Large language models (LMs) often struggle to generate diverse, human-like creative content, raising concerns about the long-term homogenization of human thought through repeated exposure to similar outputs. Yet scalable methods for evaluating LM output diversity remain limited, especially beyond narrow tasks such as random number or name generation, or beyond repeated sampling from a single model. To address this gap, we introduce Infinity-Chat, a large-scale dataset of 26K diverse, real-world, open-ended user queries that admit a wide range of plausible answers with no single ground truth. We introduce the first comprehensive taxonomy for characterizing the full spectrum of open-ended prompts posed to LMs, comprising 6 top-level categories (e.g., creative content generation, brainstorm & ideation) that further breaks down to 17 subcategories. Using Infinity-Chat, we present a large-scale study of mode collapse in LMs, revealing a pronounced Artificial Hivemind effect in open-ended generation of LMs, characterized by (1) intra-model repetition, where a single model consistently generates similar responses, and more so (2) inter-model homogeneity, where different models produce strikingly similar outputs. Infinity-Chat also includes 31,250 human annotations, across absolute ratings and pairwise preferences, with 25 independent human annotations per example. This enables studying collective and individual-specific human preferences in response to open-ended queries. Our findings show that state-of-the-art LMs, reward models, and LM judges are less well calibrated to human ratings on model generations that elicit differing idiosyncratic annotator preferences, despite maintaining comparable overall quality. Overall, INFINITY-CHAT presents the first large-scale resource for systematically studying real-world open-ended queries to LMs, revealing critical insights to guide future research for mitigating long-term AI safety risks posed by the Artificial Hivemind.

Knowledge workers often need to extract and analyze information from a collection of documents to solve complex information tasks in the workplace, e.g., hiring managers reviewing resumes or analysts assessing risk in contracts. However, foraging for relevant information can become tedious and repetitive over many documents and criteria of interest. We introduce Marco, a mixed-initiative workspace supporting sensemaking over diverse business document collections. Through collection-centric assistance, Marco reduces the cognitive costs of extracting and structuring information, allowing users to prioritize comparative synthesis and decision making processes. Users interactively communicate their information needs to an AI assistant using natural language and compose schemas that provide an overview of a document collection. Findings from a usability study (n=16) demonstrate that when using Marco, users complete sensemaking tasks 16% more quickly, with less effort, and without diminishing accuracy. A design probe with seven domain experts identifies how Marco can benefit various real-world workflows.

Benjamin Newman, Yoonjoo Lee, Aakanksha Naik, Pao Siangliulue, Raymond Fok, Juho Kim, Daniel S Weld, Joseph Chee Chang, Kyle Lo. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

Navigating the vast scientific literature often starts with browsing a paper’s abstract. However, when a reader seeks additional information, not present in the abstract, they face a costly cognitive chasm during their dive into the full text. To bridge this gap, we introduce recursively expandable abstracts, a novel interaction paradigm that dynamically expands abstracts by progressively incorporating additional information from the papers’ full text. This lightweight interaction allows scholars to specify their information needs by quickly brushing over the abstract or selecting AI-suggested expandable entities. Relevant information is synthesized using a retrieval-augmented generation approach, presented as a fluid, threaded expansion of the abstract, and made efficiently verifiable via attribution to relevant source-passages in the paper. Through a series of user studies, we demonstrate the utility of recursively expandable abstracts and identify future opportunities to support low-effort and just-in-time exploration of long-form information contexts through LLM-powered interactions.

Scholars need to keep up with an exponentially increasing flood of scientific papers. To aid this challenge, we introduce Scim , a novel intelligent interface that helps scholars skim papers to rapidly review and gain a cursory understanding of its contents. Scim supports the skimming process by highlighting salient content within a paper, directing a scholar’s attention. These automatically-extracted highlights are faceted by content type, evenly-distributed across a paper, and have a density configurable by scholars. We evaluate Scim with an in-lab usability study and a longitudinal diary study, revealing how its highlights facilitate the more efficient construction of a conceptualization of a paper. Finally, we describe the process of scaling highlights from their conception within Scim , a research prototype, to production on over 521,000 papers within the Semantic Reader, a publicly-available augmented reading interface for scientific papers. We conclude by discussing design considerations and tensions for the design of future skimming tools with augmented intelligence.

Many real-world applications (e.g., note taking, search) require extracting a sentence or paragraph from a document and showing that snippet to a human outside of the source document.Yet, users may find snippets difficult to understand as they lack context from the original document.In this work, we use language models to rewrite snippets from scientific documents to be read on their own.First, we define the requirements and challenges for this user-facing decontextualization task, such as clarifying where edits occur and handling references to other documents.Second, we propose a framework that decomposes the task into three stages: question generation, question answering, and rewriting.Using this framework, we collect gold decontextualizations from experienced scientific article readers.We then conduct a range of experiments across state-ofthe-art commercial and open-source language models to identify how to best provide missingbut-relevant information to models for our task.Finally, we develop QADECONTEXT, a simple prompting strategy inspired by our framework that improves over end-to-end prompting.We conclude with analysis that finds, while rewriting is easy, question generation and answering remain challenging for today's models. github.com/bnewm0609/qa-decontext

Scholars need to keep up with an exponentially increasing flood of scientific papers. To aid this challenge, we introduce Scim, a novel intelligent interface that helps experienced researchers skim – or rapidly review – a paper to attain a cursory understanding of its contents. Scim supports the skimming process by highlighting salient paper contents in order to direct a reader’s attention. The system’s highlights are faceted by content type, evenly distributed across a paper, and have a density configurable by readers at both the global and local level. We evaluate Scim with both an in-lab usability study and a longitudinal diary study, revealing how its highlights facilitate the more efficient construction of a conceptualization of a paper. We conclude by discussing design considerations and tensions for the design of future intelligent skimming tools.

We present the first large-scale longitudinal analysis of missing label accessibility failures in Android apps. We developed a crawler and collected monthly snapshots of 312 apps over 16 months. We use this unique dataset in empirical examinations of accessibility not possible in prior datasets. Key large-scale findings include missing label failures in 55.6% of unique image-based elements, longitudinal improvement in ImageButton elements but not in more prevalent ImageView elements, that 8.8% of unique screens are unreachable without navigating at least one missing label failure, that app failure rate does not improve with number of downloads, and that effective labeling is neither limited to nor guaranteed by large software organizations. We then examine longitudinal data in individual apps, presenting illustrative examples of accessibility impacts of systematic improvements, incomplete improvements, interface redesigns, and accessibility regressions. We discuss these findings and potential opportunities for tools and practices to improve label-based accessibility.

How do sidewalks change over time? Are there geographic or socioeconomic patterns to this change? These questions are important but difficult to address with current GIS tools and techniques. In this demo paper, we introduce three preliminary crowd+AI (Artificial Intelligence) prototypes to track changes in street intersection accessibility over time—specifically, curb ramps—and report on results from a pilot usability study.

Many researchers motivate explainable AI with studies showing that human-AI team performance on decision-making tasks improves when the AI explains its recommendations. However, prior studies observed improvements from explanations only when the AI, alone, outperformed both the human and the best team. Can explanations help lead to complementary performance, where team accuracy is higher than either the human or the AI working solo? We conduct mixed-method user studies on three datasets, where an AI with accuracy comparable to humans helps participants solve a task (explaining itself in some conditions). While we observed complementary improvements from AI augmentation, they were not increased by explanations. Rather, explanations increased the chance that humans will accept the AI’s recommendation, regardless of its correctness. Our result poses new challenges for human-centered AI: Can we develop explanatory approaches that encourage appropriate trust in AI, and therefore help generate (or improve) complementary performance?

Despite the central importance of research papers to scientific progress, they can be difficult to read. Comprehension is often stymied when the information needed to understand a passage resides somewhere else—in another section, or in another paper. In this work, we envision how interfaces can bring definitions of technical terms and symbols to readers when and where they need them most. We introduce ScholarPhi, an augmented reading interface with four novel features: (1) tooltips that surface position-sensitive definitions from elsewhere in a paper, (2) a filter over the paper that “declutters” it to reveal how the term or symbol is used across the paper, (3) automatic equation diagrams that expose multiple definitions in parallel, and (4) an automatically generated glossary of important terms and symbols. A usability study showed that the tool helps researchers of all experience levels read papers. Furthermore, researchers were eager to have ScholarPhi’s definitions available to support their everyday reading.

Crowdsourcing is a common means of collecting image segmentation training data for use in a variety of computer vision applications. However, designing accurate crowd-powered image segmentation systems is challenging, because defining object boundaries in an image requires significant fine motor skills and hand-eye coordination, which makes these tasks error-prone. Typically, special segmentation tools are created and then answers from multiple workers are aggregated to generate more accurate results. However, individual tool designs can bias how and where people make mistakes, resulting in shared errors that remain even after aggregation. In this article, we introduce a novel crowdsourcing approach that leverages tool diversity as a means of improving aggregate crowd performance. Our idea is that given a diverse set of tools, answer aggregation done across tools can help improve the collective performance by offsetting systematic biases induced by the individual tools themselves. To demonstrate the effectiveness of the proposed approach, we design four different tools and present FourEyes, a crowd-powered image segmentation system that uses aggregation across different tools. We then conduct a series of studies that evaluate different aggregation conditions and show that using multiple tools can significantly improve aggregate accuracy. Furthermore, we investigate the idea of applying post-processing for multi-tool aggregation in terms of correction mechanism. We introduce a novel region-based method for synthesizing more accurate bounds for image segmentation tasks through averaging surrounding annotations. In addition, we explore the effect of adjusting the threshold parameter of an EM-based aggregation method. Our results suggest that not only the individual tool’s design, but also the correction mechanism, can affect the performance of multi-tool aggregation. This article extends a work presented at ACM IUI 2018 [46] by providing a novel region-based error-correction method and additional in-depth evaluation of the proposed approach.

Mobile, wearable, and other ubiquitous computing devices are increasingly creating a context in which conventional keyboard and screen-based inputs are being replaced in favor of more natural speech-based interactions. Digital personal assistants use speech to control a wide range of functionality, from environmental controls to information access. However, many deaf and hard-of-hearing users have speech patterns that vary from those of hearing users due to incomplete acoustic feedback from their own voices. Because automatic speech recognition (ASR) systems are largely trained using speech from hearing individuals, speech-controlled technologies are typically inaccessible to deaf users. Prior work has focused on providing deaf users access to aural output via real-time captioning or signing, but little has been done to improve users' ability to provide input to these systems' speech-based interfaces. Further, the vocalization patterns of deaf speech often make accurate recognition intractable for both automated systems and human listeners, making traditional approaches to mitigate ASR limitations, such as human captionists, less effective. To bridge this accessibility gap, we investigate the limitations of common speech recognition approaches and techniques---both automatic and human-powered---when applied to deaf speech. We then explore the effectiveness of an iterative crowdsourcing workflow, and characterize the potential for groups to collectively exceed the performance of individuals. This paper contributes a better understanding of the challenges of deaf speech recognition and provides insights for future system development in this space.

Crowdsourcing is a common means of collecting image segmentation training data for use in a variety of computer vision applications. However, designing accurate crowd-powered image segmentation systems is challenging because defining object boundaries in an image requires significant fine motor skills and hand-eye coordination, which makes these tasks error-prone. Typically, special segmentation tools are created and then answers from multiple workers are aggregated to generate more accurate results. However, individual tool designs can bias how and where people make mistakes, resulting in shared errors that remain even after aggregation. In this paper, we introduce a novel crowdsourcing workflow that leverages multiple tools for the same task to increase output accuracy by reducing systematic error biases introduced by the tools themselves. When a task can no longer be broken down into more-tractable subtasks (the conventional approach taken by microtask crowdsourcing), our multi-tool approach can be used to further improve accuracy by assigning different tools to different workers. We present a series of studies that evaluate our multi-tool approach and show that it can significantly improve aggregate accuracy in semantic image segmentation.

Email is more than just a communication medium. Email serves as an external memory for people---it contains our reservation numbers, meeting details, phone numbers, and more. Often, people need access to this information while on the go, which is cumbersome from mobile devices with limited I/O bandwidth. In this paper, we introduce WearMail, a conversational interface to retrieve specific information in email. WearMail is mostly automated but is made robust to information extraction tasks via a novel privacy-preserving human computation workflow. In WearMail, crowdworkers never have direct access to emails, but rather (i) generate an email filter to help the system find messages that may contain the desired information, and (ii) generate examples of the requested information that are then used to create custom, low-level information extractors that run automatically within the set of filtered emails. We explore the impact of varying levels of obfuscation on result quality, demonstrating that workers are able to deal with highly-obfuscated information nearly as well as with the original. WearMail introduces general mechanisms that let the crowd search and select private data without having direct access to the data itself.