Aff - AI Ethics 2 - MNDI 2022 PHA
Aff - AI Ethics 2 - MNDI 2022 PHA.docx
Indexed text preview
AI Ethics Aff Extend – Inherency - DOD The DOD has not committed enough resources to TEVV, despite support for AI 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] Machine learning applications can increase the speed and quality of human decision-making on the battlefield, enable human-machine teaming to maximize performance and minimize the risk to soldiers, and greatly improve the accuracy and speed of analysis that relies on very large data sets. ML can also strengthen the United States’ ability to defend its networks against cyberattacks at machine speeds and has the power to automate critical components of labor-intensive enterprise functions, such as predictive maintenance and personnel management. Advances in AI and machine learning are not the sole province of the United States, however. Indeed, U.S. global leadership in AI remains in doubt in the face of an aggressive Chinese challenge in the field. Numerous DOD and academic reports reflect on the need to invest more in AI T 2 research and development, train and recruit a skilled workforce, and promote an international environment supportive of American AI innovation—all while promoting safety, security, privacy, and ethical development and use. However, far too little attention is placed on the issue of trust, and especially testing, evaluation, verification, and validation (TEVV) of these systems. Building a robust testing and evaluation ecosystem is a critical component of harnessing this technology responsibly, reliably, and urgently. Failure to do so will mean falling behind. Extend – Harms - Interoperability Diverse AI ethics undermine military interoperability because it prevents different nations militaries from cooperating. Coordinated standards improve cohesion and interoperability van der Merwe, 2021 - Fellow Center for European Policy Analysis Defense Tech Initiative [Joanna, Feb 17, “NATO Leadership on Ethical AI is Key to Future Interoperability” https://cepa.org/nato-leadership-on-ethical-ai-is-key-to-future-interoperability/ BK] If individual nations or groups are left to develop their own ethical principles without wider alignment to NATO, the result will be a number of AI-based systems with varying technical specifications based on the legal and policy decisions made by individual governments when answering the key questions. As has been demonstrated in areas such as facial recognition and policing algorithms, the assumptions made by those developing the tools and answering the key questions have a significant impact on the real-world functioning of the tool and societal acceptance of its ethics. The risk of tools failing to gain acceptance depends on the legal and ethical decisions made by governments. For the military, this may mean one state using an AI-based system that is seen as unacceptable by another, and in a joint operation one state fielding a system that cannot be used by another. Or worse yet, this could render a joint operation impossible. Without the ability to interoperate across NATO, the inability to effectively and efficiently respond to future threats would undermine the Alliance. The role of the private sector is another aspect of ethical AI development that has proved a challenge to governments and the transatlantic relationship. Within states, governments have struggled to adequately regulate Big Tech firms, which has led to these companies encroaching on government responsibilities to protect and uphold the public interest. This encroachment permeates all aspects of government, including defense and security. As Deputy Secretary of Defense Kathleen Hicks discussed during her confirmation hearings, the lack of competition is also a challenge to innovation in the private defense industry. This, along with a lack of regulation, feeds into the power imbalance between the sectors. Consequently, private sector companies building the AI and AI systems that are or will be deployed on the battlefield are deciding the ethics policies for themselves. The transatlantic partnership must focus on coordinating these core principles and systematic governance to ensure AI systems development aligns with the rule of law and democracy. In particular, this must ensure answering questions about human dignity, human control, and accountability. NATO is the ideal defense and security forum for this alignment. Given the US lead on adopting ethical principles for the entire DoD and the EU’s drive to assert checks and balances for private-sector tech companies, NATO remains the organization that can bring these two together and establishes the ethical bottom line. These will then ensure the diverging legal and ethical stances towards Big Tech do not lead to an interoperability barrier in the future. If developments surrounding the General Data Protection Regulation (GDPR) and the challenges it brought for U.S.-based, data-driven companies are any indication, a strong transatlantic led initiative is needed in order to ensure the same challenges do not hinder NATO. The solution to the challenge that ethical AI poses for the future of interoperability within NATO is for the Alliance to establish shared transatlantic ethical principles, informed by the US DoD, the EU, and others. Establishing these principles will not only strengthen transatlantic political relations; more technically, it will allow for the establishment of standardization agreements and inform training and education initiatives of the Alliance in the future. Extend – Harms - Instability AI undermines human control of warfare – it speeds up war beyond our abilities to keep up. Sanderson, 2019 - The Times arts correspondent [David, “Humans may lose control of war with AI weapons” The times. https://go.gale.com/ps/i.do?p=GIC&u=gotitans&id=GALE|A601230754&v=2.1&it=r&sid=summon LMSi] The head of the armed forces has said that artificial intelligence weaponry is evolving so quickly that in the future humans may have limited control over how battles are fought. General Sir Nick Carter, chief of the defence staff, said that "technologies are beginning to question the ethical and moral basis on which we apply the rule of conflict", Speaking at the Cliveden Literary Festival, he said: "There's a debate we need to have about what does the future of warfare look like when you end up with a combination of manned, unmanned and autonomous technologies fighting that war for you ... We may not have the same control as we had in the past." At the same event, David Petraeus, the former head of the CIA, said that advances in AI would make it difficult to "keep the person in the loop". Sir Nick said that space and cyberspace were the new battlefields and that the advance of cyberwarfare meant that we were constantly "at war, just not in the way we defined it in the past". He also warned against thinking that the battle against Islamic State had been won. "We should all realise we are a very long way from dealing with that," he said. "People talked rather too early about beating the phenomenon that was Daesh in Syria and Iraq. "We now have Isis in west Africa, in South Africa, we will have Isis in east Africa quite soon, we have it in China, we have it in the Philippines and, of course, in Afghanistan." Extend – Harms - Dignity AI weapons violate human dignity – they are not accountable and treat people as objects. 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” https://www-tandfonline-com.proxy.lib.umich.edu/doi/full/10.1080/03932729.2020.1864995 TM] Second, the Report contends that AWS are likely to cause an accountability gap, especially in connection with war crimes (Heyns 2013, paras. 75-81). One cannot exclude that AWS might make targeting decisions that, were they taken by human agents, would trigger individual criminal responsibility (Crootof 2016). Authors who oppose an AWS ban observe that, in such cases, the deploying officers would be held legally responsible under the doctrine of superior responsibility (see, for example, Margulies 2017). The rejoinder, however, is that deploying officers may cast an effective defence against criminal prosecution in terms of AWS complexities and their unexpected battlefield behaviours. The difficulty in predicting what these systems will actually do depends on both internal factors – their interacting components and functional modules – and external factors (for example, hostile battleground interactions with other artificial agents). Therefore, there may be cases in which the presence of that mental element required under international criminal law (ICL) to ascribe individual responsibility cannot be ascertained. Consequently, no one would be held criminally liable, even though the conduct in question would materially amount to an international crime (Liu 2016). This is hardly reconcilable with the accountability requirement for actions of military commanders and operators, as well as with the related principle of individual criminal responsibility under ICL. Third, it is maintained that AWS violate the principle of protection of human dignity, in that the latter dictates that decisions affecting the life and physical integrity of human beings involved in armed conflicts should be reserved for human operators (Heyns 2013, paras. 89-97). From this point of view, autonomous targeting is unacceptable because it “objectifies” human beings, reducing them to algorithmically processed “data points”, thereby systematically denying their inherent value as human beings (Moyes 2019, 6). Suppressing human life is ethically and legally justifiable only if it is based on human judgement, for only human decision-making offers a guarantee that the values at stake (human life, physical integrity and so on) can be fully appreciated (Heyns 2016). Extend - Harms – Authoritarian AI Ethical values for AI are crucial to prevent Authoritarian abuse – AI can be misused for propaganda or government surveillance Scharre, 2019 - Vice President and Director of Studies at CNAS [Paul, May 2019, Foreign affairs, “Killer apps: The real dangers of an AI Arms race”, https://www.foreignaffairs.com/articles/2019-04-16/killer-apps, acc 6/20/22, M.A.] Harm from AI misuse isn’t hypothetical; it’s already here. Bots are regularly used to manipulate social media, amplifying some messages and suppressing others. Deepfakes, AI-generated fake videos, have been used in so-called revenge porn attacks, in which a person’s face is digitally grafted onto the body of a pornographic actor. These examples are only the start. Political campaigns will use AI-powered data analytics to target individuals with political propaganda tailored just for them. Companies will use the same analytics to design manipulative advertising. Digital thieves will use AI tools to create more effective phishing attacks. Bots will be able to convincingly impersonate humans online and over the phone by cloning a person’s voice with just a minute of audio. Any interaction that isn’t in person will become suspect. Security specialists have shown that it’s possible to hack into autonomous cars, disabling the steering and brakes. Just one person could conceivably hijack an entire fleet of vehicles with a few keystrokes, creating a traffic jam or launching a terrorist attack. AI’s power as a tool of repression is even more frightening. Authoritarian governments could use deepfakes to discredit dissidents, facial recognition to enable round-the-clock mass surveillance, and predictive analytics to identify potential troublemakers. China has already started down the road toward digital authoritarianism. It has begun a massive repression campaign against the Muslim Uighur population in Xinjiang Province. Many of the tools the government is using there are low tech, but it has also begun to use data analytics, facial recognition systems, and predictive policing (the use of data to predict criminal activity). Vast networks of surveillance cameras are linked up to algorithms that can detect anomalous public behavior, from improperly parked vehicles to people running where they are not allowed. The Chinese company Yuntian Lifei Technology boasts that its intelligent video surveillance system has been deployed in nearly 80 Chinese cities and has identified some 6,000 incidents related to “social governance.” Some of the ways in which Chinese authorities now use AI seem trivial, such as tracking how much toilet paper people use in public restrooms. Their proposed future uses are more sinister, such as monitoring patterns of electricity use for signs of suspicious activity. AI Harms Disadvantaged Groups because “neutral” algorithms and big data reinforce existing discrimination Artificial Intelligence/Machine Learning Risk & Security Working Group, 2020 [Wharton, ‘Artificial Intelligence Risk & Governance’, https://ai.wharton.upenn.edu/artificial-intelligence-risk-governance/] 4. Interpretability and Discrimination Interpretability (presenting the AI system’s results in human understandable format), and discrimination (unfairly biased outcomes) are crucial concepts that factor into the risks associated with AI/ML systems used for certain use cases. In this section, we explore potential risks associated with discrimination and interpretability as they relate to certain applications of AI, e.g., loan approvals. 4.1 Discrimination in AI Depending on the use case, AI may potentially lead to discriminatory and/or unfairly biased outcomes if not implemented appropriately. Poor implementation may arise from biased data, the AI system itself not being properly trained or when there are alternate systems and data sources that could potentially be used to generate better outcomes for disadvantaged groups. Ultimately, the use of an AI system which may cause potentially unfair biased outcomes may lead to regulatory non-compliance issues, potential lawsuits and reputational risk. That said, these risks could be managed. There is even growing evidence that AI/ML systems could be harnessed to more effectively control for discriminatory outcomes. Existing Legal and Regulatory Frameworks Federal and state statutes prohibit discrimination in areas that impact our daily lives, including employment, housing, and lending, to name a few. By way of example, a potential impact in the use of AI for lending is described in greater detail below. The primary U.S. federal statutes that define illegal discrimination in lending are the Equal Credit Opportunity Act (ECOA) and the Fair Housing Act (FHA); however, lenders are subject to many other federal regulations and state laws addressing fairness. Each statute defines types of “protected classes,” such as gender, race, or ethnicity, that a lender cannot legally disfavor. Generally speaking, three types of discrimination are recognized by federal banking regulators: overt discrimination, disparate treatment, and disparate impact when not supported by a legitimate business justification. Disparate treatment discrimination could occur when similarly situated individuals are treated differently based on a prohibited basis, but the treatment does not have to be motivated by prejudice or an intent to discriminate. In an AI context, this may potentially occur, for example, when a firm explicitly uses protected class status in an AI system used to underwrite creditworthiness. Disparate impact, on the other hand, occurs when a system includes features that lead to disproportionately unfavorable outcomes for a protected class. Importantly, evidence of disparate impact is almost always assessed independently of the accuracy and validity of the system. In other words, just because a given system is statistically sound does not mean that it is legally non-discriminatory. Such systems are generally not considered legally discriminatory if they and their constituent features could be demonstrated to meet a legitimate business need and where no less discriminatory alternative system or process could be identified that also meets those needs. Concerns over using and potentially amplifying implicitly biased data also arise in other contexts. For instance, the New York Department of Financial Security (NY DFS) discussed [7] the use of external consumer data and information sources in insurance underwriting, noting the potential of leveraging these sources to help establish lifestyle indicators that may inform the review of an application for insurance coverage. In doing so, however, NY DFS observed that such data may be inaccurate or unreliable, and its use may result in a significant detrimental impact to the insured. Similarly, in a speech [8] by Charles Randell, Chair of the UK Financial Conduct Authority, concerns over misuse of big data to inform potentially detrimental outcomes were raised, with a real-world example in the use of data mining credit card charges for services such as marriage counseling, and reducing cardholders’ credit limits on the basis of the correlation between marriage breakdown and debt default. The use and potential for misuse of big data is no longer a theoretical concern and should be considered in determining the types of data that may be used in developing AI/ML systems. We reference these legal and regulatory considerations to illustrate existing standards that already apply to many algorithmic activities of financial institutions, especially as they relate to unfairly biased outcomes. Data as a Cause of Discriminatory AI A host of factors may result in AI-related illegal discrimination. Input data may cause illegal discrimination if it identifies or closely proxies class membership, if it causes protected class members to experience less favorable outcomes, or if it is differentially predictive of the outcome for the protected class. Traditional data inputs, such as many credit bureau attributes, tend to be less likely to raise disparate impact concerns because they are generally thoroughly vetted and accepted for credit worthiness. They may also be differentially predictive if the system’s weights or coefficients do not properly account for class-specific idiosyncrasies. Non-traditional data, such as utility payment history, rental payments, or a person’s digital footprint (including social media posting), may generate heightened concerns relative to traditional data. From a fairness perspective, such data may have substantial merit, as its use has been shown to expand access to the financial system for unbanked or underserved populations that are often more likely to be members of some protected groups. However, such data use often raises coverage and accuracy concerns. Algorithms as a Cause of Discriminatory AI Algorithms themselves may result in discriminatory outcomes exacerbated by their complexity and opacity. Some of this concern arises from the fact that some machine learning algorithms create variable interactions and non-linear relationships that are too complex for humans to identify and review. These relationships have the potential to cause disparate treatment discrimination by creating proxies for protected class status. To some degree, these concerns have been lessened by advances in explainable AI techniques that allow additional insight into these complex relatio
Preview shortened. Download the file for the full document.