Free beta access. 5 free AI tokens each week. Optional top-ups available. See what is included
Opening Word document…

AI for the purposes of NATO means data analytics, control, and decision support systems.

Amy Ertan Maggie Gray 2021 “An Overview of NATO Member States’ Strategies and Deployment” https://ccdcoe.org/uploads/2021/12/Strategies_and_Deployment_A4.pdf

NATO’s Science and Technology Organisation (STO) defines AI relatively broadly, as ‘the ability of machines to perform tasks that typically require human intelligence’.8 The STO and the US’s Defense Advanced Research Projects Agency (DARPA) categorise AI innovation in several waves,9 including ‘first-wave’ or ‘knowledge-based’ AI and ‘second-wave’ or ‘data-based’ AI.10 → ‘First-wave’ AI (sometimes called ‘knowledge-based’ or ‘expert systems’) has existed for decades and is already integrated into a number of systems, both military and commercial. Knowledge-based AI systems rely on rules-based decision-making, facilitating automation by using expert knowledge hand-crafted by humans and a number of if-then statements to dictate their actions.11 Knowledge-based systems cannot reason about situations outside of their carefully crafted if-then knowledge, nor can they learn from their experiences, making them incredibly brittle.12 → ‘Second-wave’ or ‘data-based’ AI systems solve specific problems by using statistical models that are trained on large, sometimes pre-labelled data sets.13 Data-based AI includes machine learning (ML) and its sub-set deep learning (DL), which have seen considerable growth in recent years, in large part because of advances in deep learning neural networks. Machine learning algorithms include supervised, unsupervised, and reinforcement learning.14 Data-based AI models are entirely reliant on the data they are trained on. For example, a drone equipped with second-wave AI data analytics and surveillance technology may be able to identify a mobile intercontinental ballistic missile (ICBM) base from a photo by using a statistics-based image recognition15 model trained on pre-labelled images of mobile ICBM bases. Throughout this report the terms ‘autonomous’ and ‘AI’ will both be used. The authors acknowledge that to many familiar with the technical terminology, ‘autonomy’ and ‘artificial intelligence’ may be considered distinct terms. Agreement on what the terms ‘autonomy’ and ‘AI’ mean varies within academic literature, often in inconsistent ways. For instance, some no longer consider ‘first-wave’ systems, which do not use ML, to be AI. Scharre and Horowitz (2015) propose a basic definition of autonomy when writing about autonomous weapons. They write that ‘autonomy is the ability of a machine to perform a task without human input’, though they point out that this may or may not involve aspects of artificial-intelligence technology.16 Some also make a distinction between ‘autonomous’ systems and ‘automated’ or ‘automatic’ systems. For example, NATO’s Allied Command Transformation differentiates between ‘autonomous’ systems and ‘automated’ systems in the following way: → autonomous functioning refers to the ability of a system, platform, or software to complete a task without human intervention, using behaviours resulting from the interaction of computer programming with the external environment. Tasks or functions executed by a platform, or distributed between a platform and other parts of the system, may be performed using a variety of behaviours, which may include reasoning and problem solving, adaptation to unexpected situations, self-direction, and learning. Which functions are autonomous – and the extent to which human operators can direct, control, or cancel functions – is determined by system design trade-offs, mission complexity, external operating environment conditions, and legal or policy constraints. This can be contrasted with automated functions, which (although they require no human intervention) operate using a fixed set of inputs, rules, and outputs, the behaviour of which is deterministic and largely predictable. Automatic functions do not permit the dynamic adaptation of inputs, rules, or outputs.17 Given the frequent conflation of the terms ‘autonomy’, ‘autonomous systems’, and ‘AI’ in documents reviewed for this report, the authors’ terminology follows the tendency of a number of NATO members to define AI and autonomy as technologies capable of going further than automated behaviour. Within this report, the word ‘autonomous’ will generally be used to describe first-wave-like systems that physically operate on their own and respond to their external environments. The term ‘AI’ will be used to describe second-wave data analytics, control systems, and decision support systems. When describing specific military systems, the authors use the same terms used by the manufacturers or by third-party experts describing the system (for example, L3Harris describes its MAST-9 ASV as ‘autonomous’, so the authors describe the MAST-9 as ‘autonomous’). Within this scope, autonomous systems do not necessarily represent or contain AI-enabled or ML capabilities.

Sign in to recut, update, or save this card.