The Ethics of Automated Warfare and Artificial Intelligence

Essay series introduced by Bessma Momani, Aaron Shull and Jean-François Bélanger: “…begins with a piece written by Alex Wilner titled “AI and the Future of Deterrence: Promises and Pitfalls.” Wilner looks at the issue of deterrence and provides an account of the various ways AI may impact our understanding and framing of deterrence theory and its practice in the coming decades. He discusses how different countries have expressed diverging views over the degree of AI autonomy that should be permitted in a conflict situation — as those more willing to cut humans out of the decision-making loop could gain a strategic advantage. Wilner’s essay emphasizes that differences in states’ technological capability are large, and this will hinder interoperability among allies, while diverging views on regulation and ethical standards make global governance efforts even more challenging.

Looking to the future of non-state use of drones as an example, the weapon technology transfer from nation-state to non-state actors can help us to understand how next-generation technologies may also slip into the hands of unsavoury characters such as terrorists, criminal gangs or militant groups. The effectiveness of Ukrainian drone strikes against the much larger Russian army should serve as a warning to Western militaries, suggests James Rogers in his essay “The Third Drone Age: Visions Out to 2040.” This is a technology that can level the field by asymmetrically advantaging conventionally weaker forces. The increased diffusion of drone technology enhances the likelihood that future wars will also be drone wars, whether these drones are autonomous systems or not. This technology, in the hands of non-state actors, implies future Western missions against, say, insurgent or guerilla forces will be more difficult.

Data is the fuel that powers AI and the broader digital transformation of war. In her essay “Civilian Data in Cyber Conflict: Legal and Geostrategic Considerations,” Eleonore Pauwels discusses how offensive cyber operations are aiming to alter the very data sets of other actors to undermine adversaries — whether through targeting centralized biometric facilities or individuals’ DNA sequence in genomic analysis databases, or injecting fallacious data into satellite imagery used in situational awareness. Drawing on the implications of international humanitarian law, Pauwels argues that adversarial data manipulation constitutes another form of “grey zone” operation that falls below a threshold of armed conflict. She evaluates the challenges associated with adversarial data manipulation, given that there is no internationally agreed upon definition of what constitutes cyberattacks or cyber hostilities within international humanitarian law (IHL).

In “AI and the Actual International Humanitarian Law Accountability Gap,” Rebecca Crootoff argues that technologies can complicate legal analysis by introducing geographic, temporal and agency distance between a human’s decision and its effects. This makes it more difficult to hold an individual or state accountable for unlawful harmful acts. But in addition to this added complexity surrounding legal accountability, novel military technologies are bringing an existing accountability gap in IHL into sharper focus: the relative lack of legal accountability for unintended civilian harm. These unintentional acts can be catastrophic, but technically within the confines of international law, which highlights the need for new accountability mechanisms to better protect civilians.

Some assert that the deployment of autonomous weapon systems can strengthen compliance with IHL by limiting the kinetic devastation of collateral damage, but AI’s fragility and apparent capacity to behave in unexpected ways poses new and unexpected risks. In “Autonomous Weapons: The False Promise of Civilian Protection,” Branka Marijan opines that AI will likely not surpass human judgment for many decades, if ever, suggesting that there need to be regulations mandating a certain level of human control over weapon systems. The export of weapon systems to states willing to deploy them on a looser chain-of-command leash should be monitored…(More)”.

Machine Learning in Public Policy: The Perils and the Promise of Interpretability

Report by Evan D. Peet, Brian G. Vegetabile, Matthew Cefalu, Joseph D. Pane, Cheryl L. Damberg: “Machine learning (ML) can have a significant impact on public policy by modeling complex relationships and augmenting human decisionmaking. However, overconfidence in results and incorrectly interpreted algorithms can lead to peril, such as the perpetuation of structural inequities. In this Perspective, the authors give an overview of ML and discuss the importance of its interpretability. In addition, they offer the following recommendations, which will help policymakers develop trustworthy, transparent, and accountable information that leads to more-objective and more-equitable policy decisions: (1) improve data through coordinated investments; (2) approach ML expecting interpretability, and be critical; and (3) leverage interpretable ML to understand policy values and predict policy impacts…(More)”.

AI Audit-Washing and Accountability

Report by Ellen P. Goodman and Julia Tréhu: “.. finds that auditing could be a robust means for holding AI systems accountable, but today’s auditing regimes are not yet adequate to the job. The report assesses the effectiveness of various auditing regimes and proposes guidelines for creating trustworthy auditing systems.

Various government and private entities rely on or have proposed audits as a way of ensuring AI systems meet legal, ethical and other standards. This report finds that audits can in fact provide an agile co-regulatory approach—one that relies on both governments and private entities—to ensure societal accountability for algorithmic systems through private oversight.

But the “algorithmic audit” remains ill-defined and inexact, whether concerning social media platforms or AI systems generally. The risk is significant that inadequate audits will obscure problems with algorithmic systems. A poorly designed or executed audit is at best meaningless and at worst even excuses harms that the audits claim to mitigate.

Inadequate audits or those without clear standards provide false assurance of compliance with norms and laws, “audit-washing” problematic or illegal practices. Like green-washing and ethics-washing before, the audited entity can claim credit without doing the work.

The paper identifies the core specifications needed in order for algorithmic audits to be a reliable AI accountability mechanism:

  • Who” conducts the audit—clearly defined qualifications, conditions for data access, and guardrails for internal audits;
  • What” is the type and scope of audit—including its position within a larger sociotechnical system;
  • Why” is the audit being conducted—whether for narrow legal standards or broader ethical goals, essential for audit comparison, along with potential costs; and
  • How” are the audit standards determined—an important baseline for the development of audit certification mechanisms and to guard against audit-washing.

Algorithmic audits have the potential to increase the reliability and innovation of technology in the twenty-first century, much as financial audits transformed the way businesses operated in the twentieth century. They will take different forms, either within a sector or across sectors, especially for systems that pose the highest risk. Ensuring that AI is accountable and trusted is key to ensuring that democracies remain centers of innovation while shaping technology to democratic values…(More)”

We could run out of data to train AI language programs 

Article by Tammy Xu: “Large language models are one of the hottest areas of AI research right now, with companies racing to release programs like GPT-3 that can write impressively coherent articles and even computer code. But there’s a problem looming on the horizon, according to a team of AI forecasters: we might run out of data to train them on.

Language models are trained using texts from sources like Wikipedia, news articles, scientific papers, and books. In recent years, the trend has been to train these models on more and more data in the hope that it’ll make them more accurate and versatile.

The trouble is, the types of data typically used for training language models may be used up in the near future—as early as 2026, according to a paper by researchers from Epoch, an AI research and forecasting organization, that is yet to be peer reviewed. The issue stems from the fact that, as researchers build more powerful models with greater capabilities, they have to find ever more texts to train them on. Large language model researchers are increasingly concerned that they are going to run out of this sort of data, says Teven Le Scao, a researcher at AI company Hugging Face, who was not involved in Epoch’s work.

The issue stems partly from the fact that language AI researchers filter the data they use to train models into two categories: high quality and low quality. The line between the two categories can be fuzzy, says Pablo Villalobos, a staff researcher at Epoch and the lead author of the paper, but text from the former is viewed as better-written and is often produced by professional writers…(More)”.

AI Localism in Practice: Examining How Cities Govern AI

Report by Sara Marcucci, Uma Kalkar, and Stefaan Verhulst: “…serves as a primer for policymakers and practitioners to learn about current governance practices and inspire their own work in the field. In this report, we present the fundamentals of AI governance, the value proposition of such initiatives, and their application in cities worldwide to identify themes among city- and state-led governance actions. We close with ten lessons on AI localism for policymakers, data, AI experts, and the informed public to keep in mind as cities grow increasingly ‘smarter’, which include: 

  • Principles provide a North Star for governance;
  • Public engagement provides a social license;
  • AI literacy enables meaningful engagement;
  • Tap into local expertise;
  • Innovate in how transparency is provided;
  • Establish new means for accountability and oversight;
  • Signal boundaries through binding laws and policies;
  • Use procurement to shape responsible AI markets;
  • Establish data collaboratives to tackle asymmetries; and
  • Make good governance strategic.

Considered together, we look to use our understanding of governance practices, local AI governance examples, and the ten overarching lessons to create an incipient framework for implementing and assessing AI localism initiatives in cities around the world….(More)”

Measuring the environmental impacts of artificial intelligence compute and applications

OECD Paper: “Artificial intelligence (AI) systems can use massive computational resources, raising sustainability concerns. This report aims to improve understanding of the environmental impacts of AI, and help measure and decrease AI’s negative effects while enabling it to accelerate action for the good of the planet. It distinguishes between the direct environmental impacts of developing, using and disposing of AI systems and related equipment, and the indirect costs and benefits of using AI applications. It recommends the establishment of measurement standards, expanding data collection, identifying AI-specific impacts, looking beyond operational energy use and emissions, and improving transparency and equity to help policy makers make AI part of the solution to sustainability challenges…(More)”.

Algorithms in the Public Sector. Why context matters

Paper by Georg Wenzelburger, Pascal D. König, Julia Felfeli, and Anja Achtziger: “Algorithms increasingly govern people’s lives, including through rapidly spreading applications in the public sector. This paper sheds light on acceptance of algorithms used by the public sector emphasizing that algorithms, as parts of socio-technical systems, are always embedded in a specific social context. We show that citizens’ acceptance of an algorithm is strongly shaped by how they evaluate aspects of this context, namely the personal importance of the specific problems an algorithm is supposed to help address and their trust in the organizations deploying the algorithm. The objective performance of presented algorithms affects acceptance much less in comparison. These findings are based on an original dataset from a survey covering two real-world applications, predictive policing and skin cancer prediction, with a sample of 2661 respondents from a representative German online panel. The results have important implications for the conditions under which citizens will accept algorithms in the public sector…(More)”.

Artificial Intelligence and Life in 2030: The One Hundred Year Study on Artificial Intelligence

Paper by Peter Stone et al: “In September 2016, Stanford’s “One Hundred Year Study on Artificial Intelligence” project (AI100) issued the first report of its planned long-term periodic assessment of artificial intelligence (AI) and its impact on society. It was written by a panel of 17 study authors, each of whom is deeply rooted in AI research, chaired by Peter Stone of the University of Texas at Austin. The report, entitled “Artificial Intelligence and Life in 2030,” examines eight domains of typical urban settings on which AI is likely to have impact over the coming years: transportation, home and service robots, healthcare, education, public safety and security, low-resource communities, employment and workplace, and entertainment. It aims to provide the general public with a scientifically and technologically accurate portrayal of the current state of AI and its potential and to help guide decisions in industry and governments, as well as to inform research and development in the field. The charge for this report was given to the panel by the AI100 Standing Committee, chaired by Barbara Grosz of Harvard University….(More)”.

The political imaginary of National AI Strategies

Paper by Guy Paltieli: “In the past few years, several democratic governments have published their National AI Strategies (NASs). These documents outline how AI technology should be implemented in the public sector and explain the policies that will ensure the ethical use of personal data. In this article, I examine these documents as political texts and reconstruct the political imaginary that underlies them. I argue that these documents intervene in contemporary democratic politics by suggesting that AI can help democracies overcome some of the challenges they are facing. To achieve this, NASs use different kinds of imaginaries—democratic, sociotechnical and data—that help citizens envision how a future AI democracy might look like. As part of this collective effort, a new kind of relationship between citizens and governments is formed. Citizens are seen as autonomous data subjects, but at the same time, they are expected to share their personal data for the common good. As a result, I argue, a new kind of political imaginary is developed in these documents. One that maintains a human-centric approach while championing a vision of collective sovereignty over data. This kind of political imaginary can become useful in understanding the roles of citizens and governments in this technological age….(More)”.

Algorithms Quietly Run the City of DC—and Maybe Your Hometown

Article by Khari Johnson: “Washington, DC, IS the home base of the most powerful government on earth. It’s also home to 690,000 people—and 29 obscure algorithms that shape their lives. City agencies use automation to screen housing applicants, predict criminal recidivism, identify food assistance fraud, determine if a high schooler is likely to drop out, inform sentencing decisions for young people, and many other things.

That snapshot of semiautomated urban life comes from a new report from the Electronic Privacy Information Center (EPIC). The nonprofit spent 14 months investigating the city’s use of algorithms and found they were used across 20 agencies, with more than a third deployed in policing or criminal justice. For many systems, city agencies would not provide full details of how their technology worked or was used. The project team concluded that the city is likely using still more algorithms that they were not able to uncover.

The findings are notable beyond DC because they add to the evidence that many cities have quietly put bureaucratic algorithms to work across their departments, where they can contribute to decisions that affect citizens’ lives.

Government agencies often turn to automation in hopes of adding efficiency or objectivity to bureaucratic processes, but it’s often difficult for citizens to know they are at work, and some systems have been found to discriminate and lead to decisions that ruin human lives. In Michigan, an unemployment-fraud detection algorithm with a 93 percent error rate caused 40,000 false fraud allegations. A 2020 analysis by Stanford University and New York University found that nearly half of federal agencies are using some form of automated decisionmaking systems…(More)”.