Sarah H. Cen

Julien Perolat, Bart de Vylder, Daniel Hennes et al.

We introduce DeepNash, an autonomous agent capable of learning to play the imperfect information game Stratego from scratch, up to a human expert level. Stratego is one of the few iconic board games that Artificial Intelligence (AI) has not yet mastered. This popular game has an enormous game tree on the order of $10^{535}$ nodes, i.e., $10^{175}$ times larger than that of Go. It has the additional complexity of requiring decision-making under imperfect information, similar to Texas hold'em poker, which has a significantly smaller game tree (on the order of $10^{164}$ nodes). Decisions in Stratego are made over a large number of discrete actions with no obvious link between action and outcome. Episodes are long, with often hundreds of moves before a player wins, and situations in Stratego can not easily be broken down into manageably-sized sub-problems as in poker. For these reasons, Stratego has been a grand challenge for the field of AI for decades, and existing AI methods barely reach an amateur level of play. DeepNash uses a game-theoretic, model-free deep reinforcement learning method, without search, that learns to master Stratego via self-play. The Regularised Nash Dynamics (R-NaD) algorithm, a key component of DeepNash, converges to an approximate Nash equilibrium, instead of 'cycling' around it, by directly modifying the underlying multi-agent learning dynamics. DeepNash beats existing state-of-the-art AI methods in Stratego and achieved a yearly (2022) and all-time top-3 rank on the Gravon games platform, competing with human expert players.

Cindy Y. Zhang, Sarah H. Cen, Devavrat Shah

In recent years, multiple notions of algorithmic fairness have arisen. One such notion is individual fairness (IF), which requires that individuals who are similar receive similar treatment. In parallel, matrix estimation (ME) has emerged as a natural paradigm for handling noisy data with missing values. In this work, we connect the two concepts. We show that pre-processing data using ME can improve an algorithm's IF without sacrificing performance. Specifically, we show that using a popular ME method known as singular value thresholding (SVT) to pre-process the data provides a strong IF guarantee under appropriate conditions. We then show that, under analogous conditions, SVT pre-processing also yields estimates that are consistent and approximately minimax optimal. As such, the ME pre-processing step does not, under the stated conditions, increase the prediction error of the base algorithm, i.e., does not impose a fairness-performance trade-off. We verify these results on synthetic and real data.

Sarah H. Cen, Aleksander Madry, Devavrat Shah

People form judgments and make decisions based on the information that they observe. A growing portion of that information is not only provided, but carefully curated by social media platforms. Although lawmakers largely agree that platforms should not operate without any oversight, there is little consensus on how to regulate social media. There is consensus, however, that creating a strict, global standard of "acceptable" content is untenable (e.g., in the US, it is incompatible with Section 230 of the Communications Decency Act and the First Amendment). In this work, we propose that algorithmic filtering should be regulated with respect to a flexible, user-driven baseline. We provide a concrete framework for regulating and auditing a social media platform according to such a baseline. In particular, we introduce the notion of a baseline feed: the content that a user would see without filtering (e.g., on Twitter, this could be the chronological timeline). We require that the feeds a platform filters contain "similar" informational content as their respective baseline feeds, and we design a principled way to measure similarity. This approach is motivated by related suggestions that regulations should increase user agency. We present an auditing procedure that checks whether a platform honors this requirement. Notably, the audit needs only black-box access to a platform's filtering algorithm, and it does not access or infer private user information. We provide theoretical guarantees on the strength of the audit. We further show that requiring closeness between filtered and baseline feeds does not impose a large performance cost, nor does it create echo chambers.

Anish Agarwal, Sarah H. Cen, Devavrat Shah et al.

We propose a generalization of the synthetic controls and synthetic interventions methodology to incorporate network interference. We consider the estimation of unit-specific potential outcomes from panel data in the presence of spillover across units and unobserved confounding. Key to our approach is a novel latent factor model that takes into account network interference and generalizes the factor models typically used in panel data settings. We propose an estimator, Network Synthetic Interventions (NSI), and show that it consistently estimates the mean outcomes for a unit under an arbitrary set of counterfactual treatments for the network. We further establish that the estimator is asymptotically normal. We furnish two validity tests for whether the NSI estimator reliably generalizes to produce accurate counterfactual estimates. We provide a novel graph-based experiment design that guarantees the NSI estimator produces accurate counterfactual estimates, and also analyze the sample complexity of the proposed design. We conclude with simulations that corroborate our theoretical findings.

Sarah H. Cen, Hannah Ismael, Lucia Zheng

Evidence lies at the core of litigation, but it is increasingly difficult to obtain in AI-related disputes. Even when a claimant's position has merit, cases are often settled or dismissed because decisive facts are hidden inside proprietary models, platform logs, and protected databases. Grounding our discussion in past and ongoing cases, we investigate how asymmetries in access, resources, and expertise can create significant barriers to evidence in AI-related cases. We show how developers and deployers resist disclosure through various strategies challenging the value of the evidence to the requesting party and the cost of evidence production. From these patterns we identify seven recurring sources of asymmetry -- access to models, data, documentation, logs, expertise, compute, and infrastructure -- that reflect a broader pattern that we call the privatization of proof: when control over proof falls in the hands of private actors that can demand justification for access while ensuring that justification remains out of reach. We further argue that different types of access can be fungible: in the absence of a certain type of access (e.g., to model internals), one may be able to use alternative forms of access (e.g., sufficient compute, query access, and access to user logs) and to obtain a functionally equivalent amount of information. We propose a three-part test that can help resolve AI access disputes in litigation, drawing on concepts such as proportionality and reasonable alternatives. Our test relies on a few observations, including that the cause of action can provide a baseline for access.

Judy Hanwen Shen, Ken Liu, Angelina Wang et al.

Data transparency has emerged as a rallying cry for addressing concerns about AI: data quality, privacy, and copyright chief among them. Yet while these calls are crucial for accountability, current transparency policies often fall short of their intended aims. Similar to nutrition facts for food, policies aimed at nutrition facts for AI currently suffer from a limited consideration of research on effective disclosures. We offer an institutional perspective and identify three common fallacies in policy implementations of data disclosures for AI. First, many data transparency proposals exhibit a specification gap between the stated goals of data transparency and the actual disclosures necessary to achieve such goals. Second, reform attempts exhibit an enforcement gap between required disclosures on paper and enforcement to ensure compliance in fact. Third, policy proposals manifest an impact gap between disclosed information and meaningful changes in developer practices and public understanding. Informed by the social science on transparency, our analysis identifies affirmative paths for transparency that are effective rather than merely symbolic.

Sarah H. Cen, Andrew Ilyas, Hedi Driss et al.

The 2024 US presidential election is the first major contest to occur in the US since the popularization of large language models (LLMs). Building on lessons from earlier shifts in media (most notably social media's well studied role in targeted messaging and political polarization) this moment raises urgent questions about how LLMs may shape the information ecosystem and influence political discourse. While platforms have announced some election safeguards, how well they work in practice remains unclear. Against this backdrop, we conduct a large-scale, longitudinal study of 12 models, queried using a structured survey with over 12,000 questions on a near-daily cadence from July through November 2024. Our design systematically varies content and format, resulting in a rich dataset that enables analyses of the models' behavior over time (e.g., across model updates), sensitivity to steering, responsiveness to instructions, and election-related knowledge and "beliefs." In the latter half of our work, we perform four analyses of the dataset that (i) study the longitudinal variation of model behavior during election season, (ii) illustrate the sensitivity of election-related responses to demographic steering, (iii) interrogate the models' beliefs about candidates' attributes, and (iv) reveal the models' implicit predictions of the election outcome. To facilitate future evaluations of LLMs in electoral contexts, we detail our methodology, from question generation to the querying pipeline and third-party tooling. We also publicly release our dataset at https://huggingface.co/datasets/sarahcen/llm-election-data-2024

Sarah H. Cen, Andrew Ilyas, Aleksander Madry

Many human-facing algorithms -- including those that power recommender systems or hiring decision tools -- are trained on data provided by their users. The developers of these algorithms commonly adopt the assumption that the data generating process is exogenous: that is, how a user reacts to a given prompt (e.g., a recommendation or hiring suggestion) depends on the prompt and not on the algorithm that generated it. For example, the assumption that a person's behavior follows a ground-truth distribution is an exogeneity assumption. In practice, when algorithms interact with humans, this assumption rarely holds because users can be strategic. Recent studies document, for example, TikTok users changing their scrolling behavior after learning that TikTok uses it to curate their feed, and Uber drivers changing how they accept and cancel rides in response to changes in Uber's algorithm. Our work studies the implications of this strategic behavior by modeling the interactions between a user and their data-driven platform as a repeated, two-player game. We first find that user strategization can actually help platforms in the short term. We then show that it corrupts platforms' data and ultimately hurts their ability to make counterfactual decisions. We connect this phenomenon to user trust, and show that designing trustworthy algorithms can go hand in hand with accurate estimation. Finally, we provide a formalization of trustworthiness that inspires potential interventions.

Aspen Hopkins, Sarah H. Cen, Andrew Ilyas et al.

The widespread adoption of AI in recent years has led to the emergence of AI supply chains: complex networks of AI actors contributing models, datasets, and more to the development of AI products and services. AI supply chains have many implications yet are poorly understood. In this work, we take a first step toward a formal study of AI supply chains and their implications, providing two illustrative case studies indicating that both AI development and regulation are complicated in the presence of supply chains. We begin by presenting a brief historical perspective on AI supply chains, discussing how their rise reflects a longstanding shift towards specialization and outsourcing that signals the healthy growth of the AI industry. We then model AI supply chains as directed graphs and demonstrate the power of this abstraction by connecting examples of AI issues to graph properties. Finally, we examine two case studies in detail, providing theoretical and empirical results in both. In the first, we show that information passing (specifically, of explanations) along the AI supply chains is imperfect, which can result in misunderstandings that have real-world implications. In the second, we show that upstream design choices (e.g., by base model providers) have downstream consequences (e.g., on AI products fine-tuned on the base model). Together, our findings motivate further study of AI supply chains and their increasingly salient social, economic, regulatory, and technical implications.

Sarah H. Cen, Salil Goyal, Zaynah Javed et al.

AI audits play a critical role in AI accountability and safety. One branch of the law for which AI audits are particularly salient is anti-discrimination law. Several areas of anti-discrimination law implicate the "less discriminatory alternative" (LDA) requirement, in which a protocol (e.g., model) is defensible if no less discriminatory protocol that achieves comparable performance can be found with a reasonable amount of effort. Notably, the burden of proving an LDA exists typically falls on the claimant (the party alleging discrimination). This creates a significant hurdle in AI cases, as the claimant would seemingly need to train a less discriminatory yet high-performing model, a task requiring resources and expertise beyond most litigants. Moreover, developers often shield information about and access to their model and training data as trade secrets, making it difficult to reproduce a similar model from scratch. In this work, we present a procedure enabling claimants to determine if an LDA exists, even when they have limited compute, data, information, and model access. We focus on the setting in which fairness is given by demographic parity and performance by binary cross-entropy loss. As our main result, we provide a novel closed-form upper bound for the loss-fairness Pareto frontier (PF). We show how the claimant can use it to fit a PF in the "low-resource regime," then extrapolate the PF that applies to the (large) model being contested, all without training a single large model. The expression thus serves as a scaling law for loss-fairness PFs. To use this scaling law, the claimant would require a small subsample of the train/test data. Then, the claimant can fit the context-specific PF by training as few as 7 (small) models. We stress test our main result in simulations, finding that our scaling law holds even when the exact conditions of our theory do not.

Sarah H. Cen, Andrew Ilyas, Jennifer Allen et al.

Most modern recommendation algorithms are data-driven: they generate personalized recommendations by observing users' past behaviors. A common assumption in recommendation is that how a user interacts with a piece of content (e.g., whether they choose to "like" it) is a reflection of the content, but not of the algorithm that generated it. Although this assumption is convenient, it fails to capture user strategization: that users may attempt to shape their future recommendations by adapting their behavior to the recommendation algorithm. In this work, we test for user strategization by conducting a lab experiment and survey. To capture strategization, we adopt a model in which strategic users select their engagement behavior based not only on the content, but also on how their behavior affects downstream recommendations. Using a custom music player that we built, we study how users respond to different information about their recommendation algorithm as well as to different incentives about how their actions affect downstream outcomes. We find strong evidence of strategization across outcome metrics, including participants' dwell time and use of "likes." For example, participants who are told that the algorithm mainly pays attention to "likes" and "dislikes" use those functions 1.9x more than participants told that the algorithm mainly pays attention to dwell time. A close analysis of participant behavior (e.g., in response to our incentive conditions) rules out experimenter demand as the main driver of these trends. Further, in our post-experiment survey, nearly half of participants self-report strategizing "in the wild," with some stating that they ignore content they actually like to avoid over-recommendation of that content in the future. Together, our findings suggest that user strategization is common and that platforms cannot ignore the effect of their algorithms on user behavior.

Sarah H. Cen, Devavrat Shah

Making an informed decision -- for example, when choosing a career or housing -- requires knowledge about the available options. Such knowledge is generally acquired through costly trial and error, but this learning process can be disrupted by competition. In this work, we study how competition affects the long-term outcomes of individuals as they learn. We build on a line of work that models this setting as a two-sided matching market with bandit learners. A recent result in this area states that it is impossible to simultaneously guarantee two natural desiderata: stability and low optimal regret for all agents. Resource-allocating platforms can point to this result as a justification for assigning good long-term outcomes to some agents and poor ones to others. We show that this impossibility need not hold true. In particular, by modeling two additional components of competition -- namely, costs and transfers -- we prove that it is possible to simultaneously guarantee four desiderata: stability, low optimal regret, fairness in the distribution of regret, and high social welfare.

Sarah H. Cen, Paul Newman

Radar detects stable, long-range objects under variable weather and lighting conditions, making it a reliable and versatile sensor well suited for ego-motion estimation. In this work, we propose a radar-only odometry pipeline that is highly robust to radar artifacts (e.g., speckle noise and false positives) and requires only one input parameter. We demonstrate its ability to adapt across diverse settings, from urban UK to off-road Iceland, achieving a scan matching accuracy of approximately 5.20 cm and 0.0929 deg when using GPS as ground truth (compared to visual odometry's 5.77 cm and 0.1032 deg). We present algorithms for keypoint extraction and data association, framing the latter as a graph matching optimization problem, and provide an in-depth system analysis.