Perm True mitigation requires multiple sets of actors – including international government regulators
UNIDIR (United Nations Institute for Disarmament Research), 2018, “Algorithmic Bias and the Weaponization of Increasingly Autonomous Technologies – A Primer”, The United Nations Institute for Disarmament Research (UNIDIR)—an autonomous institute within the United Nations—conducts research on disarmament and security. UNIDIR is based in Geneva, Switzerland, the centre for bilateral and multilateral disarmament and non-proliferation negotiations, and home of the Conference on Disarmament. The Institute explores current issues pertaining to a variety of existing and future armaments, as well as global diplomacy and local tensions and conflicts. Working with researchers, diplomats, government officials, NGOs and other institutions since 1980, UNIDIR acts as a bridge between the research community and Governments. UNIDIR activities are funded by contributions from Governments and donor foundations, https://www.unidir.org/sites/default/files/publication/pdfs/algorithmic-bias-and-the-weaponization-of-increasingly-autonomous-technologies-en-720.pdf, - Maren Lien
Responsibility for mitigating unwanted algorithmic biases does not rest with a single actor. A first set of actors are the program developers designing and creating the system. The developer is intimately familiar with each of the algorithms running in the system. To the extent that an undesirable bias can be mitigated through changes in the underlying algorithms or development process, then developers present a natural locus of intervention. In this way, some potential problems can be avoided before the system is fully built. At the same time, not all algorithmic biases can be addressed purely in the development stage. For example, appropriate training data might not be available, and the developers might have insufficient knowledge of deployment contexts to appropriately adjust their algorithms. The second set of key actors in potential mitigation of AWS algorithm biases are the acquirers of the technology. The agency or organization responsible for the purchase of the technology can require that the system have certain features, or meet specific, pre-defined standards. Alternately, the acquirer can require that the developers provide them with precise, detailed information about the training data, intended use contexts, and so forth. In the former case, the acquirer indicates which algorithmic biases are unacceptable, and the developer must find some way of producing such a system. In the latter case, the acquirer gains the knowledge needed to adapt practices (such as rules of engagement) to minimize the harms from the algorithmic biases that remain. In either case, acquisition and procurement teams can minimize the likelihood of algorithmic “failures” or negative biases. The third set of potential actors in mitigation efforts are regulators (including international policymakers) and testers. Regulators could decide to completely ban the development or use of AWS. Alternatively, they may decide to restrict or regulate some facet of development or use. In this case, they may determine which algorithmic biases are unacceptable, and not allow deployment of systems that exhibit those biases. They could prioritize various conditions, properties, and behaviours of a weapon system, and thereby impose particular ethical, legal, or social norms that the system must follow, though the developers are left with the task of determining how to satisfy those constraints. National or international regulators also have the ability to dictate regulatory constraints and processes that can help guide developers and future testers in their search for these or similar-acting system biases. Lastly, through testing, some algorithmic biases may be identified prior to approval and deployment, allowing for system revisions prior to the negative, real-world or real-life impacts that would impair efficacy or trust in future AWS deployment. The fourth set of potential actors would be the deployers or operators of the system. These actors, whether at the strategic or tactical level, would make the final decisions about whether, when and where to use the weapon system, and so have the ability to mitigate algorithmic biases simply by not using the system. Alternately, if a system is used only in settings for which it was designed with appropriate training data (and all of the other conditions), then the system’s potentially harmful impacts will be mitigated—though not necessarily completely eliminated.