AI as a Multi-domain Enabler for the Army

Artificial Intelligence (AI) – a gamut of general purpose technologies that rely on data to enable a wide spectrum of adaptive and intelligent behaviors in machines – is back in news again. The Bletchley Park Declaration, a statement signed by 29 countries including India at the end of the two-day AI Safety Summit organised by the UK on 01-02 November 2023, focuses on the risks associated with frontier large language models (LLMs) and exhorts companies developing the models to voluntarily provide context-appropriate transparency and accountability^[1]. The statement further calls on signatory countries to adopt regulatory frameworks and collaborate internationally to ensure that AI is designed, developed and deployed in a way that is human-centric, trustworthy, responsible and safe. The summit was held at the eponymous park which witnessed, during the Second World War, the coming together of war and science in a symbiotic relationship^[2] – the first time in history that the outcome of a war hinged so much on science and technology, and vice-versa. Of course, radio sets, radars, sonars, torpedos and finally the ‘destroyer of worlds’ – the atomic bomb – all were products of this epic war and shape international relations and security to this day. Alan Turing, for some the godfather of AI, cracked Germany’s Enigma cipher on the very same grounds^[3]. Incidentally, he also devised the Turing Test^[4] – a way to distinguish between human and machine intelligence.

The Bletchley Park declaration was preceded, in this year, by the US President’s Executive Order on Safe, Secure and Trustworthy AI^[5] and by the Call to Action^[6] at the Responsible AI in the Military Domain (RE AIM 2023) summit in The Hague on February 15-16, 2023. With the likely passing of the EU’s AI Act in December 2023^[7], this year has proven momentous in terms of bringing AI front and centre in various policy making initiatives. Most policies and acts concerning AI, however, are heavily skewed towards mitigating risks engendered by intentional misuse of AI such as deepfakes and generative AI. However, the military domain is one of the areas where AI, when used responsibly, has the potential to achieve exponential results. The Indian Ministry of Defence (MoD) views the utility of AI in the five key areas^[8] of lethal autonomous weapons systems (LAWS), unmanned surveillance, simulated wargames & training, cyber & aerospace security and, intelligence and reconnaissance. As per ex Chief of the Army Staff (COAS) General MM Naravane^[9], sensor fusion, situational awareness and faster decision-making are also key utilities of AI. He, however, cautioned that the organisation’s doctrines and structures must also change accordingly to effectively leverage AI. The intent of the paper therefore is to understand AI as a technology, take a look at some examples of AI being used in ongoing conflicts, analyse projects undertaken by the Indian Army in the open domain and finally offer some broad suggestions in terms of how AI can be best optimised for future warfare.

A Brief History of AI as a Technology

Before delving into the history of AI, let us first define what AI is. As a field of study^[10], AI is a “science and engineering of making intelligent machines, especially intelligent computer programs.”. As a technology^[11], AI refers to “systems that display intelligent behaviour by analysing their environment and taking actions” And, finally from a sociological perspective^[12], AI functions like “human social actors to form social relations and construct social realities”. AI, however, has its detractors amongst thinkers too, such as being called “applied statistics”^[13] by the famed sci-fi writer Ted Chiang. Coming to the common attributes of AI, these are sensing, perceiving (either through an internal world model or tokenised representation of input data) and finally computing, to either predict or generate. AI as a discipline has had its ups and downs and the window between 1956 to 1974 is considered as the Golden Years of AI^[14]. Within this timeline, researchers focused on identifying various attributes required for a general-purpose AI system. The aim was to develop these capabilities separately and later assembling them to create a genuine intelligent system. The targeted qualities were perception, problem solving, planning, reasoning and natural language understanding. However, due to limited amount of electronic memory and processing power at the time, this approach faced serious obstacles in the form of combinatorial explosion or non-deterministic polynomial time (NP)-complete problems^[15]. In simpler terms, there were some problems that theoretically had solutions but could not be solved practically with the given resources at the time within an ordinary human life span. This led to the first AI winter which lasted till the early 1980s.

The next spate of AI research was based on expert systems which used technical knowhow of professionals to create AI systems to solve narrowly defined problems. Known as Knowledge-based AI or expert systems based AI^[16], this phase lasted till the 1980s and the failure of the Cyc project^[17] – an ambitious project which tried to encapsulate the entirety of the entire human knowledge into a single program – led to the second AI winter. The scientist who lifted the field out of this gloom was Rodney Brooks who created the next paradigm in AI known as Behavioral AI^[18]. This was based on his belief that intelligence was an emergent property that resulted from the interaction of an entity in its environment. Agent-based AI^[19] which could act on its environment by sensing it was the next step in AI. These needed three attributes: reactive (adapt behavior based on change in environment), proactive (work towards accomplishing tasks given by user) and social (communicate, cooperate, coordinate and negotiate with other agents). The main theory underpinning this choice: that of the machine deciding and choosing on behalf of the humans, was derived from the expected utility value theory^[20] of John von Neumann. This model continued into the 1990s.  The current wave of AI took off in a parallel direction and is based on the concept of deep learning (DL)^[21], which is one of the ways machine learning (ML) is undertaken in the field of AI.

The overarching concept of ML^[22] is for programs to compute a desired output from a given input without an explicit instruction set. ML requires numerous examples to be trained to recognise images, texts etc. Therefore it requires lot of data and computational power. The rise of social media giants and increasing digitisation of ordinary human activities^[23], in consonance with the progressing of Moore’s law leading to processors such as graphics processing units (GPUs) has enabled the rise of ML and DL as the main form of AI. DL which uses a structure called a neural network^[24] – a type of computer program which models what a human brain may look like and consists of adjustable weights known as neurons – has manifested in various ways such as computer vision, swarming, pattern recognition and natural language processing (NLP). All these have corresponding military uses.

Use Cases in Contemporary Wars

The use of AI in modern wars is not very widespread owing to the nascent nature of the technology and the issues of trust, bias and discrimination involved in the use – and misuse – of these models. Nothwithstanding these risks, AI has been used in the ongoing conflict between Russia and Ukraine, with both sides attempting to leverage AI in multiple ways. While the former has created home-grown products and is still struggling with fully integrating AI into its operational concepts and platforms, the latter has been more successful since it is supported by a host of Western nations and companies. In three major fields, that is, information operations, drone warfare and battlefield transparency and autonomy, Ukraine has made significant advances, which can reverberate around the world, acting as portents for the kind of warfare that states can expect to be engaged in. The Ukraine conflict has witnessed the embedding of AI as part of geo-spatial intelligence^[25], combining data from satellite (military and commercial) imagery, hyperspectral maps, open source information, human intelligence (HUMINT) and drone footage to create a real-time common operational picture (COP) for dynamic targeting and neutralising options. This is being done by using neural networks wherein they ingest a huge quantity of data and find patterns. In this they are assisted by companies such as Palantir^[26], which has used this conflict to further sharpen their product lines, creating the Artificial Intelligence Platform (AIP)^[27]. This combines geo-spatial data with LLMs to create a chat-bot based model for military users where strikes can be ordered and information retrieved based on simple written requests from the user. Similarly Ukraine has used facial recognition technology (FRT)^[28] to identify dead bodies of Russian soldiers and use them later for information warfare (IW). The Ukraine military has confirmed that Saker Scout drones are now conducting autonomous drones strikes on Russian targets^[29]. The drones can allegedly identify 64 different types of Russian targets. The drones also act as wingman for first person view (FPV) drones where the AI points out the target for the follow-on FPV drone. As per Mykhailo Fedorov, Ukraine’s Minister for Digital Transformation, around 2000 AI-enabled drones^[30] have been supplied to the Ukrainian forces under the Army of Drones initiative. Russia, on the other hand, has used low level AI in loitering munitions (LMs)^[31] to attack Ukrainian military and civilian infrastructure. Both sides have used AI to launch disinformation campaigns against each other^[32].

Use of AI in the Indian Army

The Indian Army has started exploring AI as an enabler in various domains. Publicly available information, however, is very limited in this case. As a result, the analysis here is based exclusively on four major sources: newspaper reports, official websites and resources, speeches by military authorities and selected peer-reviewed articles published in the Defence Science Journal issued by the Defence Research and Development Organisation (DRDO).

In terms of requirements of the Indian Army in future warfare, object detection, pattern recognition, NLP, swarming (including collaborative behavior, collision avoidance and path detection) and up to Level 4 autonomy^[33] for unmanned systems in certain environments are likely to be the main requirements. Object detection is a critical requirement along the Line of Control (LC) and the Line of Actual Control (LAC) owing to the nature of terrain, weather conditions and continuous operational deployments. Object detection algorithms, based both on computer vision and sensor fusion involves detection of the environment by sensors such as drone-based hyperspectral imaging (HSI), RGB (denoting the three colour wavelengths of red, green and blue that closely simulates human vision) or regular digital cameras, identifying objects resembling, for example, tanks, barracks, installations etc and then labelling them appropriately for future action. At times, when one mode of vision does not fully identify the scene, sensor fusion is required, which is the process of merging data from multiple sensors to reduce the amount of uncertainty in a particular navigation task. Similarly, in counterterrorism (CT) scenarios, AI-based platforms can be used to detect suspicious objects such as vehicle-based improvised explosive devices (VB-IEDs) or IEDs placed for patrols. A DRDO paper^[34] looks at the use of sensor fusing of two streams of data: one each from HSI and Light Detection and Ranging (LiDAR) sensors to detect and distinguish camouflaged objects from their background. Another paper^[35] uses neural networks to analyse a particular terrain from the point of view of movement of mechanised forces. This is done using a convolutional neural network (CNN) which is best suited for image and video analysis. The “friction” is obtained in advance using a combination of images from drones and satellites and then feeding them to a CNN so that avenues of approach can easily be deciphered. These papers point to a thinking within the broader defence establishment that the use of AI can enable faster decision making by giving commanders at local levels tools required for practical analyses. In fact, as per news reports, the Indian Army has already deployed close to 140 AI-based surveillance systems^[36] which include high-resolution cameras, sensors, unmanned aerial vehicles (UAVs) and radars. Input streams from these are fed to algorithms to create a robust common operational picture (COP).

Pattern recognition is another area which is critical for the Indian Army. Apart from the threat of conventional warfare, India also faces a disinformation and cyber attack onslaught almost on a daily basis. There is a growing need for AI-based cyber tools for automatic threat detection, defence and offence^[37]. Similarly, sifting through social media content for sentiment analysis^[38], perceiving and countering disinformation attacks requires an AI-enabled effort for tracking patterns. On the conventional front, the Indian Army’s Northern Command is in the process of operationalising the Situational Awareness Module for the Army or SAMA^[39] which integrates inputs from the Artillery Combat Command and Control System (ACCCS), battlefield surveillance system (BSS), e-sitreps and other systems on an enterprise platform. This is the first step towards creating a data repository. The next logical step should be to create context-specific input data streams in the future for use in a hybrid LLM based chatbot mode, similar to Palantir’s AIP.

NLP is an ongoing need for the Indian Army due to the requirement of real-time translation and interpretation of languages like Mandarin along the LAC^[40] and localised dialects of Punjabi, Urdu, Dari and other languages^[41] used by terror groups along India’s western borders. Similarly swarming, with its varied use in air defence, saturation attack, autonomous targeting and communication relay, among others, are in the process of being deployed. The success of AI-enabled drones and loitering munitions (LMs) in the conflicts between Armenia and Azerbaijan and the ongoing one between Russia and Ukraine have brought home the potency and scalability of these weapons. In fact, the Indian Army is arguably one of the world’s first armies^[42] to have inducted an operationalised drone swarm, when it contracted with the Bengaluru-based startup NewSpace Research on the sidelines of Aero India in February 2023.

The induction of AI-based products within the Indian Army has received a major fillip from the policies and directions of the Indian Ministry of Defence (MoD) as well as the Service Headquarters. The Army Design Bureau (ADB) has nominated the Mhow-based Military College of Telecommunication and Engineering (MCTE) as the Indian Army’s AI Centre of Excellence (CoE)^[43]. The CoE has undertaken multiple in-house projects and has reportedly driven^[44] the process of deploying the 140 AI-based surveillance platforms. The DRDO has also two dedicated labs for AI-based research on defence platforms: the Centre for AI and Robotics (CAIR)^[45] and the DRDO Young Scientist Laboratory (DYSL)^[46], both based in Bengaluru. The three services have also handheld various defence startups in the field of AI through the innovations for defence excellence (iDEX) challenges – an exercise which resulted in the showcasing of 75 AI based products at the AIDef symposium^[47], organised by the Department of Defence Production (DDP) on 11 July 2022.

AI as a general purpose technology is both an actor and an enabler. Now, with the transformer-based LLMs being released, and especially their multimodal versions, many are talking about the capabilities of these platforms and the products that can be built over these. AI also enables the integration, synthesis and fusion of multiple streams of data, provided (especially in the context of military usage) that they are standardised, cleaned and labelled. Usage of AI-based systems in places involving interaction of these systems with humans may also create issues of bias and discrimination. These issues have to be kept in mind. The use of AI will only increase in the future and with adversaries making long strides in this field, India also needs to look at the implications of AI from a national security and defence point of view equally seriously.