"Using Graph Neural Networks to Model Biased Crowdsourced Data for Urban Applications": Graph neural networks (GNNs) are widely used on graph-structured data in urban spatiotemporal forecasting applications, such as predicting infrastructure problems and weather events. In such settings, nodes have a true latent state (e.g., street condition) that is sparsely observed (e.g., via government inspection ratings). We more frequently observe biased proxies for the latent state (e.g., via crowdsourced reports) that correlate with resident demographics.

"Fostering Tech Entrepreneurialism & Entrepreneurial Infrastructure Among Marginalized": CommunitiesMarginalized entrepreneurs and those at society’s fringes often lack access to entrepreneurial infrastructure such as business education, access to capital, and payment processing. Drawing from several multi-year interview case studies with digital sex workers, influencers, content creators, tech innovators, leaders, startup founders and impact investors, this thesis explores individual, institutional, and philanthropic approaches to addressing the lack of access marginalized populations have to tech entrepreneurship and representation in global startup culture.

Democracies augment their bureaucratic decision-making increasingly with artificial intelligence (AI). This AI-augmentation is a threat to democratic governance. And this threat is much deeper than is commonly understood, or so I argue in this paper: The usual metrics to assess AI risks and benefits—i.e., bias, or procedural and substantive fairness—fail to capture the full extent to which the AI-augmented state undermines democracy. Instead, norms of responsible public service offer a more comprehensive basis for assessing AI’s risks and benefits.

In both computer science and economics, efficiency is a cherished property. The field of algorithms is almost solely focused on their efficiency. The goal of AI research is to increase efficiency by reducing human labor. In economics, the main advantage of the free market is that it promises “economic efficiency.” A major lesson from many recent disasters is that both fields have overemphasized efficiency and underemphasized resilience.

Laws regulating pornographic deepfakes are often characterized as protecting privacy or preventing defamation. But privacy and defamation laws paradigmatically regulate true or false assertions of fact about persons. Anti- deepfakes laws do not: the typical law bans even media that no reasonable observer could understand as factual. Instead of regulating statements of fact, anti-deepfakes laws ban outrageous depictions per se. This is a significant and unrecognized departure from the established dignitary torts, and it is important to acknowledge for two reasons.

Animation increasingly defines the cultural contours of the twenty-first century and is broadly used across many forms of digital media. More than just cartooning, puppetry, or CGI, animation is a paradigm involving the projection of qualities perceived as human such as power, agency, will, and personality outside of the self and onto objects in the environment. In this talk, I argue ChatGPT and similar interactive AI systems, both powered by Large Language Models (LLMs) and not, can be best understood as animated entities.

Influential groups of thinkers, entrepreneurs, and activists argue that networked digital technologies offer potent mechanisms for counteracting and neutralizing state and private power. In particular, they argue that blockchain-based technologies for decentralized, secure authentication of identity and provenance supply the key to unwinding contemporary relations of dominance and replacing them with durably decentralized, bottom-up governance arrangements. This seminar pulls on two interrelated threads in emergent narratives about decentralized digital futures.

The use of artificially intelligent algorithms in local public decision-making is under immense scrutiny. Policymakers, activists, and scholars increasingly question why agencies and courts should have unfettered discretion to deploy privately developed algorithms that affect citizens’ rights, liberties, and opportunities without adequate democratic accountability. In recent years, the injustices faced by those subjected to inaccurate, biased, or procedurally unfair algorithmic predictions have captured the media, igniting a nationwide movement to govern the state’s use of algorithms in the public sphere.

Although targeted advertising has been touted as a way to give advertisers a choice in who they reach, increasingly, ad delivery algorithms designed by the ad platforms are invisibly refining those choices. In this talk, Aleksandra Korolova will present findings from "black-box" auditing of the role of ad delivery algorithms in shaping who sees opportunity and political ads using only the tools and data accessible to any advertiser. She will then discuss legal and policy efforts to mitigate the harmful effects of ad delivery in these domains, including their shortcomings and potential paths forward.

What counts as legitimate AI ethics labor, and consequently, what are the epistemic terms on which AI ethics claims are rendered legitimate? Based on 75 interviews with technologists including researchers, developers, open source contributors, and activists, this talk explores various epistemic bases from which AI ethics is discussed and practiced. In the context of outside attacks on AI ethics as an impediment to “progress,” David Widder shows how some AI ethics practices have reached toward scholarly authority, automation and quantification and achieved some legitimacy as a result, while those based on richly embodied and situated lived experience have not.

As automation and AI re-organise social, political, and economic life, law’s reflex is often towards the human. By ensuring automation is “human-centred” and subject to “human oversight”, the hope is that automation’s economic promises can be safely realised. Despite the intuitive appeal, analysts have started documenting the empirical failure of human oversight to improve decision quality. But this creates a conundrum. There may be little scientific evidence supporting legal human oversight requirements, but abdicating human agency in automated decision processes is ethically and politically unfeasible.

Several jurisdictions are now expanding their data protection laws in response to proliferating AI-driven evaluations of consumers, workers, borrowers, and internet users. As they digitize judgment, these evaluations risk imposing benefits and burdens in opaque and unaccountable ways via automated decision-making. Information access rights guaranteed via data protection law can assist those who have been treated unfairly—but only if they are clarified and enforced well. The key to doing so, Pasquale claims, is close attention to the practical consequences of data access and explainability.

"Bright Patterns: Towards Extra Ethical User Experience Design": Dark Patterns have received significant attention from the academic Human-Computer Interaction community since 2010, when Brignull launched darkpatterns.org and initially described them as "tricks used in websites and apps that make you do things that you didn’t mean to." After years of collective action, governments worldwide are recognizing the need to regulate them—with the GDPR in the EU, the FTC in the U.S., and, more recently, the CCPA in India, all prohibiting specific types of Dark Patterns. – Hauke Sandhaus

"Explorable Explainable AI: Improving AI Understanding for Community Health Workers in India": AI technologies are increasingly deployed to support community health workers (CHWs) in high-stakes healthcare settings, from malnutrition diagnosis to diabetic retinopathy. Yet, little is known about how such technologies are understood by CHWs with low digital literacy and what can be done to make AI more understandable for them. – Ian René Solano-Kamaiko

The advent of text-to-image models that can produce sophisticated digital images in a matter of seconds has raised alarm bells within the artistic community. Human artists working with more traditional media have warned about the effects of generative models on the market for their works, and the future of human creativity. While artists whose works were involuntarily used to train generative models have legitimate cause for complaint (and reasonable demands for compensation and attribution), the emergence of generative models is not as apocalyptic as framed. By providing amateur creators with powerful tools for creative expression, generative models help to democratize cultural participation, and diversify public discourse. Accordingly, those who view human creativity as a relational and dialogic practice should embrace generative AI’s capacity to extend this practice to previously excluded communities.

From autocomplete and smart replies to video filters and deepfakes, we increasingly live in a world where communication between humans is augmented by artificial intelligence. AI often operates on behalf of a human communicator by recommending, suggesting, modifying, or generating messages to accomplish communication goals. We call this phenomenon AI-Mediated Communication (or AI-MC). While AI-MC has the potential of making human communication more efficient, it impacts other aspects of our communication in ways that are not yet well understood. Over the last six years, my collaborators and I have been documenting the impact of AI-MC on communication outcomes, language use, interpersonal trust, and more. The talk will outline experimental findings from this work.

"Asynchronous Workflows for Maintaining Hardware": Hardware is challenging for end-users to troubleshoot, repair, and maintain on their own. As a result, they seek out expert technicians to remedy their issues. This kind of synchronous support is limited by expert availability and is challenging to scale. DLI Doctoral Fellow Amritansh Kwatra will introduce SplatOverflow, a workflow that enables asynchronous maintenance of hardware; allowing experts to assist end-users without the need for live communication.

"Algorithmic Monoculture: Modeling Risks and Benefits": More than half of the 100 largest companies in the U.S. now use HireVue's screening algorithms. What are the risks and (perhaps) benefits of an emerging algorithmic monoculture, where many decision-makers rely on the same algorithm to make consequential decisions? In this talk, DLI Doctoral Fellow Kenny Peng will present findings from a new mathematical model of algorithmic monoculture, challenging some prevailing intuitions.

Modern copyright law seems determined to impede people’s engagement with creative expression. Rock-bottom creativity requirements, the abolition of formal prerequisites to copyright ownership, and an irrationally long copyright term ensure that nearly all recorded culture is encumbered not merely for years, but for generations. Today, however, change is in the air—for all the wrong reasons. By historical accident, the same foundational properties of copyright law that have long undermined creators and audiences now pose an existential threat to generative AI. Tech companies and their allies are pushing to reform the very aspects of copyright law that impoverish traditional readership and authorship. But by and large, their proposals would change these doctrines only in ways that benefit the generative AI enterprise. This essay offers an alternative: copyright accelerationism.

Social technologies such as chat, smart speakers, and personal robots highlight the growing concern and nuance around the safety and privacy of vulnerable populations in mediated environments. In this paper, we first critique the way law treats vulnerability as binary or status based. Next, drawing from various phenomena and literature such as dark patterns, digital market manipulation, and computers as social actors, we develop a theory of socio-digital vulnerability.

Recent policy developments suggest that consumer sovereignty models of regulation have substantial, if not fatal, limitations. Binding decisions by the European Data Protection Board in 2023, as well as other recent public law enactments in the EU and the US, overtly reject the assumption that individuals are best situated to manage how companies process or use their personal information.

You can’t question a secret you haven’t been told. The criminal legal system depends on fair and open proceedings to expose and regulate unlawful and unconstitutional police conduct through the courts. If police can use claims of secrecy to systematically thwart criminal defendants' access to evidence, judicial review will fail. And yet that is exactly what is happening under a common-law doctrine called the “law enforcement privilege.” The privilege empowers police and prosecutors to rely on the results of secret investigative methods while withholding information from the defense about how those methods work.

After increasing rapidly over seven years, the number of active contributors to English Wikipedia peaked in 2007 and has been in decline since. I will present a body of evidence that suggests that English Wikipedia's pattern of growth and decline appears to be a general feature of "peer production"—the model of collaborative production that has produced millions of wikis, free/open source software projects, websites like OpenStreetMap, and more.

Understanding the impact of AI on society requires understanding how people feel about their impact and what people, outside those building them, think AI should do. However, while performance of algorithms is measured in terms of a now field-standard set of performance metrics, these metrics and their interpretation are not standard knowledge for most people who are impacted by these systems. To express their thoughts reliably, non-expert study participants need to understand what they are being asked.

I will begin with an overview of mechanism design, the field of microeconomic theory focused on the design of institutions and rules that produce good social outcomes. I will discuss a prominent case in which mechanism design was used to advocate for a major policy change in the name of equity – a change in the algorithms used to assign K-12 students to public school in Boston. I will talk about how, inevitably, a gap emerges between what the theory suggests, and what stakeholders think the theory suggests.

This talk outlines the general approach of the ADM+S Centre to issues of automated decision-making in the news and media domain, addressing both the challenges for researchers seeking to understand developments in this field, and the questions of policy, ethics and regulation that arise from it. We use this lens to introduce several research projects underway in the Centre, focussing on observability and explainability in relation to key applications and infrastructures (search engines), business systems (social media advertising), and the social distribution of human capabilities (digital inclusion among vulnerable communities).

Since it was introduced in 2006 by theoretical computer scientists Dwork, McSherry, Nissim, and Smith, differential privacy has become the leading framework for ensuring that individual-level information is not leaked through statistical releases or machine learning models built from sensitive datasets. In addition to a rich theoretical literature, differential privacy has also started to make the transition to practice, with large-scale applications by the US Census Bureau and technology companies like Google, Apple, Microsoft, and Meta.

Algorithms, data extraction, digital marketers monetizing "eyeballs": these all seem like such recent features of our lives. And yet, Lee McGuigan tells us in this eye-opening book, digital advertising was well underway before the widespread use of the Internet. Explaining how marketers have brandished the tools of automation and management science to exploit new profit opportunities, Selling the American People traces data-driven surveillance all the way back to the 1950s, when the computerization of the advertising business began to blend science, technology, and calculative cultures in an ideology of optimization. With that ideology came adtech, a major infrastructure of digital capitalism.

We examine three common principles underlying a slew of “privacy-enhancing” techniques recently deployed or scheduled for deployment by big tech companies: (1) “We deny (or throttle) access to your data by third parties!”; (2) “We minimize the use and retention of raw data!”; and (3) “Your data never leaves your device!” Our article challenges these principles, not on the grounds that techniques offered to implement them are unsuccessful in achieving their stated goals. Instead, we argue that the principles themselves fall short because their underlying conception of privacy is flawed.

Effectively reducing hate speech on social media is a defining challenge of the digital age. Hate speech expressions inflict non-trivial harm to individuals or groups based on their ethnicity, gender, religion and other characteristics. Hate speech’s lasting effects silence and marginalize vulnerable communities. When people see hate speech on social media they sometimes counter it publicly by condemning the transgression itself and/or by showing solidarity with the victim. In an in-the-wild study on X (formerly Twitter), we controlled user accounts to deliberately counter racist slurs and investigated whether transgressors would change their transgression behavior following our intervention.

Open-source licenses are infrastructure that collaborative communities in-habit. These licenses don’t just define the legal terms under which members (and outsiders) can use and build on the contributions of others. They also reflect a community’s consensus on the reciprocal obligations that define it as a community. A license is a statement of values, in legally executable form, adapted for daily use. As such, a license must be designed, much as the software and hardware that open-source developers create. Sometimes an existing license is fit to purpose and can be adopted without extensive discussion. However, often the technical or social needs of a community do not precisely map onto existing licenses, or the community itself is divided about the norms a license should enforce.

Lawyers lead the investigations for many cybersecurity incidents, ranging from data breaches to ransomware, in part because they can often shield any materials produced after a breach from discovery under either attorney-client privilege or work product immunity. Moreover, by limiting and shaping the documentation that is produced by breached firms’ personnel and third-party consultants in the wake of a cyberattack, attorneys can limit the availability of potentially damaging information to plaintiffs’ attorneys, regulators, or media, even if their attorney-client privilege and work product immunity arguments falter.

Machine learning (ML) algorithms play an increasingly prominent role in the distribution of benefits and burdens in sensitive domains. However, ML systems risk introducing biases that undermine values of justices and fairness, for instance, by perpetuating or even exacerbating unjustifiable harms against vulnerable communities. In response to this concern, a burgeoning field of fair ML has emerged, with researchers developing various fairness measures and methodologies for quantifying and mitigating algorithmic harms. From this perspective, philosophical theorizing about ML tools is often constrained to clarifying the normative underpinnings of fairness measures to help resolve disagreements arising from impossibility results and formal trade-offs. In this talk, Sina Fazelpour argue that this type of fair ML strategy is usefully characterized as a problematic form of ideal theorizing about justice, and thus suffers from limitations known to plague that approach more broadly.

As social media continues to have a significant influence on public opinion, understanding the impact of the machine learning algorithms that filter and curate content is crucial. However, existing studies have yielded inconsistent results, potentially due to limitations such as reliance on observational methods, use of simulated rather than real users, restriction to specific types of content, or internal access requirements that may create conflicts of interest. To overcome these issues, we conducted a pre-registered controlled experiment on Twitter's algorithm without internal access. The key to our design was to, for a large group of active Twitter users, simultaneously collect (a) the tweets the algorithm shows, and (b) the tweets the user would have seen if they were just shown the latest tweets from people they follow; we then surveyed users about both sets of tweets in a random order.

Privacy risk assessments have been touted as an objective, principled way to encourage organizations to implement privacy-by-design. They are central to a new regulatory model of collaborative governance, as embodied by the GDPR. However, existing guidelines and methods are vague, and there is little empirical evidence on privacy harms, questioning the suitability of privacy risk assessments as an effective policy instrument. In this talk, Ero Balsa will present a close analysis of US NIST’s Privacy Risk Assessment Methodology, highlighting multiple sites of discretion that create countless opportunities for adversarial organizations to engage in performative compliance.

The increasing harms caused by hate, harassment, and other forms of abuse online have motivated major platforms to explore hierarchical governance. The idea is to allow communities to have designated members take on moderation and leadership duties; meanwhile, members can still escalate issues to the platform. But these promising approaches have only been explored in plaintext settings where community content is public to the platform. It is unclear how one can realize hierarchical governance in the huge and increasing number of online communities that utilize end-to-end encrypted (E2EE) messaging for privacy.

Much research on the law and policy of algorithms has focused on ways to detect or prevent algorithmic misbehavior or mistake. However, there are also problems that result when algorithms perform their assigned tasks too well rather than too poorly. This presentation makes the case that significant individual harms and social welfare losses alike can and do occur in the face of the ever more common phenomenon of the too accurate algorithm.

Artificial intelligence has become increasingly ingrained in the fabric of everyday life, yet sociologists know little about how technology producers design artificial intelligence. Margot will present her research which draws upon in-depth interviews with twenty-one voice assistant designers at major technology companies. The study examines how engineers weighed multiple and sometimes competing organizational goals in making decisions about how to produce “personhood” in voice assistant devices.

Mental health "digital biomarkers" purport to measure mental health and well-being using data collected from mobile devices and technology platforms (eg, location, usage patterns). Translating digital biomarkers into clinical practice raises multiple sociotechnical questions that need to be addressed if these computational tools intend to improve care. How do we assess whether digital biomarkers will be reliable in clinical settings, and even in cases where measurement is reliable, will digital biomarkers improve mental health?

Increasingly concerned about the way in which content spreads on the internet, scholars reach for the concept of "algorithmic amplification" (AA) as both an explanation and a warning. Although these researchers frequently acknowledge the metaphorical and conceptual haziness around the term, they continue to rely on it to carry both descriptive and normative intent. This project aims to do the foundational work to give "algorithmic amplification" conceptual precision and normative teeth.

When humans first imagined robots and computers in the workplace, they envisioned them as servants or supporters of humankind. However, the integration of technology in the workplace has evolved. In recent years, numerous online platforms, such as Uber or Doordash, have been "employing" large numbers of human workers for tasks supervised, controlled, and managed by algorithms - a phenomenon known as algorithmic management. This shift in the role of technology raises several critical questions, such as how it feels to have an AI algorithm as a boss and how human workers react to maltreatment by algorithms.

Research on algorithmic imaginaries related to creators and influencers often focuses on their efforts to understand and best navigate algorithms to maximize visibility, and thus profits. However, there has been less work on how non-influencers work together to redistribute algorithmically-produced visibility through beliefs about how algorithms ought to work. Using "algorithmic ethnography" (Christin, 2020), Caplan and her co-authors, Elena Maris (UIC) and Hibby Thach (UIC) have identified three TikTok genres that they argue are emblematic of how practices of mutual aid (Spade, 2020) are unfolding over platforms.

Rethinking the regulation of advertising-based platform business models such as Facebook/Meta and Google/Alphabet, which I call attention platforms, is an urgent task. Two decades of regulatory apathy and intellectual fragmentation have produced siloed approaches to the regulation of data and content that leave many urgent political, economic and environmental issues unaddressed. In this paper, I argue that current approaches to regulating data and datafication – approaches that regulate control over personal data or that focus on regulating social data – fail to address the most pervasive forms of extraction and harm in the attention platform economy: those that stem from addiction, over-consumption, virality, and fragmentation of the public sphere.

Attention capitalism has generated design processes and product development decisions that prioritize platform growth over all other considerations. To the extent limits have been placed on these incentives, interventions have primarily taken the form of content moderation. While moderation is important for what we call “acute harms,” societal-scale harms – such as negative effects on mental health and social trust – require new forms of institutional transparency and scientific investigation, which we group under the name accountability infrastructure.

A data-generation revolution is underway. Up until recently, most of the data used for decision-making was collected from events that take place in the physical world ("real" data). Yet, it is forecasted that by 2024, 60% of data used to train artificial intelligence systems around the world will be synthetic (!). Synthetic data is artificially-generated data that has analytical value. For some purposes, synthetic datasets can replace real data by preserving or mimicking their properties. For some others it can complement real data in a way which increases their accuracy or their privacy protection. The importance of this data revolution for our economies and societies cannot be over-stated. It affects data access and data flows, potentially changing the competitive dynamics in markets where real data is not easily collected, and potentially affecting decision-making functions for many spheres of our lives.

In this talk, Erin Miller will challenge the First Amendment orthodoxy that speech rights bind only the state. She will argue that the primary justification for the freedom of speech is to protect interests like autonomy, democracy, and knowledge from the kind of extraordinary power available to the state. If so, it applies with nearly equal force to any private agents with power over speech rivaling that of the state. Such a class of private agents, which she calls quasi-state agents, turns out to be a live possibility once we recognize that state power is more limited than it seems and can be broken down into multiple, equally threatening parts. They might include, for example, the largest social media platforms and powerful private employers.

Niva Elkin-Koren is a Professor of Law at Tel-Aviv University Faculty of Law and a Faculty Associate at the Berkman Klein Center for Internet & Society at Harvard University. She is a former Dean of University of Haifa Faculty of Law, and the founding director of the Center for Cyber, Law and Policy (CCLP) and of the Haifa Center for Law & Technology (HCLT).

A century ago, John Dewey observed that '[s]team and electricity have done more to alter the conditions under which men associate together than all the agencies which affected human relationships before our time'. In the last few decades, computing technologies have had a similar effect. Political philosophy's central task is to help us decide how to live together, by analysing our social relations, diagnosing their failings, and articulating ideals to guide their revision. But these profound social changes have left scarcely a dent in the model of social relations that most (analytical) political philosophers assume.

In this talk, Judith Simon will first turn to the question whether we can sensibly talk about trust in AI systems. Proposing a socio-technical view on AI, she will argue that we can trust AI systems, if we conceive them as systems consisting of networks of technologies and human actors, but that we should trust them if and only if they are trustworthy. Simon will conclude her talk by outlining some epistemic and ethical requirements for trustworthy systems and two caveats.

Human mistakes are inevitable, but fortunately heterogenous. Not so with machine decision-making. Using the same machine learning model for high-stakes decisions in many settings amplifies the strengths, weaknesses, biases, and idiosyncrasies of the original model. When the same person re-encounters the same model again and again, or models trained on the same dataset, she might be wrongly rejected again and again.

In this paper, we defend what we call the Interpretability Thesis which states that, in many contexts, decision-makers are morally obligated to avoid basing their decisions about how to treat decision-subjects on the outputs of non-interpretable ("black box") algorithmic decision systems. Others have defended this thesis, typically by arguing that we have duties of transparency to decision-subjects which require us to make certain information available to them. However, this approach to defending the interpretability thesis has been met with skepticism with skeptics worrying about the grounds of these duties of transparency and concerned that we hold algorithmic decisions to higher standards than human decision systems which also fail to meet duties of transparency.

Big tech “exploits” us. This has become a common refrain among critics of digital platforms. It gives voice to a shared sense that technology firms are somehow mistreating people—taking advantage of us, extracting from us—in a way that other data-driven harms, such as surveillance and algorithmic bias, fail to capture. But what does “exploitation” entail, exactly, and how do platforms perpetrate it? What would a theory of digital exploitation add to existing discussions about platform governance?

When firms contract with government agencies to provide services, they regularly assert that some subset of their work is proprietary and confidential. At the same time, public agencies are also subject to transparency requirements. Within the State of Washington, agencies are subject to a strongly transparent Public Records Act, the state's freedom of information law, under which members of the public are granted access to a large share of government information by request. Public agencies also seek access to firms' data to advance accountability, equity, and oversight objectives. In both respects, data access is constrained in practice when firms assert it to be trade secret. Specifically, I analyze two public-private data sharing relationships as a site of contestation over data access and control.

As the artificial intelligence and deep learning revolutions have swept over a number of industries, medicine has stood out as a prime area for beneficial innovation. The maturation of key areas of AI – computer vision, natural language processing, etc. – have led to their successive adoption in certain application areas of medicine. The field has seen thousands of researchers and companies begin pioneering new and creative ways of benefiting healthcare with AI. Here we’ll discuss two vitally important areas – therapeutics, and workflows.

Most of our online activities are, at least in part, powered by personalized recommender systems. While automatic pattern extraction as a technology holds great promise, it can also have alarming adverse impacts. This talk will give a high-level overview of common techniques for personalized recommender systems, and how they connect to problems on both the personal and social level. It will also discuss some alternative approaches addressing these issues, and why a solution cannot come from technology alone.

This talk will feature some preliminary thoughts regarding how the current regulatory paradigm in privacy law enables digital technology companies to “dodge” privacy regulations with which other companies offering similar services must comply, and exploring how this “dodge” results in unfair rules for companies and undermines privacy protection for people. Analyzing how the current privacy regime facilitates the dodge is important for diagnosing the shortcomings of existing laws, for revealing harmful effects on individuals and social institutions, and for developing effective alternatives. Such diagnosis gains urgency in light of the growing scholarly and policymaker consensus around privacy law reform.

Privacy engineering seeks to provide tools and methods to design privacy-preserving systems or patch privacy invasive ones. Obfuscation is one of the essential tools in the privacy engineering toolkit. But what can we learn from the plethora of methods and techniques that one may categorize as obfuscation? What can we learn from the role obfuscation plays in privacy engineering? In this talk, Ero Balsa will provide an overview of the two main reasons why privacy engineers resort to obfuscation: to enable people to protect themselves against unnecessarily privacy-invasive systems, and to modulate the level of exposure that providing utility to untrusted parties requires.

From July to January of 2020, Stanford Internet Observatory researchers worked with a coalition of researchers, government entities, tech companies, and civil society organizations in a multi-stakeholder partnership called the Election Integrity Partnership (EIP). Its mission was to rapidly detect high-velocity and potentially impactful false and misleading narratives related to voting. This talk will discuss findings from the partnership: how incidents became narratives, the rise of bottom-up misinformation, the dynamics of repeat spreaders, and the way in which platform policies shape message propagation.

Cryptographic proof of liabilities (PoL) is a cryptographic primitive to prove the size of funds a bank owes to its customers in a decentralized manner and can be used for solvency audits with better privacy guarantees. Most PoL schemes follow the same principle, i.e., a prover aggregates all of the user balances and enables users to verify balance inclusion in the reported total amount. This process is probabilistic and the more the users who verify inclusion the better the guarantee of a non-cheating prover. In this presentation, Yan Ji introduces generalized PoL, which was originally proposed for proving financial solvency, by extending the state-of-the-art PoL scheme with extra privacy features, and making it applicable to domains outside finance, including transparent and private donations, new algorithms for disapproval voting and negative reviews, and publicly verifiable COVID-19 cases.

This is an early-stage project about misinformation. While legal scholars have been active over the last 4 years identifying legal definitions of and developing legal responses to the problem of misinformation, including assessing the constitutionality of those responses under the current Supreme Court's First Amendment jurisprudence, less attention has been paid to how the law is already changing as a result of misinformation and how current legal doctrines and institutions are vulnerable to erosion because of misinformation already in the mix. This project brings together literature in sociology and social network theory about how information spreads and doctrinal standards used in judicial review of government action.

In information technology for development and related fields, action-oriented researchers aim to design and evaluate how technology can be used to improve the lives of underserved populations around the globe. However, improvement is a value-laden concept with normative, causal, and methodological assumptions. There are many alternative definitions that can be difficult to engage with and tempting for an action researcher to ignore. However, these definitions can heavily influence the direction, design, and evaluation of such work. In this presentation, Anthony Poon discusses some potential perspectives on improvement, including human development, empowerment, and post-development, and how they have influenced some of his past and current work.

DLI's inaugural debate was inspired by thinking through the provocations posed by the impact of ‘intelligent’ technologies on the future of human life. Will robots take over the planet? Will they undermine or erode what it means to be human in other more subtle or unanticipated ways? Is the preoccupation with intelligent machines a red herring? Or is the biggest threat posed by intelligent machines the affordances they provide to the humans who wield them?

Although an increasing number of ethical data science and AI courses is available, pedagogical approaches used in these courses rely exclusively on texts rather than on algorithmic development or data analysis. Technical students often consider these courses unimportant and a distraction from the “real” material. To develop instructional materials and methodologies that are thoughtful and engaging, we must strive for balance: between texts and coding, between critique and solution, and between cutting-edge research and practical applicability. In this talk, Julia Stoyanovich will discuss responsible data science courses that she has been developing and teaching to technical students at New York University since 2019, and will also speak to some ongoing work on teaching responsible data science to members of the public in a peer learning setting.

Can games make the world a better place? Is it possible that we use games to make a difference in global challenges such as climate change or public health? Can we reduce societal biases, or encourage people to intervene in situations of danger, such as sexual assault? And how do we know the games are doing what they set out to do?

This joint session is about the digital advertising ecosystem: we highlight some of its disturbing practices against users’ privacy, explain the puzzle of lack of GDPR enforcement over its clear data protection violations, provide a glimpse on how a major publisher with a significant ad operation, The New York Times, has been trying to safeguard the privacy of its readers without forgoing revenue, and conclude by looking ahead at the current conversations in the web standards community on how to build an ad ecosystem without ubiquitous tracking.

Design does more than supply the market with new products and services; it can raise provocations, critique existing socio-technical arrangements, seed conversations around matters of concern, and imagine radical alternatives. However, even when design is used as a critical provocation or political contestation, the focus is often on ‘making’ something new - a product, interface or artifact. That is because ‘unmaking’, a natural aspect of the designerly transformations always underway in the worlds around us, remains invisibilized and rarely theorized as its own explicit and intentional strategy.

Machine learning (ML) is at the core of many Internet services and operates on users’ personal information. The deciding metric for deploying ML models is often test performance, which measures if the models learned the given task well. Test performance, however, does not measure other important properties of ML models such as security vulnerabilities, privacy leakage and compliance with regulations.

In light of the US election, there will be no Digital Life Seminar scheduled for today. We look forward to seeing you next week!

Much of urban tech exploits today’s most ethically-charged technologies and business practices—such as indiscriminate location tracking, facial recognition, and gig work to fundamentally reprogram how urban systems function. As these failures become clearer, and broader awareness of systemic injustice in society grows, how can the emerging field of urban tech clarify choices between right and wrong?

How well can ordinary people do in identifying the veracity of news in real time? Using a unique research design that has involved crowdsourcing popular news articles from both mainstream and suspect news sources that have appeared in the past 24 hours to both ordinary citizens and professional fact checkers, Professor Tucker will report on the individual level characteristics of those likely to incorrectly identify false news stories as true, the results of interventions to attempt to reduce the prevalence of this behavior, and the prospects for crowdsourcing to serve as a viable means for identifying false news stories in real time.

Mechanism design is a form of optimization developed in economic theory. It casts economists as institutional engineers, choosing an outcome and then arranging a set of market rules and conditions to achieve it. In this paper, Lee McGuigan, Jake Goldenfein, and Salome Viljoen argue that mechanism design, applied in algorithmic environments, has become a tool for producing information domination, distributing social costs in ways that benefit designers, and controlling and coordinating participants in multi-sided platforms.

Modern software development has embraced the concept of "code reuse," which is the practice of relying on third-party code to avoid "reinventing the wheel" (and rightly so). While this practice saves developers time and effort, it also creates liabilities: the resulting app may behave in ways that the app developer does not anticipate. This can cause very serious issues for privacy compliance: while an app developer did not write all of the code in their app, they are nonetheless responsible for it. In this talk, I will present research that my group has conducted to automatically examine the privacy behaviors of mobile apps vis-à-vis their compliance with privacy regulations.

In this paper, we develop a model of COVID spread that uses dynamic mobility networks, derived from US cell phone data, to capture the hourly movements of millions of people from local neighborhoods (census block groups, or CBGs) to points of interest (POIs) such as restaurants, grocery stores, or religious establishments.

Software has eaten the world and crapped out a dystopia: a place where Abbot Labs uses copyright claims to stop people with diabetes from taking control over their insulin dispensing and where BMW is providing seat-heaters as an-over-the-air upgrade that you have to pay for by the month. Companies have tried this stuff since the year dot, but Thomas Edison couldn't send a patent enforcer to your house to make sure you honored the license agreement on your cylinder by only playing it on an Edison phonograph. Today, digital systems offer perfect enforcement for the pettiest, greediest grifts imaginable.

Drawing from my forthcoming book The Voice Catchers (Yale U Press, early 2021), I pose two key questions about this new development in the United States: How has the voice intelligence industry been able to gain the kind of social traction that has tens of millions of people giving their up voiceprints to so-called “intelligent assistants”? And in the face of this widespread shift to voice bio-profiling, what social policies should concerned citizens advocate to slow the process and implement regulations regarding this new form of surveillance?

Who bears responsibility for the real-world consequences of technology? This question has been unduly complicated for decades by the 1996 legislation that provides immunity from liability to platforms that host third-party content: Section 230 of the Communications Decency Act.

What happens when two different technologies, historically governed by different regulatory regimes, are combined into a single, hybrid, consumer device?

In this talk, Omid Pouraseed will discuss recent methods for adversarial data manipulation, and mention possible defense strategies against them. Although manipulations of visual and auditory media are as old as media themselves, the recent advent of deepfakes has marked a turning point in the creation of fake content.

Privacy is contextual. Everyday, we manage different contexts and adjust our privacy expectations accordingly. The theory of Contextual Integrity offers a way to capture contextual norms and a heuristic to analyze privacy. This analysis is especially helpful to detect situations in which the system designers take advantage of well-established, contextual privacy expectations, to encourage user disclosures without adhering to governing norms. For example, imagine an app that is marked to you as a patient/doctor communication tool in a medical context, yet it is actually an insurance company trying to get more information on you.

Diana will present her research on technology-mediated abuse in IPV, threat models, and recent work from Cornell Tech's Intimate Partner Violence clinic.

Over the past decade, algorithmic accountability has become an important concern for social scientists, computer scientists, journalists, and lawyers.

United States v. Apple: Competition in America examines the misunderstandings and exaggerations that firms have raised throughout antitrust history to justify collusion and monopoly.

Recommender systems have come to serve as the “homepage” for users to access informational items such videos, music, books, etc.

Supervised learning algorithms are increasingly operationalized in real-world decision-making systems. Unfortunately, the nature and desiderata of real-world tasks rarely fit neatly into the supervised learning contract.

For a decade and a half, Facebook has dominated the landscape of digital social networks and has evolved to become one of the most powerful arbiters of online speech.

Modern applications frequently require access to sensitive data, such as facial images, typing history, or health records, thereby increasing the need for expressive privacy protection.

A growing body of scholarship is addressing the risks of opaque analyses as well as the fear of hidden biases and discrimination that may come along with automated decision-making.

Over the last decade, behavioral science made significant progress and impact in academic research as well as impacted policy in commercial organizations and governments. At the same time, the rise of digital technologies and the digital economy provides exciting opportunities and presents challenges for the next decade of behavioral science. In this talk, Sobolev will explore novel avenues for behavioral science research in the digital economy.

Hill will discuss the ethics of building facial recognition databases that use the faces of people who have not consented to taking part.

​

​

​

​

​

​

​

​

​

​

​

​

​

​