2022
MNDI
Affirmative
TEVV AFF – MNDI HPA › Solvency › AT AIA CP
Dufour 2018 - Colonel in the Canadian Army, currently working with NATO [Martin, NDC Policy Brief No. 6 December “Will artificial intelligence challenge NATO interoperability?” https://www.ndc.nato.int/news/news.php?icode=1239 Acc. 4/21/22 TA]
The 2017 report Future War NATO argues that the “technologically-driven US military strategy is advancing so fast compared to European allies that, sooner rather than later, all-important NATO military interoperability might well become a thing of the past”.11 This echoed a 2016 report from the Armament Industry European Research Group, which concluded that “a further boost in US defence technology could promote a wider US-Europe gap and the emergence of a two-tier alliance”.12 It was furthermore observed that the USD3.6 billion invested in Third Offset Strategy technologies in 2017, while representing only 5 percent of the overall US military research and development budget corresponded “to more than 40 percent of the overall EU-European R&D budgets”.13 Beyond interoperability, a profound technological mismatch between allies could also erode the political and military cohesion of the Alliance by creating resentment between countries that bear a greater share of the burden as a result, and those that do not.
2022
MNDI
Affirmative
TEVV AFF – MNDI HPA › Solvency › AT AIA CP
Horowitz and Scharre 2021 - Director of the Emerging Capabilities Policy Office in the Office of the Under Secretary of Defense for Policy and Vice President and Director of Studies at CNAS [Michael, Paul, January 12 2021, “AI and International Stability: Risks and Confidence- Building Measures”, , Acc 6/18/22. M.A.]
One challenge to establishing rules of the road for autonomous systems’ behavior would be if there were incentives to defect from the rules. For example, in World War I, technological developments enabled submarines, which were highly effective in attacking ships but unable to feasibly comply with existing prize law without putting themselves at risk of attack by surfacing. Despite attempts in the early 20th century to regulate submarines, the incentives for defecting from the existing rules were too great (and the rules failed to adapt), and the result was unrestricted submarine warfare. Another challenge to a potential autonomous incidents agreement is fully exploring the incentives for trustworthiness, both in the signals that countries send about the behavior of their autonomous systems and adversaries’ responses. Some declaratory policies would not be credible, such as the claim to have created a “dead hand” system such that if a country engaged in a particular type of action, an autonomous system would start a war and there would be nothing a leader could do to stop it.
2022
MNDI
Affirmative
TEVV AFF – MNDI HPA › Solvency › AT Military Readiness DA
Horowitz and Kahn, 2021 – Senior and Research Fellows at the Council on Foreign Relations [Michael and Lauren, The Washington Quarterly 3-19-2021 “Leading in Artificial Intelligence through Confidence Building Measures” TM]
One might argue that US leadership in multilateral AI standards could create risks for the United States if it involves commitments that prevent the US deployment and use of militarily important AI-enabled systems. These risks, however, are minimal. The standards the United States would promote involve commitments to international humanitarian law and responsible behavior that the United States already follows when it comes to the development and use of military systems. Thus, it would not require changes in US behavior that might slow down responsible military AI adoption.
2022
MNDI
Affirmative
TEVV AFF – MNDI HPA › Solvency › AT Military Readiness DA
Amoroso and Tamburrini, 2021 - Prof of International Law at the University of Cagliari and Prof of Philosophy of Science and Technology at the University of Naples Federico [Daniele and Guglielmo, Feb Italian Journal of International Affairs “In Search of the ‘Human Element’: International Debates on Regulating Autonomous Weapons Systems” TM]
While these arguments, involving human duties and fundamental rights, are deontological in character, others are consequentialist since they focus on expected consequences of AWS use. The main normative argument for autonomy in weapons systems is indeed consequentialist: should AWS one day ensure better ethical performances than human soldiers in terms of reduced IHL breaches in the battlefield, then permitting their deployment ought to be preferred to banning it (US 2018a; Russian Federation 2019, para. 2). Yet, this is also contested in the light of a broader consequentialist appraisal, which considers expected repercussions outside the battlefield. The spread of AWS is likely to have negative consequences for international security and peace that outweigh any military or humanitarian advantages one can envisage on the battlefield (Bode and Huelss 2018). Destabilisation risks would include a new arms race among state actors; fewer disincentives to start wars on account of reduced numbers of soldiers involved; cyber-vulnerabilities leading to unintended conflicts; and an accelerated pace of war beyond human reactive abilities (Tamburrini 2016, 127-8).
2022
MNDI
Affirmative
TEVV AFF – MNDI HPA › Solvency › AT Military Readiness DA
Trabucco and Stanley-Lockman, 2022 – prof of Political Science, University of Copenhagen and prof of Defense and Strategic Studies, Nanyang Technological University [Lena and Zoe, The Oxford Handbook of AI Governance, March, “NATO’s Role in Responsible AI Governance in Military Affairs” https://www.oxfordhandbooks.com/view/10.1093/oxfordhb/9780197579329.001.0001/oxfordhb-9780197579329-e-69 Acc 4/16/22 TA]
In any discussion of AI as an emerging military technology, it is necessary to strike a balance between acknowledging the transformative potential of AI in the security environment, while simultaneously recognizing the “hype” that may, thus far, be unfounded. But some conclusions are clear. The risks and opportunities of military AI can pose significant challenges for future military operations, and this necessarily means there are many stakeholders with a vested interest in developing, promoting, and implementing responsible military AI. As multinational coalitions and military operations are a foundational security policy for much of the world, this means NATO is also a stakeholder with a vital interest in promoting safe and secure technology among its partners, both traditional and non-traditional. As the international security environment continues to shift, there is space for NATO to pursue its agenda to maintain technological superiority not just to protect and defend its way of life, but also to build on its pillars of AI governance to steward military innovation on a responsible trajectory.
2022
MNDI
Affirmative
TEVV AFF – MNDI HPA › Solvency › AT Military Readiness DA
Christie, 2020 - former Deputy Head of Innovation in NATO’s International Staff [Edward, Nov 24 “Artificial Intelligence at NATO: dynamic adoption, responsible use” nato.int/docu/review/articles/2020/11/24/artificial-intelligence-at-nato-dynamic-adoption-responsible-use/index.html, BK]
The Alliance’s success with AI will also depend on new and well-designed principles and practices relating to good governance and responsible use. Certain Allied governments have already made certain public commitments in the area of responsible use, addressing concepts such as lawfulness, responsibility, reliability, and governability, among others. In parallel, Allies have taken part in the Group of Governmental Experts on Lethal Autonomous Weapon Systems under the auspices of the United Nations, leading to the formulation of 11 guiding principles. Importantly, there is a good case for viewing work on adopting AI and work on principles of responsible use as complementary and synergistic. In effect, there are certain essential principles or goals that will underpin and facilitate both engineering good practice, as well as responsible state behaviour. Certain national principles imply a need for specific design requirements. For example, a principle of governability may be linked to technical abilities to detect and avoid unintended consequences, and to disengage or deactivate in case of unintended behaviour. The technical characteristics required to ensure that these and other objectives are met will necessarily be part of the design and testing phases of relevant systems. In turn, the relevant engineering work will be an opportunity to refine understanding, leading to more granular and more mature principles. Further work in the area of Testing, Evaluating, Verifying and Validating (TEVV) will be essential, as will support from relevant Modelling and Simulation efforts. NATO’s well-established strengths in the area of standardization will help frame these lines of effort, while also ensuring interoperability between Allied forces.
2022
MNDI
Affirmative
TEVV AFF – MNDI HPA › Solvency › AT Military Readiness DA
Horowitz and Scharre 2021 - Director of the Emerging Capabilities Policy Office in the Office of the Under Secretary of Defense for Policy and Vice President and Director of Studies at CNAS [Michael, Paul, January 12 2021, “AI and International Stability: Risks and Confidence- Building Measures”, , Acc 6/18/22. M.A.]
The downsides of publicly signaling the prioritization of AI T&E are relatively limited. A critic might argue that, to the extent that accidents are a necessary part of the innovation and capabilities development process, an overemphasis on T&E might discourage experimentation. However, promoting experimentation and innovation does not have to come at the expense of building robust and assured systems, especially since it is through experimentation and testing that accident risks are likely to be revealed, leading to the deployment of more capable systems. Ensuring that AI systems function as intended is part of fielding effective military capabilities, and effective T&E processes are aligned with the goal of fielding superior military capabilities. Rigorous T&E processes would, by definition, add time to the development process in order to ensure that systems are robust and secure before deployment, but the result would be more effective systems once deployed. In peacetime, taking additional measures to ensure that military systems will perform properly in wartime has little downside, so long as accident risk does not become a bureaucratic excuse for inaction
2022
MNDI
Affirmative
TEVV AFF – MNDI HPA › Solvency › AT Military Readiness DA
Horowitz and Kahn, 2021 – Senior and Research Fellows at the Council on Foreign Relations [Michael and Lauren, The Washington Quarterly 3-19-2021 “Leading in Artificial Intelligence through Confidence Building Measures” TM]
Leadership in standard-setting and confidence-building measures will also enhance US military capability, rather than constraining the US military, assisting the United States in strategic competition. Unsafe AI systems do not just risk unintentional conflict and inadvertent escalation—they are less likely to be effective systems. Committing to lead the world in AI safety could create a ripple effect in the US defense enterprise and the private sector about putting a premium on safe and ethical AI, in turn making it more likely that military (and civilian) uses of algorithms are reliable, improving their utility for the military. Such signaling will also likely improve the reputation of the US military with Silicon Valley and AI/ML researchers, whose concerns about military uses of AI have loomed large since a protest halted Google’s renewal of its Project Maven contract in 2018. While concerns about Silicon Valley’s opposition to the Department of Defense are overstated according to some survey research,37 other surveys of AI/ML professionals show that increasing the emphasis on safety is a high priority.38 A public commitment to safety will thus help the Department of Defense improve at attracting top STEM talent, including AI/ML talent, increasing its ability to keep the American military ahead.
2022
MNDI
Affirmative
TEVV AFF – MNDI HPA › Solvency › Solvency – Run Time Monitoring
Haugh, 2018 - project leader for the Institute for Defense Analysis [Brian A., September “The Status of Test, Evaluation, Verification, and Validation (TEV&V) of Autonomous Systems,” , 6/23/22 MD]
D. Run-Time Monitoring Given the difficulty of assuring that a system will not exhibit specific undesired behaviors, a natural thought is to instead monitor the system during operations and intervene when bad behavior is imminent. This approach is already common in engineering practice for safety-critical systems. It is mentioned explicitly in the recent US Unmanned Systems Integrated Roadmap 2017-2042: For the most demanding adaptive and non-deterministic systems, a new approach to traditional TEVV will be needed. For these types of highly complex autonomous systems, an alternate method leveraging a run-time architecture that can constrain the system to a set of allowable, predictable, and recoverable behaviors should be integrated early into the development process. Emergent behaviors from large-scale deployment of interacting autonomous systems poses a difficult challenge. The analysis and test burden would thereby, be shifted to a simpler, more deterministic run-time assurance mechanism. The effort for new approaches to TEVV endeavors to provide a structured argument, supported by evidence, justifying that a 3-6 system is acceptably safe and secure not only through offline tests, but also through reliance on real-time monitoring, prediction, and fail-safe recovery.1 Although this mechanism might indeed be simpler than avoiding unpredictable behaviors in the first place, it is not without its own challenges. In general, any behavior whose dependability cannot be adequately assured through system design and training would need to be monitored, with a robust intervention standing by. This means not only intervening when the system is about to execute some undesired physical action (e.g., one that might risk harm to the system or to humans), but also intervening in any case where the system is making a bad decision or misinterpreting its environment. Detecting such cases and handling them gracefully will not always be easy. Research is required into architectures to support this concept, instrumentation needs and control algorithms to predict and avoid specific failure modes, systematic identification of conditions to be monitored for, robustness against attacks designed to invoke fail-safe behaviors, and so forth. It goes without saying that the fail-safe systems would themselves need to be verified and validated as well.
2022
MNDI
Affirmative
TEVV AFF – MNDI HPA › Solvency › Solvency – Attached Funding
Tarraf et. al 2019 - Senior Information Scientist at the RAND Corporation [Danielle with William Shelton, Edward Parker, Brien Alkire, Diana Gehlhaus, Justin Grana, Alexis Levedahl, Jasmin Léveillé, Jared Mondschein, James Ryseff, “The Department of Defense Posture for Artificial Intelligence: Assessment and Recommendations”, LMSi]
Recommendation T-5: All funded AI efforts should include a budget for AI VVT&E. This recommendation, in support of Recommendation S-2, is a forcing function that is relatively simple to implement and that can help ensure that the consideration of VVT&E is baked into the R&D of AI techniques and the design of AI solutions rather than considered as an afterthought further down the line. Although VVT&E during early R&D phases should be commonplace, we make this recommendation to explicitly reinforce its critical importance, to highlight the present lack of foundations for VVT&E in AI, and the importance of developing that science Our next recommendation is a tactical one, unrelated to VVT&E, but speaks to the need to bring developers of AI technologies together with users and operators to enable success in technology development.
2022
MNDI
Affirmative
TEVV AFF – MNDI HPA › Solvency › Solvency – Adversarial Testing
Haugh, 2018 - project leader for the Institute for Defense Analysis [Brian A., September “The Status of Test, Evaluation, Verification, and Validation (TEV&V) of Autonomous Systems,” , 6/23/22 MD]
C. Adversarial Testing As noted in Challenge 6, many defense applications will require very low probabilities of high-cost events occurring. For systems with extremely large and highly nonlinear state spaces, it will not be possible to provide that assurance statistically using traditional Design of Experiments techniques. Adversarial testing, especially when combined with machine learning and automated test design methods, provides a potentially more efficient means to identify and eliminate potential failure modes. In this approach, analytical techniques are used to identify test scenarios in which the system is most likely 3-5 to perform unacceptably and focuses testing in those portions of the state space to maximize diagnostic information and inform robust design. As a specific example, one could apply reinforcement learning to the selection of environmental factors as input to the test cases in a modeling and simulation driven test bed that included the AI software running the autonomous systems. The environment could be treated as a “thinking adversary” in an asymmetric game. The environment would learn optimal strategies to defeat the autonomous system, even as the autonomous system learned improved strategies to counter the environment. This would both maximize the chance of finding key vulnerabilities and weaknesses and help discover ways to mitigate or avoid those vulnerabilities. The vulnerabilities need not be isolated within the autonomous system itself—they might also arise from the proposed CONOPS, especially in the case of significant HMT. They could also be associated with bias or incompleteness in the training data used in supervised learning to develop the autonomous capabilities. Cognitive instrumentation would be needed to distinguish inadequate algorithms or models from inadequate CONOPS or flawed training. Finally, we note that this approach requires domain expertise as well. The operating environment and mission must be well enough understood so that the apparent “weaknesses” are not associated with impossible circumstances. If the weakness occurs only during heavy rainfall at temperatures below –40° C, it will be a weakness we can live with. More subtly, the domain expertise will need to be sufficient to estimate probability of occurrence of the conditions that exposed the weakness. As mentioned before, utilization in the defense sector will require attention to the likelihood of seriously adverse outcomes.
2022
MNDI
Affirmative
TEVV AFF – MNDI HPA › Solvency › Solvency – Transparency
Horowitz and Scharre 2021 - Director of the Emerging Capabilities Policy Office in the Office of the Under Secretary of Defense for Policy and Vice President and Director of Studies at CNAS [Michael, Paul, January 12 2021, “AI and International Stability: Risks and Confidence- Building Measures”, , Acc 6/18/22. M.A.]
Increased Transparency about T&E Processes. A related unilateral or multilateral CBM could involve countries publicly releasing details about the T&E processes used for military applications of AI without revealing details about specific technical capabilities. This is similar to existing U.S. policy regarding legal weapons reviews. Currently, the U.S. military promotes norms in favor of stringent legal weapons reviews but does not share the actual reviews of specific weapons. Since this CBM would build on existing norms that the United States already promotes, transparency about T&E processes for military AI systems might be more likely to receive American support than more intrusive measures. Moreover, increasing knowledge about T&E processes might bring other countries that want to learn from the American military on board. The potential drawbacks of transparency surrounding T&E processes stem from what happens if the CBM succeeds. If successful, all countries, including potential adversaries, would have greater knowledge of how to design effective T&E processes for their military AI applications. This could improve their ability to field more effective military AI systems. This downside may be somewhat mitigated if a country only shares high-level information about its T&E bureaucratic processes and refrains from sharing technical information that could actually help an adversary execute more effective T&E. Nevertheless, an overarching concern with any T&E-related CBM that aims to reduce the risk to international stability from states building unsafe AI systems is that actually succeeding in improving other states’ T&E could also lead to adversaries deploying more effective AI systems. Whether an adversary’s improved AI capabilities or the prospect of an adversary deploying unsafe military AI systems is more of a danger to a country’s security would need to be considered. International Military AI T&E Standards. Another CBM regarding AI safety could entail establishing and promoting specific international standards for what constitutes effective T&E practices for military AI applications. Such an effort could build on private-sector and public-private standard-setting actions for non-military uses of AI. While not enforceable or verifiable, promoting common standards for AI T&E could be a useful focal point for like-minded states to promote responsible norms concerning the safe deployment of military AI systems. The downsides of promoting common T&E standards are similar to the potential downsides of a public emphasis on AI safety. These kinds of CBMs are early building blocks: While the gains are likely to be relatively limited, the downsides are limited as well, because they do not expose key information or require national commitments that limit capabilities. As with increasing transparency about T&E processes, the most significant downside to effective T&E standards would be that, if successful, this CBM could increase the reliability of military AI systems by adversary states. The relative balance of danger between more reliable, and therefore more effective, adversary AI systems versus unreliable and more accident-prone AI systems would again need to be carefully weighed.
2022
MNDI
Affirmative
TEVV AFF – MNDI HPA › Solvency › AT Militaries are Conservative
Scharre 2021 – Director of the Technology and National Security Program at the Center for a New American Security [Paul, 6/28/21,https://tnsr.org/2021/06/debunking-the-ai-arms-race-theory/, “Debunking the AI Arms Race Theory”, 6/18/22, LND]
That militaries may incur risks from moving too quickly in adopting new technology is counter to the common caricature of military culture as conservative, hidebound, and resistant to innovation. While this caricature is not entirely fair — militaries do innovate even in peacetime49 — the lack of direct, observational feedback on performance in a realistic competitive environment, akin to marketplace dynamics for commercial companies, can mean that militaries are often slow to adapt to changing circumstances. A variety of factors can affect military adoption of new technologies,50 and adoption rates can vary considerably depending on the technology, state, and military community. Across a range of contemporary technologies militaries lag the private sector. For example, modern-day militaries are behind the private sector in adopting information technology, human performance optimizing technologies, and personnel best practices. With an increasing amount of technological innovation occurring outside of the defense sector, this lag is likely to continue.51 Yet, the key risk factor for military AI systems is not the timing of when militaries begin the process of adoption, but the taking of shortcuts in safety to accelerate fielding new AI capabilities. Adopting technology is a multi-stage process, involving research and development, experimentation, prototyping, technology maturation, production, testing, and fielding. It is possible for militaries to move slowly in one stage and quickly (or via shortcuts) in others. While there are many areas in which the U.S. military’s adoption of AI, autonomy, robotics, and uninhabited vehicles is moving slowly due to a variety of bureaucratic obstacles, it is also possible that the United States could rush parts of the adoption process and end up with immature technology in production or even in the field. This mixed dynamic, of moving slowly in some aspects of technology development and taking shortcuts in others, has been present in other defense programs. The F-35 fighter jet went into production before the first test flight, a decision that the Defense Department’s top acquisition official, Frank Kendall, later characterized as “acquisition malpractice.”52 Yet, the whole acquisition program took 25 years from its initial conception to its first operational deployment.53 The F-35 is still not in full rate production, 28 years after it was initially conceived of.54 The F-35 program moved too quickly in some areas, introducing unnecessary risk, even while it was overall encumbered by the laborious pace typical of major defense acquisition programs. Militaries’ sluggish bureaucracy is, therefore, no defense against shoddy testing and premature fielding.
2022
MNDI
Affirmative
TEVV AFF – MNDI HPA › Solvency › AT Private Circumvention
Flournoy, Haines et al 2020 - former Under Secretary of Defense for Policy and Director of National Intelligence [Michèle and Avril, Gabrielle Cheftiz Special Assistant to the Under Secretary of Defense for Policy October, “Building Trust through Testing Adapting DOD’s Test & Evaluation, Validation & Verification (TEVV) Enterprise for Machine Learning Systems, including Deep Learning Systems” https://cset. georgetown.edu/wp-content/uploads/Building-Trust-Through-Testing.pdf Acc 6/23/22 JZ]
4. Translate the testing framework into testable, verifiable requirements to be used by the private sector and build an integrated team to leverage this approach. DoD should establish a process for translating the testing framework into testable and verifiable requirements. DoD requirements would help standardize processes for industry contractors who develop AI for DoD and promote a faster and cheaper TEVV process by enabling the private sector to do some testing throughout development. The development of such requirements and standard processes would allow DoD to leverage the talent and expertise in the private sector, while maintaining DoD’s safety and risk standards and employing DoD’s operational knowledge and adversarial testing capabilities. Such requirements could additionally serve as the basis for standards that are promoted by the U.S. Government across the private sector and internationally, as discussed in a later recommendation. To realize this approach, DoD should build integrated, multi-disciplinary teams that reflect the entire development, testing, and sustainment life cycle. One model is the JAIC’s project manager model, which incorporates experts on product, policy, legal, test and evaluation, and requirements into one team. A similar approach is the Navy’s AI "DevRon” concept, a single entity accountable start to finish for the life cycle of capability development.29 Scaling this approach would help ensure requirements take into account the unique challenges of ML/DL TEVV. Requirements should also be written to advance the integration of ethics into the design and testing process. One case study for how this is already being done is DARPA’s Urban Reconnaissance through Supervised Autonomy (URSA) program, which is developing AI-infused drones that can help prevent friendly fire and civilian casualties in urban battles.30 This program brought in ethicists to anticipate challenges before development even began, with the aim of integrating ethics into the development loop. This URSA program offers useful lessons on how to conduct ethically aligned design of military systems that enables testing for robustness, safety, and security and ensures these systems are reliable and governable. Developing requirements and standards for ML/DL will help strengthen the link between DoD and industry, ensure a hybrid approach that leverages both private and public sector resources, and inject DoD’s safety and ethical requirements into industry development practices.
2022
MNDI
Affirmative
TEVV AFF – MNDI HPA › Solvency › AT Private Circumvention
Flournoy, Haines et al 2020 - former Under Secretary of Defense for Policy and Director of National Intelligence [Michèle and Avril, Gabrielle Cheftiz Special Assistant to the Under Secretary of Defense for Policy October, “Building Trust through Testing Adapting DOD’s Test & Evaluation, Validation & Verification (TEVV) Enterprise for Machine Learning Systems, including Deep Learning Systems” https://cset. georgetown.edu/wp-content/uploads/Building-Trust-Through-Testing.pdf Acc 6/23/22 JZ]
As a general matter, investments in the science of ML/DL TEVV are a critical prerequisite to developing the most effective and efficient standards, tools, and methodologies needed to assure system performance. As these technologies and their applications evolve, new areas of research will undoubtedly arise that warrant investment. 3. Develop a tailored, risk-based framework for ML/DL testing and safety. The AI/ML TEVV CFT should lead on the development of a framework that establishes architecture and testing standards for TEVV. DoD cannot have a one-sizefits- all approach; it needs a flexible testing framework that is mission and use-case dependent. A DoD-wide testing framework for AI/ML will help shorten the testing cycle and make test results interpretable and comparable across the Department. This framework should also incorporate DoD’s legal and ethical requirements, serving as implementation guidance for the AI ethics principles and 3000.09. The DIB AI ethics principles call on the JAIC to create a taxonomy of DoD use cases of AI, based on their ethical, safety, and legal risk considerations. The CFT should leverage this taxonomy and develop corresponding testing criteria and safety standards. Testing standards, for levels of interpretability or assurance, for example, should be determined based on several dimensions of risk and performance, including: the likelihood of error detection, the consequence of the error given the complexity of the operating context, the potential for unintended escalation, the size of the attack surface, system performance relative to that of a human operator, impact on human decision-making, and the risk associated with not adopting a system (e.g., the risk may be a 10 percent error for DL, but a 30 percent error without the system). For example, AI for business process automation would likely score low on all of these risk criteria, while AI for critical network cybersecurity would likely score high on all and, therefore, necessitate stricter and more expansive TEVV requirements. For higher-risk applications, DoD may need to require systems to be designed with fail-safe systems or operated only as part of a human-machine team to help mitigate risks and govern system performance. Researchers are currently developing methods for monitoring system performance and constraining a system to a set of allowable, predictable behaviors and mitigate the risk of failure and unintended escalation. A DoD-wide AI TEVV framework will help decision-makers manage the tradeoff between the risks of failure and the value of deployment, while advancing the development of clear and consistent requirements for system design and metrics for performance evaluation.
2022
MNDI
Affirmative
TEVV AFF – MNDI HPA › Solvency › AT No Definition of AI
Artificial Intelligence/Machine Learning Risk & Security Working Group, 2020 [Wharton, ‘Artificial Intelligence Risk & Governance’, ]
1.3 Definitions & Assumptions We use several terms throughout this document specific to AI/ML, some of which are subject to vigorous discussions and debates in the research community. Because this document attempts to form a starting point for broader AI/ML governance and risk management efforts, we purposely leverage and encourage readers to refer to various papers for the definition of AI. As such, we note that specific definitions should be (and generally are) tailored to each organization depending on the scope, risk appetite, internal structure, culture, and implementation details of AI/ML efforts. While there is no universally accepted definition of AI, it is generally understood to refer to “a branch of computer science dealing with the simulation of intelligent behavior in computers, or the capability of a machine to imitate intelligent human behavior.” Generally, machine learning is referred to as, “a field of computer science that uses algorithms to process large amounts of data and learn from it.”[2] The term AI is broadly used and typically includes aspects of machine learning and natural language processing. For purposes of this paper, the focus is largely on the use of and potential risks related to machine learning, though the overarching discussion applies more broadly to the abovementioned areas.
2022
MNDI
Affirmative
TEVV AFF – MNDI HPA › Solvency › AT No Definition of AI
Amoroso and Tamburrini, 2021 - Prof of International Law at the University of Cagliari and Prof of Philosophy of Science and Technology at the University of Naples Federico [Daniele and Guglielmo, Feb Italian Journal of International Affairs “In Search of the ‘Human Element’: International Debates on Regulating Autonomous Weapons Systems” TM]
Despite heated diplomatic debates, agreement is still lacking among states about what an AWS is. Nevertheless, discussion about this fundamental issue is not fragmented into a myriad of competing notions of ‘autonomy’. Rather, it is polarised around two basic construals, epitomised by those advanced by the UK Ministry of Defence (MoD) and the US Department of Defense (DoD). The UK MoD sets a very demanding requirement for a weapons system to be autonomous. Indeed, the MoD’s Joint Doctrine Publications devoted to Unmanned Aircraft Systems describe AWS as systems “capable of understanding higher level intent and direction”, thereby being able “to take appropriate action to bring about the desired state” (UK MoD 2011, 14; 2017, 13). This is a tall order. Neither existing nor foreseeable weapons systems genuinely comprehend the meaning of higher-level goals and intentions. And there are presently no educated guesses as to the prospects for their development. Thus, weapons systems that are genuinely autonomous according to the UK MoD are projected into an undetermined technological future. If one endorses this approach, then ethical and legal discussions about AWS are simply premature in that they concern neither present nor foreseeable technological developments. This conclusion, however, conflicts with the intuitions underlying current academic and diplomatic debates, according to which normative analysis and deliberation on AWS are an urgent issue (Amoroso and Tamburrini 2017, 3-4). Remarkably absent in the US DoD Directive on ‘Autonomy in Weapons System’ is any attempt to account for autonomy in terms of high-level intent comprehension. Instead, this Directive introduces a functional, task-oriented condition for weapons systems to be considered autonomous: a weapon system must be able, after activation, to “select and engage targets without further intervention by a human operator” (US DoD 2012). It is of the utmost relevance for the international debate that the US DoD requirement on autonomy was embraced by the International Committee of the Red Cross (ICRC 2016, 1) and the Campaign (Campaign 2013, 1). Therefore, the chief stakeholders – the technologically most advanced military power, the main guardian of international humanitarian law (IHL) and the coalition of NGOs advocating a ban on AWS – converged on this formulation of necessary properties of autonomy.2 Such an international convergence is a good reason to endorse the DoD’s functional condition of autonomy as a starting point for AWS legal and ethical debates. There is, however, another reason for preferring the DoD over the UK MoD’s approach: it supports the intuition that normative analysis and deliberation about AWS is an urgent and impending issue since a variety of existing and technologically foreseeable systems satisfy the DoD functional condition.
2022
MNDI
Affirmative
TEVV AFF – MNDI HPA › Solvency › AT No Definition of AI
Batarsheh et al. 2021 - Professor of Electrical and Computer Engineering, Virginia Polytechnic Institute [Feras, Laura Freeman Director, Intelligent Systems Division – National Security Institute Research Associate Professor, Chih Hao Huang, Data Analyst at Turing Research. April 26 “A survey on artificial intelligence assurance” https://link-springer-com.proxy.lib.umich.edu/article/10.1186/s40537-021-00445-7#Sec8]
Verification: “The process of evaluating a system or component to determine whether the products of a given development phase satisfy the conditions imposed at the start of that phase”. Validation: “The process of evaluating a system or component during or at the end of the development process to determine whether it satisfies specified requirements” (Gonzalez and Barr, 2020). Another definition for V&V is from the Department of Defense, as they applied testing practices to simulation systems, it states the following: Verification is the “process of determining that a model implementation accurately represents the developer’s conceptual descriptions and specifications”, and Validation is the process of “determining the degree to which a model is an accurate representation” [60]. Testing: according to the American Software testing Qualification Board, testing is “the process consisting of all lifecycle activities, both static and dynamic, concerned with planning, preparation and evaluation of software products and related work products to determine that they satisfy specified requirements, to demonstrate that they are fit for purpose and to detect defects”. Based on that (and other reviewed definitions), testing includes both validation and verification. Assurance: this term has been rarely applied to conventional software engineering; rather, it is used in the context of AI and learning algorithms. In this manuscript, based on prior definitions and recent AI challenges, we propose the following definition for AI assurance: A process that is applied at all stages of the AI engineering lifecycle ensuring that any intelligent system is producing outcomes that are valid, verified, data-driven, trustworthy and explainable to a layman, ethical in the context of its deployment, unbiased in its learning, and fair to its users. Our definition is by design generic and therefore applicable to all AI domains and subareas. Additionally, based on our review of a wide variety of existing definitions of assurance, it is evident that the two main AI components of interest are the data and the algorithm; accordingly, those are the two main pillars of our definition. Additionally, we highlight that the outcomes the AI enable system (intelligent system) are evaluated at the system level, where the decision or action is being taken.
2022
MNDI
Affirmative
TEVV AFF – MNDI HPA › Solvency › AT Will Shift to Semi Autonomous
Scharre, 2016 -- Vice President and Director of Studies at CNAS [Paul, Feb 2016, Center for New American Security, “Autonomous weapons and operational risk”, https://s3.us-east-1.amazonaws.com/files.cnas.org/documents/CNAS_Autonomous-weapons-operational-risk.pdf?mtime=20160906080515&focal=none, Acc 6/21/22, M. A.]
Autonomous weapon systems—potential future weapons that would select and engage targets on their own—raise a host of legal, ethical, and moral questions. They also raise critically important considerations regarding safety and risk. Autonomous weapons have a qualitatively different degree of risk than equivalent semi-autonomous weapons that would retain a human in the loop. The consequences of a failure that causes the weapon to engage an inappropriate target could be far greater with an autonomous weapon. The result could be fratricide, civilian casualties, or unintended escalation in a crisis. Humans are not immune from errors, and semi-autonomous weapons can also fail. However for semi-autonomous weapons, requiring a human in the loop to authorize each engagement creates a natural fail-safe. If the weapon system begins to fail, the human controller can modify the weapon system’s operation or halt the engagement before further damage is done. With an autonomous weapon, however, the damage potential before a human controller is able to intervene could be far greater. In the most extreme case, an autonomous weapon could continue engaging inappropriate targets until it exhausts its magazine, potentially over a wide area. If the failure mode is replicated in other autonomous weapons of the same type, a military could face the disturbing prospect of large numbers of autonomous weapons failing simultaneously, with potentially catastrophic consequences.
2022
MNDI
Affirmative
TEVV AFF – MNDI HPA › Solvency › AT Testing Fails
Batarsheh et al. 2021 - Professor of Electrical and Computer Engineering, Virginia Polytechnic Institute [Feras, Laura Freeman Director, Intelligent Systems Division – National Security Institute Research Associate Professor, Chih Hao Huang, Data Analyst at Turing Research. April 26 “A survey on artificial intelligence assurance” https://link-springer-com.proxy.lib.umich.edu/article/10.1186/s40537-021-00445-7#Sec8]
The state of AI assurance This section introduces some milestone methods and discussion in AI assurance. Many of the discussed works rely on standard software validation and verification methods. Such methods are inadequate for AI systems, because they have a dimension of intelligence, learning, and re-learning, as well as adaptability to certain contexts. Therefore, errors in AI system “may manifest themselves because of autonomous changes” [211], and among other scenarios would require extensive assurance. For instance, in expert systems, the inference engine component creates rules and new logic based on forward and backward propagation [20]. Such processes require extensive assurance of the process as well as the outcome rules. Alternatively, for other AI areas such as neural networks, while propagation is used, taxonomic evaluations and adversarial targeting are more critical to their assurance [145]. For other subareas such as machine learning, the structure of data, data collection decisions, and other data-relevant properties need step-wise assurance to evaluate the resulted predictions and forecasts. For instance, several types of bias can occur in any phase of the data science lifecycle or while extracting outcomes. Bias can begin during data collection, data wrangling, modeling, or any other phase. Biases and variances which arise in the data are independent of the sample size or statistical significance, and they can directly affect the context or the results or the model. Other issues such as incompleteness, data skewness, or lack of structure have a negative influence on the quality of outcomes of any AI model and require data assurance [117]. While the historic majority of methods for knowledge-based systems and expert systems (as well as neural networks) aimed at finding generic solutions for their assurance [21, 218], and [166], other “more recent” methods were focused on one AI subarea and one domain. For instance, in Mason et al. [142, 144], assurance was applied to reinforcement learning methods for safety–critical systems. Prentzas et al. [174] presented an assurance method for machine learning as its applied to stroke predictions, similar to Pawar’s et al.’s [167] XAI for healthcare framework. Pepe et al. [169], and Chittajallu et al.’s [42] developed a method for surgery video detection methods. Moreover, domains such as law and society would generally benefit from AI subareas such as natural language processing for analyzing legal contracts [135], but also require assurance.