Progress in AI is driven largely by the scale and quality of training data. Despite this, there is a deficit of empirical analysis examining the attributes of well-established datasets beyond text. In this work we conduct the largest and first-of-its-kind longitudinal audit across modalities --- popular text, speech, and video datasets --- from their detailed sourcing trends and use restrictions to their geographical and linguistic representation. Our manual analysis covers nearly 4000 public datasets between 1990-2024, spanning 608 languages, 798 sources, 659 organizations, and 67 countries. We find that multimodal machine learning applications have overwhelmingly turned to web-crawled, synthetic, and social media platforms, such as YouTube, for their training sets, eclipsing all other sources since 2019. Secondly, tracing the chain of dataset derivations we find that while less than 33% of datasets are restrictively licensed, over 80% of the source content in widely-used text, speech, and video datasets, carry non-commercial restrictions. Finally, counter to the rising number of languages and geographies represented in public AI training datasets, our audit demonstrates measures of relative geographical and multilingual representation have failed to significantly improve their coverage since 2013. We believe the breadth of our audit enables us to empirically examine trends in data sourcing, restrictions, and Western-centricity at an ecosystem-level, and that visibility into these questions are essential to progress in responsible AI. As a contribution to ongoing improvements in dataset transparency and responsible use, we release our entire multimodal audit, allowing practitioners to trace data provenance across text, speech, and video.

Hang Jiang, Xiajie Zhang, Robert Mahari, Daniel Kessler, Eric Ma, Tal August, Irene Li, Alex Pentland, Yoon Kim, Deb Roy, Jad Kabbara. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

We present the Legal Passage Retrieval Dataset, LePaRD.LePaRD contains millions of examples of U.S. federal judges citing precedent in context.The dataset aims to facilitate work on legal passage retrieval, a challenging practice-oriented legal retrieval and reasoning task.Legal passage retrieval seeks to predict relevant passages from precedential court decisions given the context of a legal argument.We extensively evaluate various approaches on LePaRD, and find that classification-based retrieval appears to work best.Our best models only achieve a recall of 59% when trained on data corresponding to the 10,000 most-cited passages, underscoring the difficulty of legal passage retrieval.By publishing LePaRD, we provide a large-scale and high quality resource to foster further research on legal passage retrieval.We hope that research on this practiceoriented NLP task will help expand access to justice by reducing the burden associated with legal research via computational assistance.

Tax returns contain 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 third-parties.Tax returns also contain sensitive data such that sharing them in their entirety undermines privacy.We outline how zero-knowledge cryptography may be applied to address this tension by allowing individuals and organizations to make provable claims about select information in their tax returns without revealing additional information, in a way that can be independently verified by third parties.We highlight key system goals and design specifications for this zero-knowledge tax disclosure system (zkTax) and present a prototype implementation.The prototype consists of three distinct services that can be distributed: a tax authority that provides signed tax documents; a Redact & Prove Service that enables users to redact tax documents and produce a zero-knowledge proof attesting the provenance of the redacted data; and a Verify Service to check the validity of claims.We demonstrate how zkTax could be implemented with minimal changes to existing tax infrastructure, allowing the system 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.

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.

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.

Data generated by users on digital platforms are a crucial resource for advocates and researchers interested in uncovering digital inequities, auditing algorithms, and understanding human behavior. Yet data access is often restricted. How can researchers both effectively and ethically collect user data? This paper shares an innovative approach to crowdsourcing user data to collect otherwise inaccessible Amazon purchase histories, spanning 5 years, from more than 5,000 U.S. users. We developed a data collection tool that prioritizes participant consent and includes an experimental study design. The design allows us to study multiple important aspects of privacy perception and user data sharing behavior, including how socio-demographics, monetary incentives and transparency can impact share rates. Experiment results (N=6,325) reveal both monetary incentives and transparency can significantly increase data sharing. Age, race, education, and gender also played a role, where female and less-educated participants were more likely to share. Our study design enables a unique empirical evaluation of the 'privacy paradox', where users claim to value their privacy more than they do in practice. We set up both real and hypothetical data sharing scenarios and find measurable similarities and differences in share rates across these contexts. For example, increasing monetary incentives had a 6 times higher impact on share rates in real scenarios. In addition, we study participants' opinions on how data should be used by various third parties, again finding that gender, age, education, and race have a significant impact. Notably, the majority of participants disapproved of government agencies using purchase data yet the majority approved of use by researchers. Overall, our findings highlight the critical role that transparency, incentive design, and user demographics play in ethical data collection practices, and provide guidance for future researchers seeking to crowdsource user generated data.