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

NATO 2021 guidelines are a good starting point but agency, funding, and sufficient infrastructure are required to guide AI towards ethical development.

Christie and Ertan 22 (Edward Hunter Christie is the owner and founder of AI Policy Consulting. He served as a NATO official from 2014 to 2020, ending his tenure in the role of Deputy Head of NATO’s Innovation Unit. Amy Ertan is a cybersecurity fellow at the Belfer Center for Science and International Affairs (Harvard Kennedy School), researcher at the NATO Cooperative Cyber Defence Centre of Excellence (CCDCOE) and doctoral candidate at the Information Security Group, Royal Holloway, University of London.) “NATO and Artificial Intelligence” 14 Jun 2022 https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4133397 // ZX

The practical adoption of AI presents certain challenges for Allied defence institutions. Christie (2021) identifies four areas to consider: iterative development, access to human capital, access to data, and engagement with civilian-oriented technology institutions through supportive innovation mechanisms. Iterative development refers to the fact that software development in general, and AI development in particular, requires highly iterative and dynamic work processes that are generally referred to as Agile development. From a project management perspective, traditional development generally follows a Waterfall model, a linear process where each step is completed before moving on to the next step. Agile development, by contrast, allows for development steps to be revisited and adjusted multiple times in a more dynamic manner, with much shorter development and testing timelines. The need to adopt Agile development approaches is well understood in key parts of the NATO Enterprise, notably Allied Command Transformation (ACT) and the NATO Communications and Information Agency (NCIA), and in many national defence institutions. What is less clear is what types of reforms need to be enacted at higher organisational levels. Access to data refers to a range of organisational and normative challenges to ensure that Allied defence institutions can leverage the data they have as effectively as possible. To address these challenges, a NATO Data Exploitation Policy was developed, based on detailed consultations with internal NATO Enterprise stakeholders and with representatives of Allied governments. The Policy is based on the recognition of the need for a comprehensive life-cycle approach to data resources (including collection, documentation, storage, exploitation, retention, and disposal) and for the further development of relevant enabling factors for each lifecycle stage. NATO’s emerging data policy efforts also include the formulation of guiding principles that partly mirror the principles of responsible use that should hold for AI, with the addition of specific considerations that relate to data ownership and management. To the extent that the NATO Enterprise carries out its own data exploitation and AI development efforts, the relevant NATO staffs will need the relevant structures, resources, and staff to do so, in a manner that mirrors what occurs in national defence institutions. Additional considerations are necessary to ensure that the NATO Enterprise can play a full role as a facilitator for common multinational and Alliance-wide efforts. A particularly salient issue in that context is the collaborative training of Machine Learning algorithms. While individual Allies may choose in many cases to remain the sole owners and custodians of the real-world military data they collect, there is a good case for seeking economies of scale and scope through collaborative data exploitation and algorithm training efforts. Intuitively, one may expect this to imply a need to pool datasets from multiple institutions under the custodianship of a single entity, for example a NATO body. This need not necessarily be the case, thanks to the existence of technical solutions that allow for the collaborative development of Machine Learning algorithms based on sharing successive trained instances of algorithms without actual data pooling. However, significant practical work will be necessary to ensure that the Alliance is able to pursue both types of effort – informally, data sharing versus model sharing – in efficient and seamless ways. Allied governments should also consider increasing public funding levels for research & development in AI and related technologies, as well as enhancing collaborative public-private financing mechanisms to support technological innovation at various stages of maturity. To support these goals, two major initiatives championed by NATO’s Innovation Unit were approved by Allies in 2021, namely the creation of a Defence Innovation Accelerator for the North Atlantic (DIANA) and the launch of a NATO Innovation Fund (NATO, 2021c). The versatility and adaptability of AI technologies pose specific challenges in terms of governance. A large volume of analysis and commentary has already been devoted to ethical considerations, leading to the adoption by many organisations of a range of principles. In the course of 2021, NATO developed and negotiated its first-ever AI Strategy. The final text as approved by Allies was formally agreed by Allied Defence Ministers in October 2021. A summary of the Strategy was released to the public, consistent with the stated aim of the Alliance “to lead by example and encourage the development and use of AI in a responsible manner” (NATO, 2021). The Strategy and its summary contain a set of Principles of Responsible Use which were closely modelled on the DoD principles and other existing national principles. The NATO Principles constitute a political commitment that applies to both the Alliance and the NATO Enterprise (the expression “Allies and NATO” is used in the text), in other words it is also a commitment by each national government regarding its national activities, regardless of whether they are intended for NATO activities or for national or other multinational activities outside of the NATO context. As in the DoD case, the NATO principles “apply across all types of AI applications” (NATO, 2021). An additional issue on the scope of application which is clearer in the NATO case is that the principles apply to “the AI applications they [Allies and NATO] develop and consider for deployment” (NATO, 2021). By implication, AI applications that are not intended for deployment could deviate from the principles. The underlying reasoning for this caveat is to allow for experimental applications or systems that entail deviations from principles that are necessary, for example for the development of countermeasures against actual or hypothetical adversarial capabilities that do not conform to the principles or for enhancing understanding of the technical characteristics necessary to ensure compliance with all principles. A process of learning-by-doing will be necessary across Allied defence institutions and relevant NATO bodies in order to operationalise NATO’s Principles of Responsible Use. Over time and based on experience in the implementation of the principles, Allies and NATO should be able to formulate with greater precision how the principles are to be applied, potentially leading to new NATO military-technical standards, to reliance on emerging civilian technical standards where that is more efficient, and to new or updated certification processes. In that context, the development of relevant review methodologies and of risk and impact assessment frameworks, and relevant updates to security certification requirements will be important (Stanley-Lockman and Christie, 2021). NATO’s cooperative structures and consultation mechanisms will provide a basis for the testing, evaluation, validation and verification (TEVV) of AI-enabled capabilities against intended use cases and the NATO Principles of Responsible Use. Ongoing applied research within the NATO Science & Technology Organisation’s networks on matters such as human-machine teaming, and machine-machine and humans ystems integration, among others, will provide essential contributions (Stanley-Lockman and Christie, 2021). Existing and emerging national and NATO test centres, most likely including those foreseen under NATO’s Defence Innovation Accelerator for the North Atlantic (DIANA), will be instrumental in the operationalisation of the Principles of Responsible Use by providing facilities, personnel, and procedures for relevant TEVV and certification activities. This would include a range of simulation and modelling activities, in addition to physical testing as appropriate. Focusing on the technical characteristics to be ensured during the design and development phases, it is important to note that AI applications pose challenges in terms of testing and validation. A simple mechanical device can be subject to relatively simple testing protocols to ensure its safety, and such testing protocols, while informed by practical uses, can be standardised to a high degree. AI applications, on the other hand, are too versatile, changeable, and context-dependent to allow for an ex-ante formulation of all possibly relevant technical tests. Instead, the way forward is to design and apply adequate “methodological standards” on the processes inherent to the design and development of AI algorithms, rather than on each trained instantiation of every AI algorithm. Allies and NATO have laid important policy foundations for the military use of AI. A particularly visible development is the adoption, by the United States in 2020 and by the entire NATO Alliance in 2021, of principles of responsible development and use. This is only the beginning of a long process of technological transformation, adoption, and adaptation. Multiple challenges have been identified that need to be addressed in order to adopt AI and leverage data as a strategic resource. Looking across the range of organisational, institutional, human resources, and investment challenges, one evident general conclusion is that Allied governments will need to mobilise new and greater financial resources and hire new staff and train existing staff if they wish to adopt and adapt as required. At the same time, relevant policy adaptation should accompany this process, on the one hand to directly support technology adoption, and on the other hand to ensure good governance and responsible use. The latter topic is generally the most analysed and discussed in policy communities, both inside and outside government. In this chapter, we wished to stress that this most popular topic has already moved beyond the general discussion phase and is now in need of technical layers of additional work to ensure that recently adopted principles are implemented in practice.

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