On 7 June 2022, NATO’s International Military Staff (IMS) held its fourth substantive debate, focusing on the link between gender and emerging and disruptive technologies (TE/TR) , and more specifically artificial intelligence (AI). It discussed operationalizing responsible use principles to build trust and robust interoperability, with a focus on the importance of building data trust for intelligence development artificial by integrating the gender dimension. Presentations were made by Mr. Ulf Ehlert, Head of the Strategy and Policy Section of the NATO Science and Technology Organization (STO), and Ms. Zoe Stanley-Lockman, of the Innovation Unit of the Challenges Division of NATO Emerging Security (ESC).
Emerging and disruptive technologies based on values
Mr. Ehlert led the substantive discussion on the themes of innovation, technology and values. He said that technological fields must “evolve alongside society through a process of mutual adaptation, and that we all have different roles to play in shaping this development, the outcome of which cannot be determined by advance “. On TE/TR, he noted the differences between “emerging technologies” and “disruptive technologies”. Emerging technologies are conditioned by a recent scientific discovery or a new technological development which is expected to mature over the next 20 years and whose ultimate effects on defence, security and/or institutional functions are still uncertain. The concept to be retained is therefore that of maturity. Disruptive technologies, on the other hand, correspond to a scientific discovery or a technological evolution that is expected to have a revolutionary effect on defence, security and/or institutional functions within the next 20 years. In this case, it’s more about impacts. NATO should adopt an evolutionary policy-making process that builds on current knowledge, but provides enough flexibility so that decisions made today can be adapted or corrected tomorrow. For more on this, see Ehlert’s article, Why Our Technology Choices Should Be Guided by Our Values.
Gender dimension and artificial intelligence
Gender dimension and artificial intelligence
Ms. Zoe Stanley-Lockman focused more specifically on the convergence of gender and AI. She noted that the three waves of AI have focused, respectively, on craft knowledge, statistical learning, and contextual adaptation. Moravec’s paradox shows that for AI, high-level reasoning requires very little computation but low-level skills require more resources. Ms Stanley-Lockman added: “An AI system can easily determine which national park a person is in with an image, but more difficult if there is a bird.” AI systems do not reproduce human logic and therefore we cannot consider explaining all the decisions made by an AI system. When it comes to gender intersectionality, one of the current issues is biases in design choices. It is possible to mitigate bias by asking the following questions: what problem are we trying to solve, what value is created by the AI, what data can we use, who will use the AI system. Ms. Stanley-Lockman gave some examples of gender stereotypes regarding the under-representation of images of women in online search results for common jobs and gender issues in machine translation. These stereotypes can have negative effects for NATO if they are ignored. For example, if we train AI systems on unrepresentative data, we risk developing unpredictable performance in real-world situations.
What path for NATO?
NATO’s Artificial Intelligence Strategy (2021) focuses on the development of responsible AI systems by design. Countries endorsed six responsible use principles for AI, namely legality, accountability, understandability and traceability, reliability, governability, and mitigation of bias. Design choices for AI should start before initial development to properly integrate gender. For more information, please see the NATO Artificial Intelligence Strategy Brief. NATO has established itself as a driving force for responsible innovation and therefore appears to be strategically necessary. Furthermore, compliance with these principles provides operational value. As bias mitigation is one of the six fundamental principles adopted by NATO, the Alliance intends to integrate the gender dimension into the development of AI-enabled capabilities as well as the robustness of AI systems against attacks. As TE/TR and AI will be seen as tools to measure and accelerate military decision-making, NATO will need to ensure that they do not measure biased results.