical commands, the linkages can be hacked both through electronic warfare as well as cyber attacks. Since AI runs entirely on software, there is a finite probability of it being manipulated and used against the owner. DARPA had run a three year ‘Cyber Grand Challenge’[xxxii] to accelerate the development of advanced, autonomous systems that can detect, evaluate, and patch software vulnerabilities before adversaries have a chance to exploit them. The competition which ended on 4th of Aug 2016, achieved its aim to prove the principle that machine-speed, scalable cyber defense is possible. This would mark the beginning of a new era in much needed cyber defence of AI systems.
As far as AI is concerned it suffices to quote US deputy secretary of defense, Robert Work “…the 2017 fiscal budget request will likely ask for $12-$15bn for war gaming, experimentation and demonstrations to test out the military’s theories on AI and robotics ‘in human-machine collaboration combat teaming’…”[xxxiii]
[ii] Charniak, E., & McDermott, D. Introduction to artificial intelligence. Addison-Wesley Longman Publishing Co., Inc. Boston, MA, USA ©1985,ISBN:0-201-11945-5
[iii] Kurzweil, R. (The Age of Intelligent Machines. MIT Press, Cambridge, Massachusetts
[iv] Schalkoff, R. I. Artificial Intelligence: An Engineering Approach .McGraw-Hill, New York.
[v] Rich, E., and Knight, K. Artificial Intelligence. McGraw-Hill, New York, second edition.
[vi] Winston, P.H. Artificial Intelligence. Addison-Wesley, Reading, Massachusetts, third edition.
[vii] Luger, G.F. and Stubblefield, W.A. Artificial Intelligence: Structures and Strategies for Complex
Problem Solving. Benjamin/Cummings. Redwood City, California, second edition.
[viii] Barrat, James. Our Final Invention: Artificial Intelligence and the End of the Human Era. New York, NY: St. Martin’s Press.
[ix] Russell, Stuart, and Peter Norvig. Artificial Intelligence: A Modern Approach. Montreal, QC: Prentice Hall.
[xi] Ron Kohavi; Foster Provost (1998). “Glossary of terms”. Machine Learning. 30: 271–274.
[xii] Russell, Stuart; Norvig, Peter . Artificial Intelligence: A Modern Approach (2nd ed.). Prentice Hall. ISBN 978-0137903955.
[xiii] Li Deng and Dong Yu, Deep Learning: Methods and Applications. https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/DeepLearning-NowPublishing-Vol7-SIG-039.pdf
[xiv]Jeff Hawkins & Donna Dubinsky, What Is Machine Intelligence Vs. Machine Learning Vs. Deep Learning Vs. Artificial Intelligence (AI)? http://numenta.com/blog/machine-intelligence-machine-learning-deep-learning-artificial-intelligence.html
[xviii] Office of the Under Secretary of Defense for Acquisition, Technology and Logistics, Resilient Military Systems and the Advanced Cyber Threat, United States Department of Defense, Defense Science Board, January 2013
[xix] U.S. Government Accountability Office, “Defense Department Cyber Efforts,” May 2011, 2–3, http://www.gao.gov/new.items/d1175.pdf.
[xx] D. K. Barman, G. Khataniar, “Design Of Intrusion Detection System Based On Artificial Neural Network And Application Of Rough Set”, International Journal of Computer Science and Communication Networks, Vol. 2, No. 4, pp. 548-552
[xxi] I. Ionita, L. Ionita, “An agent-based approach for building an intrusion detection system,” 12th International Conference on Networking in Education and Research (RoEduNet), pp.1-6.
[xxii] G.V.P. Kumar, D.K. Reddy, “An Agent Based Intrusion Detection System for Wireless Network with Artificial Immune System (AIS) and Negative Clone Selection,” International Conference on Electronic Systems, Signal Processing and Computing Technologies (ICESC), pp. 429-433.
[xxiii] M. Padmadas, N. Krishnan, J. Kanchana, M. Karthikeyan, “Layered approach for intrusion detection systems based genetic algorithm,” IEEE International Conference on Computational Intelligence and Computing Research (ICCIC), pp.1-4.
[xxiv] F. Barani, “A hybrid approach for dynamic intrusion detection in ad hoc networks using genetic algorithm and artificial immune system,” Iranian Conference on Intelligent Systems (ICIS), pp.1 6.
[xxv] Bates, Madeleine; Ellard, Dan; Peterson, Pat; Shaked, Varda. http://www.aclweb.org/anthology/H91-1040