We conduct an independent, third-party audit for bias of LinkedIn's Talent Search ranking system, focusing on potential ranking bias across two attributes: gender and race. To do so, we first construct a dataset of rankings produced by the system, collecting extensive Talent Search results across a diverse set of occupational queries. We then develop a robust labeling pipeline that infers the two demographic attributes of interest for the returned users. To evaluate potential biases in the collected dataset of real-world rankings, we utilize two exposure disparity metrics: deviation from group proportions and MinSkew@k. Our analysis reveals an under-representation of minority groups in early ranks across many queries. We further examine potential causes of this disparity, and discuss why they may be difficult or, in some cases, impossible to fully eliminate among the early ranks of queries. Beyond static metrics, we also investigate the concept of subgroup fairness over time, highlighting \emph{temporal disparities} in exposure and retention, which are often more difficult to audit for in practice. In employer recruiting platforms such as LinkedIn Talent Search, the persistence of a particular candidate over multiple days in the ranking can directly impact the probability that the given candidate is selected for opportunities. Our analysis reveals demographic disparities in this temporal stability, with some groups experiencing greater volatility in their ranked positions than others. We contextualize all our findings alongside LinkedIn’s published self-audits of its Talent Search system and reflect on the methodological constraints of a black-box external evaluation, including limited observability and noisy demographic inference. Our work contributes empirical insights and practical guidance for conducting third-party audits of modern socio-technical systems which go beyond the well-studied and standard algorithmic fairness guarantees of predictors.

Retrieval-Augmented Generation (RAG) enhances Large Language Models by grounding their outputs in external documents. These systems, however, remain vulnerable to attacks on the retrieval corpus, such as prompt injection. RAG-based search systems (e.g., Google’s Search AI Overview) present an interesting setting for studying and protecting against such threats, as defense algorithms can benefit from built-in reliability signals—like document ranking—and represent a non-LLM challenge for the adversary due to decades of work to thwart SEO. Motivated by, but not limited to, this scenario, this work introduces ReliabilityRAG, a framework for adversarial robustness that explicitly leverages reliability information of retrieved documents. Our first contribution adopts a graph-theoretic perspective to identify a ``consistent majority'' among retrieved documents to filter out malicious ones. We introduce a novel algorithm based on finding a Maximum Independent Set (MIS) on a document graph where edges encode contradiction. Our MIS variant explicitly prioritizes higher-reliability documents and provides provable robustness guarantees against bounded adversarial corruption under natural assumptions. Recognizing the computational cost of exact MIS for large retrieval sets, our second contribution is a scalable weighted sample and aggregate framework. It explicitly utilizes reliability information, preserving some robustness guarantees while efficiently handling many documents. We present empirical results showing ReliabilityRAG provides superior robustness against adversarial attacks compared to prior methods, maintains high benign accuracy, and excels in long-form generation tasks where prior robustness-focused methods struggled. Our work is a significant step towards more effective, provably robust defenses against retrieved corpus corruption in RAG.

Recent reports claim that large language models (LLMs) now outperform elite humans in competitive programming. Drawing on knowledge from a group of medalists in international algorithmic contests, we revisit this claim, examining how LLMs differ from human experts and where limitations still remain. We introduce LiveCodeBench Pro, a benchmark composed of problems from Codeforces, ICPC, and IOI that are continuously updated to reduce the likelihood of data contamination. A team of Olympiad medalists annotates every problem for algorithmic categories and conducts a line-by-line analysis of failed model-generated submissions. Using this new data and benchmark, we find that frontier models still have significant limitations: without external tools, the best model achieves only 53\% pass@1 on medium-difficulty problems and 0\% on hard problems, domains where expert humans still excel. We also find that LLMs succeed at implementation-heavy problems but struggle with nuanced algorithmic reasoning and complex case analysis, often generating confidently incorrect justifications. High performance appears largely driven by implementation precision and tool augmentation, not superior reasoning. LiveCodeBench Pro thus highlights the significant gap to human grandmaster levels, while offering fine-grained diagnostics to steer future improvements in code-centric LLM reasoning.

Recently, Meta has shifted towards AI-mediated ad targeting mechanisms that do not require advertisers to provide detailed targeting criteria. The shift is likely driven by excitement over AI capabilities as well as the need to address new data privacy policies and targeting changes agreed upon in civil rights settlements. At the same time, in response to growing public concern about the harms of targeted advertising, Meta has touted their ad preference controls as an effective mechanism for users to exert control over the advertising they see. Furthermore, Meta markets their "Why this ad" targeting explanation as a transparency tool that allows users to understand the reasons for seeing particular ads and inform their actions to control what ads they see in the future. Our study evaluates the effectiveness of Meta's "See less" ad control, as well as the actionability of ad targeting explanations following the shift to AI-mediated targeting. We conduct a large-scale study, randomly assigning participants the intervention of marking "See less" to either Body Weight Control or Parenting topics, and collecting the ads Meta shows to participants and their targeting explanations before and after the intervention. We find that utilizing the "See less" ad control for the topics we study does not significantly reduce the number of ads shown by Meta on these topics, and that the control is less effective for some users whose demographics are correlated with the topic. Furthermore, we find that the majority of ad targeting explanations for local ads made no reference to location-specific targeting criteria, and did not inform users why ads related to the topics they requested to "See less" of continued to be delivered. We hypothesize that the poor effectiveness of controls and lack of actionability and comprehensiveness in explanations are the result of the shift to AI-mediated targeting, for which explainability and transparency tools have not yet been developed by Meta. Our work thus provides evidence for the need of new methods for transparency and user control, suitable and reflective of how the increasingly complex and AI-mediated ad delivery systems operate.

Rankings are ubiquitous across many applications, from search engines to hiring committees. In practice, many rankings are derived from the output of predictors. However, when predictors trained for classification tasks have intrinsic uncertainty, it is not obvious how this uncertainty should be represented in the derived rankings. Our work considers ranking functions: maps from individual predictions for a classification task to distributions over rankings. We focus on two aspects of ranking functions: stability to perturbations in predictions and fairness towards both individuals and subgroups. Not only is stability an important requirement for its own sake, but --- as we show --- it composes harmoniously with individual fairness in the sense of Dwork et al. (2012). While deterministic ranking functions cannot be stable aside from trivial scenarios, we show that the recently proposed uncertainty aware (UA) ranking functions of Singh et al. (2021) are stable. Our main result is that UA rankings also achieve group fairness through successful composition with multiaccurate or multicalibrated predictors. Our work demonstrates that UA rankings naturally interpolate between group and individual level fairness guarantees, while simultaneously satisfying stability guarantees important whenever machine-learned predictions are used.

Detailed targeting of advertisements has long been one of the core offerings of online platforms. Unfortunately, malicious advertisers have frequently abused such targeting features, with results that range from violating civil rights laws to driving division, polarization, and even social unrest. Platforms have often attempted to mitigate this behavior by removing targeting attributes deemed problematic, such as inferred political leaning, religion, or ethnicity. In this work, we examine the effectiveness of these mitigations by collecting data from political ads placed on Facebook in the lead up to the 2022 U.S. midterm elections. We show that major political advertisers circumvented these mitigations by targeting proxy attributes: seemingly innocuous targeting criteria that closely correspond to political and racial divides in American society. We introduce novel methods for directly measuring the skew of various targeting criteria to quantify their effectiveness as proxies, and then examine the scale at which those attributes are used. Our findings have crucial implications for the ongoing discussion on the regulation of political advertising and emphasize the urgency for increased transparency.

The prevalence and importance of algorithmic two-sided marketplaces has drawn attention to the issue of fairness in such settings. Algorithmic decisions are used in assigning students to schools, users to advertisers, and applicants to job interviews. These decisions should heed the preferences of individuals, and simultaneously be fair with respect to their merits (synonymous with fit, future performance, or need). Merits conditioned on observable features are always *uncertain*, a fact that is exacerbated by the widespread use of machine learning algorithms to infer merit from the observables. As our key contribution, we carefully axiomatize a notion of individual fairness in the two-sided marketplace setting which respects the uncertainty in the merits; indeed, it simultaneously recognizes uncertainty as the primary potential cause of unfairness and an approach to address it. We design a linear programming framework to find fair utility-maximizing distributions over allocations, and we show that the linear program is robust to perturbations in the estimated parameters of the uncertain merit distributions, a key property in combining the approach with machine learning techniques.

Social media platforms curate access to information and opportunities, and so play a critical role in shaping public discourse today. The opaque nature of the algorithms these platforms use to curate content raises societal questions. Prior studies have used black-box methods led by experts or collaborative audits driven by everyday users to show that these algorithms can lead to biased or discriminatory outcomes. However, existing auditing methods face fundamental limitations because they function independent of the platforms. Concerns of potential harmful outcomes have prompted proposal of legislation in both the U.S. and the E.U. to mandate a new form of auditing where vetted external researchers get privileged access to social media platforms. Unfortunately, to date there have been no concrete technical proposals to provide such auditing, because auditing at scale risks disclosure of users' private data and platforms' proprietary algorithms. We propose a new method for platform-supported auditing that can meet the goals of the proposed legislation. The first contribution of our work is to enumerate the challenges and the limitations of existing auditing methods to implement these policies at scale. Second, we suggest that limited, privileged access to relevance estimators is the key to enabling generalizable platform-supported auditing of social media platforms by external researchers. Third, we show platform-supported auditing need not risk user privacy nor disclosure of platforms' business interests by proposing an auditing framework that protects against these risks. For a particular fairness metric, we show that ensuring privacy imposes only a small constant factor increase (6.34x as an upper bound, and 4× for typical parameters) in the number of samples required for accurate auditing. Our technical contributions, combined with ongoing legal and policy efforts, can enable public oversight into how social media platforms affect individuals and society by moving past the privacy-vs-transparency hurdle.

This panel brings experts together from law, philosophy, computer science and media studies to explore how digital platforms exercise power over which content is visible online, and which content is promoted to users, with a special focus on the use of algorithmic systems to achieve these ends.

We consider the problem of allocating divisible items among multiple agents, and consider the setting where any agent is allowed to introduce {\emph diversity constraints} on the items they are allocated. We motivate this via settings where the items themselves correspond to user ad slots or task workers with attributes such as race and gender on which the principal seeks to achieve demographic parity. We consider the following question: When an agent expresses diversity constraints into an allocation rule, is the allocation of other agents hurt significantly? If this happens, the cost of introducing such constraints is disproportionately borne by agents who do not benefit from diversity. We codify this via two desiderata capturing {\em robustness}. These are {\emph no negative externality} -- other agents are not hurt -- and {\emph monotonicity} -- the agent enforcing the constraint does not see a large increase in value. We show in a formal sense that the Nash Welfare rule that maximizes product of agent values is {\emph uniquely} positioned to be robust when diversity constraints are introduced, while almost all other natural allocation rules fail this criterion. We also show that the guarantees achieved by Nash Welfare are nearly optimal within a widely studied class of allocation rules. We finally perform an empirical simulation on real-world data that models ad allocations to show that this gap between Nash Welfare and other rules persists in the wild.

Political campaigns are increasingly turning to targeted advertising platforms to inform and mobilize potential voters. The appeal of these platforms stems from their promise to empower advertisers to select (or "target") users who see their messages with great precision, including through inferences about those users' interests and political affiliations. However, prior work has shown that the targeting may not work as intended, as platforms' ad delivery algorithms play a crucial role in selecting which subgroups of the targeted users see the ads. In particular, the platforms can selectively deliver ads to subgroups within the target audiences selected by advertisers in ways that can lead to demographic skews along race and gender lines, and do so without the advertiser's knowledge. In this work we demonstrate that ad delivery algorithms used by Facebook, the most advanced targeted advertising platform, shape the political ad delivery in ways that may not be beneficial to the political campaigns and to societal discourse. In particular, the ad delivery algorithms lead to political messages on Facebook being shown predominantly to people who Facebook thinks already agree with the ad campaign's message even if the political advertiser targets an ideologically diverse audience. Furthermore, an advertiser determined to reach ideologically non-aligned users is non-transparently charged a high premium compared to their more aligned competitor, a difference from traditional broadcast media. Our results demonstrate that Facebook exercises control over who sees which political messages beyond the control of those who pay for them or those who are exposed to them. Taken together, our findings suggest that the political discourse's increased reliance on profit-optimized, non-transparent algorithmic systems comes at a cost of diversity of political views that voters are exposed to. Thus, the work raises important questions of fairness and accountability desiderata for ad delivery algorithms applied to political ads.

Ad platforms such as Facebook, Google and LinkedIn promise value for advertisers through their targeted advertising. However, multiple studies have shown that ad delivery on such platforms can be skewed by gender or race due to hidden algorithmic optimization by the platforms, even when not requested by the advertisers. Building on prior work measuring skew in ad delivery, we develop a new methodology for black-box auditing of algorithms for discrimination in the delivery of job advertisements. Our first contribution is to identify the distinction between skew in ad delivery due to protected categories such as gender or race, from skew due to differences in qualification among people in the targeted audience. This distinction is important in U.S. law, where ads may be targeted based on qualifications, but not on protected categories. Second, we develop an auditing methodology that distinguishes between skew explainable by differences in qualifications from other factors, such as the ad platform’s optimization for engagement or training its algorithms on biased data. Our method controls for job qualification by comparing ad delivery of two concurrent ads for similar jobs, but for a pair of companies with different de facto gender distributions of employees. We describe the careful statistical tests that establish evidence of non-qualification skew in the results. Third, we apply our proposed methodology to two prominent targeted advertising platforms for job ads: Facebook and LinkedIn. We confirm skew by gender in ad delivery on Facebook, and show that it cannot be justified by differences in qualifications. We fail to find skew in ad delivery on LinkedIn. Finally, we suggest improvements to ad platform practices that could make external auditing of their algorithms in the public interest more feasible and accurate.

In this work, we study notions of fairness in decision-making systems when individuals have diverse preferences over the possible outcomes of the decisions. Our starting point is the seminal work of Dwork et al. [ITCS 2012] which introduced a notion of individual fairness (IF): given a task-specific similarity metric, every pair of individuals who are similarly qualified according to the metric should receive similar outcomes. We show that when individuals have diverse preferences over outcomes, requiring IF may unintentionally lead to less-preferred outcomes for the very individuals that IF aims to protect (e.g. a protected minority group). A natural alternative to IF is the classic notion of fair division, envy-freeness (EF): no individual should prefer another individual’s outcome over their own. Although EF allows for solutions where all individuals receive a highly-preferred outcome, EF may also be overly-restrictive for the decision-maker. For instance, if many individuals agree on the best outcome, then if any individual receives this outcome, they all must receive it, regardless of each individual’s underlying qualifications for the outcome. We introduce and study a new notion of preference-informed individual fairness (PIIF) that is a relaxation of both individual fairness and envy-freeness. At a high-level, PIIF requires that outcomes satisfy IF-style constraints, but allows for deviations provided they are in line with individuals' preferences. We show that PIIF can permit outcomes that are more favorable to individuals than any IF solution, while providing considerably more flexibility to the decision-maker than EF. In addition, we show how to efficiently optimize any convex objective over the outcomes subject to PIIF for a rich class of individual preferences. Finally, we demonstrate the broad applicability of the PIIF framework by extending our definitions and algorithms to the multiple-task targeted advertising setting introduced by Dwork and Ilvento [ITCS 2019].

We explore the power of the hybrid model of differential privacy (DP), in which some users desire the guarantees of the local model of DP and others are content with receiving the trusted-curator model guarantees. In particular, we study the utility of hybrid model estimators that compute the mean of arbitrary realvalued distributions with bounded support. When the curator knows the distribution’s variance, we design a hybrid estimator that, for realistic datasets and parameter settings, achieves a constant factor improvement over natural baselines.We then analytically characterize how the estimator’s utility is parameterized by the problem setting and parameter choices. When the distribution’s variance is unknown, we design a heuristic hybrid estimator and analyze how it compares to the baselines. We find that it often performs better than the baselines, and sometimes almost as well as the known-variance estimator. We then answer the question of how our estimator’s utility is affected when users’ data are not drawn from the same distribution, but rather from distributions dependent on their trust model preference. Concretely, we examine the implications of the two groups’ distributions diverging and show that in some cases, our estimators maintain fairly high utility. We then demonstrate how our hybrid estimator can be incorporated as a sub-component in more complex, higher-dimensional applications. Finally, we propose a new privacy amplification notion for the hybrid model that emerges due to interaction between the groups, and derive corresponding amplification results for our hybrid estimators.

The goals of learning from user data and preserving user privacy are often considered to be in conflict. This presentation will demonstrate that there are contexts when provable privacy guarantees can be an enabler for better web search and data mining (WSDM), and can empower researchers hoping to change the world by mining sensitive user data. The presentation starts by motivating the rigorous statistical data privacy definition that is particularly suitable for today's world of big data, differential privacy. It will then demonstrate how to achieve differential privacy for WSDM tasks when the data collector is trusted by the users. Using Chrome's deployment of RAPPOR as a case study, it will be shown that achieving differential privacy while preserving utility is feasible even when the data collector is not trusted. The presentation concludes with open problems and challenges for the WSDM community.

The enormous financial success of online advertising platforms is partially due to the precise targeting features they offer. Although researchers and journalists have found many ways that advertisers can target---or exclude---particular groups of users seeing their ads, comparatively little attention has been paid to the implications of the platform's ad delivery process, comprised of the platform's choices about which users see which ads. It has been hypothesized that this process can "skew" ad delivery in ways that the advertisers do not intend, making some users less likely than others to see particular ads based on their demographic characteristics. In this paper, we demonstrate that such skewed delivery occurs on Facebook, due to market and financial optimization effects as well as the platform's own predictions about the "relevance" of ads to different groups of users. We find that both the advertiser's budget and the content of the ad each significantly contribute to the skew of Facebook's ad delivery. Critically, we observe significant skew in delivery along gender and racial lines for "real" ads for employment and housing opportunities despite neutral targeting parameters. Our results demonstrate previously unknown mechanisms that can lead to potentially discriminatory ad delivery, even when advertisers set their targeting parameters to be highly inclusive. This underscores the need for policymakers and platforms to carefully consider the role of the ad delivery optimization run by ad platforms themselves---and not just the targeting choices of advertisers---in preventing discrimination in digital advertising.

Today an increasing number of consumer devices such as head phones, wearables, light bulbs and even baseball bats, are Bluetooth-enabled thanks to the widespread support of the technology by phone manufacturers and mobile operating system vendors. The ability for any device to seamlessly connect and exchange information with smartphones via Bluetooth Low Energy (BLE) protocol promises unlimited room for innovation. However, it also brings about new privacy challenges. We show that the BLE protocol together with the Bluetooth permission model implemented in the Android and iOS operating systems can be used for cross-app tracking unbeknownst to the individuals.

Randomized Aggregatable Privacy-Preserving Ordinal Response, or RAPPOR, is a technology for crowdsourcing statistics from end-user client software, anonymously, with strong privacy guarantees. In short, RAPPORs allow the forest of client data to be studied, without permitting the possibility of looking at individual trees. By applying randomized response in a novel manner, RAPPOR provides the mechanisms for such collection as well as for efficient, high-utility analysis of the collected data. In particular, RAPPOR permits statistics to be collected on the population of client-side strings with strong privacy guarantees for each client, and without linkability of their reports. This paper describes and motivates RAPPOR, details its differential-privacy and utility guarantees, discusses its practical deployment and properties in the face of different attack models, and, finally, gives results of its application to both synthetic and real-world data.

Most of what we understand about data sensitivity is through user self-report (e.g., surveys), this paper is the first to use behavioral data to determine content sensitivity, via the clues that users give as to what information they consider private or sensitive through their use of privacy enhancing product features. We perform a large-scale analysis of user anonymity choices during their activity on Quora, a popular question-and-answer site. We identify categories of questions for which users are more likely to exercise anonymity and explore several machine learning approaches towards predicting whether a particular answer will be written anonymously. Our findings validate the viability of the proposed approach towards an automatic assessment of data sensitivity, show that data sensitivity is a nuanced measure that should be viewed on a continuum rather than as a binary concept, and advance the idea that machine learning over behavioral data can be effectively used in order to develop product features that can help keep users safe.

With the recent surge of social networks such as Facebook, new forms of recommendations have become possible -- recommendations that rely on one's social connections in order to make personalized recommendations of ads, content, products, and people. Since recommendations may use sensitive information, it is speculated that these recommendations are associated with privacy risks. The main contribution of this work is in formalizing trade-offs between accuracy and privacy of personalized social recommendations. We study whether "social recommendations", or recommendations that are solely based on a user's social network, can be made without disclosing sensitive links in the social graph. More precisely, we quantify the loss in utility when existing recommendation algorithms are modified to satisfy a strong notion of privacy, called differential privacy. We prove lower bounds on the minimum loss in utility for any recommendation algorithm that is differentially private. We then adapt two privacy preserving algorithms from the differential privacy literature to the problem of social recommendations, and analyze their performance in comparison to our lower bounds, both analytically and experimentally. We show that good private social recommendations are feasible only for a small subset of the users in the social network or for a lenient setting of privacy parameters.

The question of how to publish an anonymized search log was brought to the forefront by a well-intentioned, but privacy-unaware AOL search log release. Since then a series of ad-hoc techniques have been proposed in the literature, though none are known to be provably private. In this paper, we take a major step towards a solution: we show how queries, clicks and their associated perturbed counts can be published in a manner that rigorously preserves privacy. Our algorithm is decidedly simple to state, but non-trivial to analyze. On the opposite side of privacy is the question of whether the data we can safely publish is of any use. Our findings offer a glimmer of hope: we demonstrate that a non-negligible fraction of queries and clicks can indeed be safely published via a collection of experiments on a real search log. In addition, we select an application, keyword generation, and show that the keyword suggestions generated from the perturbed data resemble those generated from the original data.

We consider a privacy threat to a social network in which the goal of an attacker is to obtain knowledge of a significant fraction of the links in the network. We formalize the typical social network interface and the information about links that it provides to its users in terms of lookahead. We consider a particular threat where an attacker subverts user accounts to get information about local neighborhoods in the network and pieces them together in order to get a global picture. We analyze, both experimentally and theoretically, the number of user accounts an attacker would need to subvert for a successful attack, as a function of his strategy for choosing users whose accounts to subvert and a function of lookahead provided by the network. We conclude that such an attack is feasible in practice, and thus any social network that wishes to protect the link privacy of its users should take great care in choosing the lookahead of its interface, limiting it to 1 or 2, whenever possible.

We consider a privacy threat to a social network in which the goal of an attacker is to obtain knowledge of a significant fraction of the links in the network. We formalize the typical social network interface and the information about links that it provides to its users in terms of lookahead. We consider a particular threat in which an attacker subverts user accounts to gain information about local neighborhoods in the network and pieces them together in order to build a global picture. We analyze, both experimentally and theoretically, the number of user accounts an attacker would need to subvert for a successful attack, as a function of his strategy for choosing users whose accounts to subvert and a function of the lookahead provided by the network. We conclude that such an attack is feasible in practice, and thus any social network that wishes to protect the link privacy of its users should take great care in choosing the lookahead of its interface, limiting it to 1 or 2, whenever possible.