Search

Anonymity is an important principle online. However, malicious actors have long used misleading identities to conduct fraud, spread disinformation, and carry out other deceptive schemes. With the advent of increasingly capable AI, bad actors can amplify the potential scale and effectiveness of their operations, intensifying the challenge of balancing anonymity and trustworthiness online. In this paper, we analyze the value of a new tool to address this challenge: "personhood credentials" (PHCs), digital credentials that empower users to demonstrate that they are real people -- not AIs -- to online services, without disclosing any personal information. Such credentials can be issued by a range of trusted institutions -- governments or otherwise. A PHC system, according to our definition, could be local or global, and does not need to be biometrics-based. Two trends in AI contribute to the urgency of the challenge: AI's increasing indistinguishability from people online (i.e., lifelike content and avatars, agentic activity), and AI's increasing scalability (i.e., cost-effectiveness, accessibility). Drawing on a long history of research into anonymous credentials and "proof-of-personhood" systems, personhood credentials give people a way to signal their trustworthiness on online platforms, and offer service providers new tools for reducing misuse by bad actors. In contrast, existing countermeasures to automated deception -- such as CAPTCHAs -- are inadequate against sophisticated AI, while stringent identity verification solutions are insufficiently private for many use-cases. After surveying the benefits of personhood credentials, we also examine deployment risks and design challenges. We conclude with actionable next steps for policymakers, technologists, and standards bodies to consider in consultation with the public., Comment: 63 pages, 7 figures, 5 tables; minor additions to acknowledgments and wording changes for clarity; corrected typo

AI progress is creating a growing range of risks and opportunities, but it is often unclear how they should be navigated. In many cases, the barriers and uncertainties faced are at least partly technical. Technical AI governance, referring to technical analysis and tools for supporting the effective governance of AI, seeks to address such challenges. It can help to (a) identify areas where intervention is needed, (b) identify and assess the efficacy of potential governance actions, and (c) enhance governance options by designing mechanisms for enforcement, incentivization, or compliance. In this paper, we explain what technical AI governance is, why it is important, and present a taxonomy and incomplete catalog of its open problems. This paper is intended as a resource for technical researchers or research funders looking to contribute to AI governance., Comment: Ben Bucknall and Anka Reuel contributed equally and share the first author position

General-purpose artificial intelligence (AI) systems are built on massive swathes of public web data, assembled into corpora such as C4, RefinedWeb, and Dolma. To our knowledge, we conduct the first, large-scale, longitudinal audit of the consent protocols for the web domains underlying AI training corpora. Our audit of 14,000 web domains provides an expansive view of crawlable web data and how codified data use preferences are changing over time. We observe a proliferation of AI-specific clauses to limit use, acute differences in restrictions on AI developers, as well as general inconsistencies between websites' expressed intentions in their Terms of Service and their robots.txt. We diagnose these as symptoms of ineffective web protocols, not designed to cope with the widespread re-purposing of the internet for AI. Our longitudinal analyses show that in a single year (2023-2024) there has been a rapid crescendo of data restrictions from web sources, rendering ~5%+ of all tokens in C4, or 28%+ of the most actively maintained, critical sources in C4, fully restricted from use. For Terms of Service crawling restrictions, a full 45% of C4 is now restricted. If respected or enforced, these restrictions are rapidly biasing the diversity, freshness, and scaling laws for general-purpose AI systems. We hope to illustrate the emerging crises in data consent, for both developers and creators. The foreclosure of much of the open web will impact not only commercial AI, but also non-commercial AI and academic research., Comment: 41 pages (13 main), 5 figures, 9 tables

New capabilities in foundation models are owed in large part to massive, widely-sourced, and under-documented training data collections. Existing practices in data collection have led to challenges in tracing authenticity, verifying consent, preserving privacy, addressing representation and bias, respecting copyright, and overall developing ethical and trustworthy foundation models. In response, regulation is emphasizing the need for training data transparency to understand foundation models' limitations. Based on a large-scale analysis of the foundation model training data landscape and existing solutions, we identify the missing infrastructure to facilitate responsible foundation model development practices. We examine the current shortcomings of common tools for tracing data authenticity, consent, and documentation, and outline how policymakers, developers, and data creators can facilitate responsible foundation model development by adopting universal data provenance standards., Comment: ICML 2024 camera-ready version (Spotlight paper). 9 pages, 2 tables

In a world of increasing closed-source commercial machine learning models, model evaluations from developers must be taken at face value. These benchmark results-whether over task accuracy, bias evaluations, or safety checks-are traditionally impossible to verify by a model end-user without the costly or impossible process of re-performing the benchmark on black-box model outputs. This work presents a method of verifiable model evaluation using model inference through zkSNARKs. The resulting zero-knowledge computational proofs of model outputs over datasets can be packaged into verifiable evaluation attestations showing that models with fixed private weights achieve stated performance or fairness metrics over public inputs. We present a flexible proving system that enables verifiable attestations to be performed on any standard neural network model with varying compute requirements. For the first time, we demonstrate this across a sample of real-world models and highlight key challenges and design solutions. This presents a new transparency paradigm in the verifiable evaluation of private models.

Tax returns contain key financial information of interest to third parties: public officials are asked to share financial data for transparency, companies seek to assess the financial status of business partners, and individuals need to prove their income to landlords or to receive benefits. Tax returns also contain sensitive data such that sharing them in their entirety undermines privacy. We introduce a zero-knowledge tax disclosure system (zkTax) that allows individuals and organizations to make provable claims about select information in their tax returns without revealing additional information, which can be independently verified by third parties. The system consists of three distinct services that can be distributed: a tax authority provides tax documents signed with a public key; a Redact & Prove Service enables users to produce a redacted version of the tax documents with a zero-knowledge proof attesting the provenance of the redacted data; a Verify Service enables anyone to verify the proof. We implement a prototype with a user interface, compatible with U.S. tax forms, and demonstrate how this design could be implemented with minimal changes to existing tax infrastructure. Our system is designed to be extensible to other contexts and jurisdictions. This work provides a practical example of how distributed tools leveraging cryptography can enhance existing government or financial infrastructures, providing immediate transparency alongside privacy without system overhauls.

While the flexible capabilities of large language models (LLMs) allow them to answer a range of queries based on existing learned knowledge, information retrieval to augment generation is an important tool to allow LLMs to answer questions on information not included in pre-training data. Such private information is increasingly being generated in a wide array of distributed contexts by organizations and individuals. Performing such information retrieval using neural embeddings of queries and documents always leaked information about queries and database content unless both were stored locally. We present Private Retrieval Augmented Generation (PRAG), an approach that uses multi-party computation (MPC) to securely transmit queries to a distributed set of servers containing a privately constructed database to return top-k and approximate top-k documents. This is a first-of-its-kind approach to dense information retrieval that ensures no server observes a client's query or can see the database content. The approach introduces a novel MPC friendly protocol for inverted file approximate search (IVF) that allows for fast document search over distributed and private data in sublinear communication complexity. This work presents new avenues through which data for use in LLMs can be accessed and used without needing to centralize or forgo privacy.

Decentralized autonomous organizations (DAOs) are a new, rapidly-growing class of organizations governed by smart contracts. Here we describe how researchers can contribute to the emerging science of DAOs and other digitally-constituted organizations. From granular privacy primitives to mechanism designs to model laws, we identify high-impact problems in the DAO ecosystem where existing gaps might be tackled through a new data set or by applying tools and ideas from existing research fields such as political science, computer science, economics, law, and organizational science. Our recommendations encompass exciting research questions as well as promising business opportunities. We call on the wider research community to join the global effort to invent the next generation of organizations., Comment: includes major coordination problems

The physical environment you navigate strongly determines which communities and people matter most to individuals. These effects drive both personal access to opportunities and the social capital of communities, and can often be observed in the personal mobility traces of individuals. Traditional social media feeds underutilize these mobility-based features, or do so in a privacy exploitative manner. Here we propose a consent-first private information sharing paradigm for driving social feeds from users' personal private data, specifically using mobility traces. This approach designs the feed to explicitly optimize for integrating the user into the local community and for social capital building through leveraging mobility trace overlaps as a proxy for existing or potential real-world social connections, creating proportionality between whom a user sees in their feed, and whom the user is likely to see in person. These claims are validated against existing social-mobility data, and a reference implementation of the proposed algorithm is built for demonstration. In total, this work presents a novel technique for designing feeds that represent real offline social connections through private set intersections requiring no third party, or public data exposure.

News media has long been an ecosystem of creation, reproduction, and critique, where news outlets report on current events and add commentary to ongoing stories. Understanding the dynamics of news information creation and dispersion is important to accurately ascribe credit to influential work and understand how societal narratives develop. These dynamics can be modelled through a combination of information-theoretic natural language processing and networks; and can be parameterised using large quantities of textual data. However, it is challenging to see "the wood for the trees", i.e., to detect small but important flows of information in a sea of noise. Here we develop new comparative techniques to estimate temporal information flow between pairs of text producers. Using both simulated and real text data we compare the reliability and sensitivity of methods for estimating textual information flow, showing that a metric that normalises by local neighbourhood structure provides a robust estimate of information flow in large networks. We apply this metric to a large corpus of news organisations on Twitter and demonstrate its usefulness in identifying influence within an information ecosystem, finding that average information contribution to the network is not correlated with the number of followers or the number of tweets. This suggests that small local organisations and right-wing organisations which have lower average follower counts still contribute significant information to the ecosystem. Further, the methods are applied to smaller full-text datasets of specific news events across news sites and Russian troll accounts on Twitter. The information flow estimation reveals and quantifies features of how these events develop and the role of groups of trolls in setting disinformation narratives.

Contrary to expectations that the increased connectivity offered by the internet and particularly Online Social Networks (OSNs) would result in broad consensus on contentious issues, we instead frequently observe the formation of polarised echo chambers, in which only one side of an argument is entertained. These can progress to filter bubbles, actively filtering contrasting opinions, resulting in vulnerability to misinformation and increased polarisation on social and political issues. These have real-world effects when they spread offline, such as vaccine hesitation and violence. This work seeks to develop a better understanding of how echo chambers manifest in different discussions dealing with different issues over an extended period of time. We explore the activities of two groups of polarised accounts across three Twitter discussions in the Australian context. We found Australian Twitter accounts arguing against marriage equality in 2017 were more likely to support the notion that arsonists were the primary cause of the 2019/2020 Australian bushfires, and those supporting marriage equality argued against that arson narrative. We also found strong evidence that the stance people took on marriage equality in 2017 did not predict their political stance in discussions around the Australian federal election two years later. Although mostly isolated from each other, we observe that in certain situations the polarised groups may interact with the broader community, which offers hope that the echo chambers may be reduced with concerted outreach to members.

The modern age of digital music access has increased the availability of data about music consumption and creation, facilitating the large-scale analysis of the complex networks that connect music together. Data about user streaming behaviour, and the musical collaboration networks are particularly important with new data-driven recommendation systems. Without thorough analysis, such collaboration graphs can lead to false or misleading conclusions. Here we present a new collaboration network of artists from the online music streaming service Spotify, and demonstrate a critical change in the eigenvector centrality of artists, as low popularity artists are removed. The critical change in centrality, from classical artists to rap artists, demonstrates deeper structural properties of the network. A Social Group Centrality model is presented to simulate this critical transition behaviour, and switching between dominant eigenvectors is observed. This model presents a novel investigation of the effect of popularity bias on how centrality and importance are measured, and provides a new tool for examining such flaws in networks.

Contagion models are a primary lens through which we understand the spread of information over social networks. However, simple contagion models cannot reproduce the complex features observed in real-world data, leading to research on more complicated complex contagion models. A noted feature of complex contagion is social reinforcement that individuals require multiple exposures to information before they begin to spread it themselves. Here we show that the quoter model, a model of the social flow of written information over a network, displays features of complex contagion, including the weakness of long ties and that increased density inhibits rather than promotes information flow. Interestingly, the quoter model exhibits these features despite having no explicit social reinforcement mechanism, unlike complex contagion models. Our results highlight the need to complement contagion models with an information-theoretic view of information spreading to better understand how network properties affect information flow and what are the most necessary ingredients when modeling social behavior., Comment: 18 pages, 9 figures, 1 table

The number of characters in a movie is an interesting feature. However, it is non-trivial to measure directly. Naive metrics such as the number of credited characters vary wildly. Here, we show that a metric based on the notion of "ecological diversity" as expressed through a Shannon-entropy based metric can characterise the number of characters in a movie, and is useful in taxonomic classification. We also show how the metric can be generalised using Jensen-Shannon divergence to provide a measure of the similarity of characters appearing in different movies, for instance of use in recommender systems, e.g., Netflix. We apply our measures to the Marvel Cinematic Universe (MCU), and show what they teach us about this highly successful franchise of movies. In particular, these measures provide a useful predictor of "success" for films in the MCU, as well as a natural means to understand the relationships between the stories in the overall film arc.