arXiv subjects

Will Epperson

Publications and source records attributed to Will Epperson.

11 recordsLinked to original sources

arXiv2026

CAVEAT: Towards Robust Computer-Use Agents in Incentive-Misaligned Environments

Computer-use agents (CUAs) increasingly act on behalf of users online. What happens when the environments they operate in have incentives that do not align with the user's? In online marketplaces, for example, platforms may favor some products over others, potentially steering agents away from the user's objective. Existing CUA benchmarks cover cooperative settings or explicit attacks, but do not test whether agents preserve user objectives when the environment itself has a stake in the outcome. We introduce CAVEAT, a controlled benchmark spanning nine marketplace environments and a taxonomy of eight common steering mechanisms. Across five model families, agents purchase the user-optimal product in 78.6% of matched-control episodes but only 17.3% when steering mechanisms are enabled. Larger models and increased reasoning improve robustness, but substantial failures persist. Our trajectory analysis and targeted ablations identify three points where steering enters the decision process: (1) agents distort the user's priorities, (2) prematurely narrow the set of alternatives they consider, and (3) commit before resolving decision-relevant evidence. Guided by this diagnosis, we develop CAVEAT-Harness, which directly targets these failure modes and raises user-optimal purchasing by 55.0%. Targeted post-training further improves a smaller open model. These results establish incentive robustness as a distinct challenge for delegated agents, diagnose how it fails, and show that targeted interventions can substantially improve it.

cs.AI↗

arXiv2025

Magentic Marketplace: An Open-Source Environment for Studying Agentic Markets

As LLM agents advance, they are increasingly mediating economic decisions, ranging from product discovery to transactions, on behalf of users. Such applications promise benefits but also raise many questions about agent accountability and value for users. Addressing these questions requires understanding how agents behave in realistic market conditions. However, previous research has largely evaluated agents in constrained settings, such as single-task marketplaces (e.g., negotiation) or structured two-agent interactions. Real-world markets are fundamentally different: they require agents to handle diverse economic activities and coordinate within large, dynamic ecosystems where multiple agents with opaque behaviors may engage in open-ended dialogues. To bridge this gap, we investigate two-sided agentic marketplaces where Assistant agents represent consumers and Service agents represent competing businesses. To study these interactions safely, we develop Magentic-Marketplace -- a simulated environment where Assistants and Services can operate. This environment enables us to study key market dynamics: the utility agents achieve, behavioral biases, vulnerability to manipulation, and how search mechanisms shape market outcomes. Our experiments show that frontier models can approach optimal welfare -- but only under ideal search conditions. Performance degrades sharply with scale, and all models exhibit severe first-proposal bias, creating 10-30x advantages for response speed over quality. These findings reveal how behaviors emerge across market conditions, informing the design of fair and efficient agentic marketplaces.

cs.MA↗

arXiv2025

Texture: Structured Exploration of Text Datasets

Exploratory analysis of a text corpus is essential for assessing data quality and developing meaningful hypotheses. Text analysis relies on understanding documents through structured attributes spanning various granularities of the documents such as words, phrases, sentences, topics, or clusters. However, current text visualization tools typically adopt a fixed representation tailored to specific tasks or domains, requiring users to switch tools as their analytical goals change. To address this limitation, we present Texture, a general-purpose interactive text exploration tool. Texture introduces a configurable data schema for representing text documents enriched with descriptive attributes. These attributes can appear at arbitrary levels of granularity in the text and possibly have multiple values, including document-level attributes, multi-valued attributes (e.g., topics), fine-grained span-level attributes (e.g., words), and vector embeddings. The system then combines existing interactive methods for text exploration into a single interface that provides attribute overview visualizations, supports cross-filtering attribute charts to explore subsets, uses embeddings for a dataset overview and similar instance search, and contextualizes filters in the actual documents. We evaluated Texture through a two-part user study with 10 participants from varied domains who each analyzed their own dataset in a baseline session and then with Texture. Texture was able to represent all of the previously derived dataset attributes, enabled participants to more quickly iterate during their exploratory analysis, and discover new insights about their data. Our findings contribute to the design of scalable, interactive, and flexible exploration systems that improve users' ability to make sense of text data.

cs.HC↗

arXiv2025

Over-Relying on Reliance: Towards Realistic Evaluations of AI-Based Clinical Decision Support

As AI-based clinical decision support (AI-CDS) is introduced in more and more aspects of healthcare services, HCI research plays an increasingly important role in designing for complementarity between AI and clinicians. However, current evaluations of AI-CDS often fail to capture when AI is and is not useful to clinicians. This position paper reflects on our work and influential AI-CDS literature to advocate for moving beyond evaluation metrics like Trust, Reliance, Acceptance, and Performance on the AI's task (what we term the "trap" of human-AI collaboration). Although these metrics can be meaningful in some simple scenarios, we argue that optimizing for them ignores important ways that AI falls short of clinical benefit, as well as ways that clinicians successfully use AI. As the fields of HCI and AI in healthcare develop new ways to design and evaluate CDS tools, we call on the community to prioritize ecologically valid, domain-appropriate study setups that measure the emergent forms of value that AI can bring to healthcare professionals.

cs.HC↗

arXiv2025

Interactive Debugging and Steering of Multi-Agent AI Systems

Fully autonomous teams of LLM-powered AI agents are emerging that collaborate to perform complex tasks for users. What challenges do developers face when trying to build and debug these AI agent teams? In formative interviews with five AI agent developers, we identify core challenges: difficulty reviewing long agent conversations to localize errors, lack of support in current tools for interactive debugging, and the need for tool support to iterate on agent configuration. Based on these needs, we developed an interactive multi-agent debugging tool, AGDebugger, with a UI for browsing and sending messages, the ability to edit and reset prior agent messages, and an overview visualization for navigating complex message histories. In a two-part user study with 14 participants, we identify common user strategies for steering agents and highlight the importance of interactive message resets for debugging. Our studies deepen understanding of interfaces for debugging increasingly important agentic workflows.

cs.MA↗

arXiv2024

Guided Statistical Workflows with Interactive Explanations and Assumption Checking

Statistical practices such as building regression models or running hypothesis tests rely on following rigorous procedures of steps and verifying assumptions on data to produce valid results. However, common statistical tools do not verify users' decision choices and provide low-level statistical functions without instructions on the whole analysis practice. Users can easily misuse analysis methods, potentially decreasing the validity of results. To address this problem, we introduce GuidedStats, an interactive interface within computational notebooks that encapsulates guidance, models, visualization, and exportable results into interactive workflows. It breaks down typical analysis processes, such as linear regression and two-sample T-tests, into interactive steps supplemented with automatic visualizations and explanations for step-wise evaluation. Users can iterate on input choices to refine their models, while recommended actions and exports allow the user to continue their analysis in code. Case studies show how GuidedStats offers valuable instructions for conducting fluid statistical analyses while finding possible assumption violations in the underlying data, supporting flexible and accurate statistical analyses.

cs.HC↗

arXiv2023

A Declarative Specification for Authoring Metrics Dashboards

Despite their ubiquity, authoring dashboards for metrics reporting in modern data analysis tools remains a manual, time-consuming process. Rather than focusing on interesting combinations of their data, users have to spend time creating each chart in a dashboard one by one. This makes dashboard creation slow and tedious. We conducted a review of production metrics dashboards and found that many dashboards contain a common structure: breaking down one or more metrics by different dimensions. In response, we developed a high-level specification for describing dashboards as sections of metrics repeated across the same dimensions and a graphical interface, Quick Dashboard, for authoring dashboards based on this specification. We present several usage examples that demonstrate the flexibility of this specification to create various kinds of dashboards and support a data-first approach to dashboard authoring.

cs.HC↗

arXiv2023

Dead or Alive: Continuous Data Profiling for Interactive Data Science

Profiling data by plotting distributions and analyzing summary statistics is a critical step throughout data analysis. Currently, this process is manual and tedious since analysts must write extra code to examine their data after every transformation. This inefficiency may lead to data scientists profiling their data infrequently, rather than after each transformation, making it easy for them to miss important errors or insights. We propose continuous data profiling as a process that allows analysts to immediately see interactive visual summaries of their data throughout their data analysis to facilitate fast and thorough analysis. Our system, AutoProfiler, presents three ways to support continuous data profiling: it automatically displays data distributions and summary statistics to facilitate data comprehension; it is live, so visualizations are always accessible and update automatically as the data updates; it supports follow up analysis and documentation by authoring code for the user in the notebook. In a user study with 16 participants, we evaluate two versions of our system that integrate different levels of automation: both automatically show data profiles and facilitate code authoring, however, one version updates reactively and the other updates only on demand. We find that both tools facilitate insight discovery with 91% of user-generated insights originating from the tools rather than manual profiling code written by users. Participants found live updates intuitive and felt it helped them verify their transformations while those with on-demand profiles liked the ability to look at past visualizations. We also present a longitudinal case study on how AutoProfiler helped domain scientists find serendipitous insights about their data through automatic, live data profiles. Our results have implications for the design of future tools that offer automated data analysis support.

cs.HC↗

arXiv2021

RECAST: Enabling User Recourse and Interpretability of Toxicity Detection Models with Interactive Visualization

With the widespread use of toxic language online, platforms are increasingly using automated systems that leverage advances in natural language processing to automatically flag and remove toxic comments. However, most automated systems -- when detecting and moderating toxic language -- do not provide feedback to their users, let alone provide an avenue of recourse for these users to make actionable changes. We present our work, RECAST, an interactive, open-sourced web tool for visualizing these models' toxic predictions, while providing alternative suggestions for flagged toxic language. Our work also provides users with a new path of recourse when using these automated moderation tools. RECAST highlights text responsible for classifying toxicity, and allows users to interactively substitute potentially toxic phrases with neutral alternatives. We examined the effect of RECAST via two large-scale user evaluations, and found that RECAST was highly effective at helping users reduce toxicity as detected through the model. Users also gained a stronger understanding of the underlying toxicity criterion used by black-box models, enabling transparency and recourse. In addition, we found that when users focus on optimizing language for these models instead of their own judgement (which is the implied incentive and goal of deploying automated models), these models cease to be effective classifiers of toxicity compared to human annotations. This opens a discussion for how toxicity detection models work and should work, and their effect on the future of online discourse.

cs.HC↗

arXiv2020

RECAST: Interactive Auditing of Automatic Toxicity Detection Models

As toxic language becomes nearly pervasive online, there has been increasing interest in leveraging the advancements in natural language processing (NLP), from very large transformer models to automatically detecting and removing toxic comments. Despite the fairness concerns, lack of adversarial robustness, and limited prediction explainability for deep learning systems, there is currently little work for auditing these systems and understanding how they work for both developers and users. We present our ongoing work, RECAST, an interactive tool for examining toxicity detection models by visualizing explanations for predictions and providing alternative wordings for detected toxic speech.

cs.CL↗

arXiv2019

FairVis: Visual Analytics for Discovering Intersectional Bias in Machine Learning

The growing capability and accessibility of machine learning has led to its application to many real-world domains and data about people. Despite the benefits algorithmic systems may bring, models can reflect, inject, or exacerbate implicit and explicit societal biases into their outputs, disadvantaging certain demographic subgroups. Discovering which biases a machine learning model has introduced is a great challenge, due to the numerous definitions of fairness and the large number of potentially impacted subgroups. We present FairVis, a mixed-initiative visual analytics system that integrates a novel subgroup discovery technique for users to audit the fairness of machine learning models. Through FairVis, users can apply domain knowledge to generate and investigate known subgroups, and explore suggested and similar subgroups. FairVis' coordinated views enable users to explore a high-level overview of subgroup performance and subsequently drill down into detailed investigation of specific subgroups. We show how FairVis helps to discover biases in two real datasets used in predicting income and recidivism. As a visual analytics system devoted to discovering bias in machine learning, FairVis demonstrates how interactive visualization may help data scientists and the general public understand and create more equitable algorithmic systems.

cs.LG↗