SIPRI Report – April 2020

This volume assembles the perspectives of eight experts on South Asia on why and how machine learning and autonomy may become the focus of an arms race among nuclear-armed states. It further explores how the adoption of these technologies may have an impact on their calculation of strategic stability and nuclear risk at the regional and transregional levels.

Executive Summary 

The ongoing renaissance of artificial intelligence (AI) is reshaping the world. Just like many other developing countries, India and Pakistan—the two nuclear-armed states of South Asia—are exploring the subsequent opportunities for economic and social change. Their political leaders seem to prioritize civilian applications of AI over the military, and public attention reflects the political priorities. National efforts to militarize AI do not receive the same public coverage as civilian AI developments.

Meanwhile, according to the available open-source information, India and Pakistan are increasingly interested in the potential benefits of AI for defence and security. This might be one of the reasons why an expert debate on the opportunities and risks posed by the AI renaissance in the military realm has started in recent years. However, the debate suffers from large gaps, particularly in the emerging discussion on the potential impact of AI on strategic stability and nuclear risk in South Asia. This issue has been underexplored by scholars studying South Asia from both inside and outside the region.

This edited volume—which follows earlier volumes on Euro-Atlantic and East Asian perspectives—tries to fill the gaps in the scholarly debate on this important topic and to facilitate further regional debate. It is based on a workshop held in Colombo in February 2019. The eight expert contributors—from South Asia and around the world—reflect the variety of issues, approaches and views.

It is clear from a comparative study of the state of adoption of AI in South Asia that India and Pakistan are playing catch-up in the world competition on military AI. Compared to the United States, China and Russia, India’s advances are modest, while Pakistan’s are even less visible. One of the reasons seems to be under-resourcing and inefficiencies in defence research and state industries. These prohibit the development and adoption of emerging technologies within a reasonable time frame.

However, according to contributors from India and Pakistan, both countries are well aware of the strategic significance of AI. They see AI as one of many enablers of the mutual strategic balance. India must also take into consideration the role of AI in the military build-up of China, one of its long-term security concerns.

In assessing the strategic significance of AI, the expert contributors—regardless of their origin—agree that AI is a double-edged sword. On the one hand, AI could enhance nuclear command and control, early warning, intelligence surveillance and reconnaissance (ISR), and the physical security of nuclear capabilities, among other areas. In this way it would improve states’ sense of security. On the other hand, the same advances could cast doubt on the survivability of their respective second-strike capabilities. This doubt would stimulate more aggressive nuclear postures that could increase nuclear risk.

There are several scenarios in which AI-enabled weapons could be involved in escalatory dynamics in South Asia. Given that there have been few military applications of AI in either India or Pakistan, the contributors do not endorse the view that the use of AI systems could cause a nuclear war between India and Pakistan or between India and China—at least for the foreseeable future. However, most agree that the introduction of AI into the nuclear capabilities and postures of India and Pakistan could affect strategic stability in South Asia. For this reason, the majority of contributors support the idea that the states of South Asia should take steps now to reduce the nuclear risk.

The question of how to design those steps is more divisive. For some, the solution lies in the development of a legally binding international agreement that would limit the military use of AI. Others argue that elaborating regional transparency and confidence-building measures would be a more feasible option. A starting point in their view would be to establish a regional dialogue on nuclear doctrines and capabilities that would include a discussion on military AI. Given the success of several track-2 dialogues on security between China, India and Pakistan, such an initiative seems to be relatively realistic.