Searcharxiv⌕ Search

arXiv subjects

Hoda Heidari

Publications and source records attributed to Hoda Heidari.

At least 19 recordsLinked to original sources

Between Rigor and Reality: How AI Safety Benchmark Developers Understand Benchmark Use and Maintenance

Safety benchmarks are cited as the primary tool to understand the risks posed by generative AI (GenAI) systems, yet growing evidence suggests they often fail to meet the needs of real-world safety evaluation. We present findings from five interviews with AI safety benchmark developers in academia and industry. We find a disconnect between perceptions of the intended usefulness of safety benchmarks and their practical usefulness in safety evaluations of deployed systems. Our participants offered perspectives on two challenges contributing to this disconnect: (1) the difficulty of achieving inter-rater reliability on safety constructs, and (2) a lack of clarity regarding how to address persistent threats to external validity. Based on our analysis, we argue that safety benchmarks must not only be grounded in deployment contexts, but actively integrated with them. To facilitate this, we outline pathways for more transparent communication between academic and industry AI safety stakeholders.

cs.AI↗

Why and How to Do Science When AI Is Better at It: A Framework Grounded in Human Understanding and Direction

As AI begins to match or exceed human researchers at core scientific tasks, two questions become pressing: \emph{why} should humans keep doing science, and \emph{how} should they use AI when they do? This paper argues that the second can be answered only by answering the first. It holds that one indispensable aim of science is \emph{durable, collective human understanding of the world, pursued in service of human ends and values}. From this principle it derives a two-part test for when a research task can be delegated to AI. The first part, \emph{human understanding}, requires that the scientific community remain able to explain, defend and contest the resulting claims, and retain the skills to do so. The second, \emph{human direction}, requires that judgments about which questions matter and what findings mean stay with accountable humans. Applying the test to each phase of research in the Contributor Roles Taxonomy (CRediT) yields recommendations that fields can tailor to their questions and methods. The framework differs from existing guidance in three ways. It rests on an account of what science is for rather than on integrity norms. It holds even if AI outperforms human researchers, rather than depending on AI's current limitations. And it judges each use of AI by its effects on human understanding and direction rather than by the type of tool, across every phase of research, including peer review and the training of new researchers. It closes with the reforms needed to keep science aimed at human understanding and direction.

cs.AI↗

On the Framing Effect of AI Red-Teaming Instructions: Evidence from Prior Exercises and Controlled Simulations

AI red-teaming, adversarially probing generative AI systems to surface harms, has become a central evaluation method. However, prior work has highlighted that red-teaming is a fairly open-ended task, laden with myriad design choices that can shape its findings. In this work, we focus on the instructions used to conduct red-teaming exercises and hypothesize that they can significantly affect outcomes in a manner analogous to the framing effect in psychology, where the format in which information is presented greatly impacts how people respond. To understand whether and why instructions impact red-teaming outcomes, first, we analyze the range of ways in which instructions have historically been formulated, reviewing 228 instances of prior work. Second, based on takeaways from this review, we run controlled simulations that hold parameters such as the risk domain, the attacker, and the model under investigation fixed while varying only the instructions. From five real instruction templates used by industry and academic organizations, we derive 45 instruction frames across three risk domains and three levels of risk granularity, generating 100 variants per frame, and deploy each against five open-weights target models, for 22,500 attacks. Using a generalized linear mixed model, we find that certain instruction properties significantly predict attack success rates: the lowest and highest levels of risk granularity both outperform intermediate specificity, as do short and long instructions relative to medium-length ones. The presence of examples is associated with higher attack success rates, while supplying explicit harm definitions yields lower ones. As alignment evaluations become increasingly automated, our results suggest that humans continue to shape red-teaming outcomes substantially through the instructions they provide to scaffold the activity.

cs.CY↗

Searching for "Harmful Refusal": A Psychometric Audit of an AI Safety Benchmark

Safety benchmarks typically report one overall score for a suite of datasets, each of which may target one or more safety-related attributes, so models with similar overall scores can have very different attribute profiles. Comparing models is more tractable at the level of individual attributes, yet it is often unclear whether even a single dataset's scores isolate any single attribute. One plausible candidate for such an attribute is harmful refusal, a model's tendency to refuse dangerous or policy-violating prompts. We examine whether it constitutes a single, measurable attribute in HELM Safety. Using a construct validity framework that stipulates that an attribute must exist before a test can measure it, we start with HELM Safety's four datasets that might plausibly target harmful refusal, but find that three are saturated. We subject the remaining dataset, HarmBench, to two psychometric tests to determine if a single attribute like harmful refusal could stand behind its score. First, multidimensional item response theory modeling strongly suggests that HarmBench does not measure a singular attribute. Second, a differential item functioning analysis finds items where models from different developers with the same refusal ability score differently. These flags largely disappear under scope-specific matching, a pattern consistent with aggregation effects but not sufficient to rule out domain-specific developer differences. Zooming out, HarmBench collapses distinct harm behaviors into one score, and the overall HELM safety aggregate further collapses HarmBench and scores from other datasets into a single top-line number. Any safety score that averages over datasets and items can hide saturation and conflate behaviors this way. We argue that a score should earn its single-attribute reading before models are compared with it.

cs.AI↗

A Framework for Generating Valid Context-Specific Benchmarks through Expert Guidance

This paper presents an end-to-end approach for generating context-specific large language model (LLM) benchmark datasets by combining expert input with synthetic data generation. Existing benchmark construction methods often trade off validity and scalability: datasets designed with domain experts can produce high-quality evaluations but are slow and costly to create, while synthetically generating data may scale efficiently but often results in unrealistic, redundant, or out-of-scope examples. To address this gap, we introduce a schema eliciting key information about the goals, scope, and context of an evaluation task, and use this information to guide synthetic data generation. We further define four criteria grounded in measurement validity for assessing dataset quality: coverage, diversity, content realism, and stylistic realism. Using these criteria, we show how expert-informed scaffolds can guide synthetic data generation toward more valid benchmarks. Through quantitative evaluations and a real-world case study with domain experts, we demonstrate that our approach improves benchmark data quality over existing methods while preserving validity. We additionally analyze how different types of schema information affect different dataset quality criteria, and provide practical guidance on which information to prioritize collecting under resource constraints.

cs.AI↗

Toward Third-Party Assurance of AI Systems: Design Requirements, Prototype, and Early Testing

As Artificial Intelligence (AI) systems proliferate, the need for systematic, transparent, and actionable processes for evaluating them is growing. While many resources exist to support AI evaluation, they have several limitations. Few address both the process of designing, developing, and deploying an AI system and the outcomes it produces. Furthermore, few are end-to-end and operational, give actionable guidance, or present evidence of usability or effectiveness in practice. In this paper, we introduce a third-party AI assurance framework that addresses these gaps. We focus on third-party assurance to prevent conflict of interest and ensure credibility and accountability of the process. We begin by distinguishing assurance from audits in several key dimensions. Then, following design principles, we reflect on the shortcomings of existing resources to identify a set of design requirements for AI assurance. We then construct a prototype of an assurance process that consists of (1) a responsibility assignment matrix to determine the different levels of involvement each stakeholder has at each stage of the AI lifecycle, (2) an interview protocol for each stakeholder of an AI system, (3) a maturity matrix to assess AI systems' adherence to best practices, and (4) a template for an assurance report that draws from more mature assurance practices in business accounting. We conduct early validation of our AI assurance framework by applying the framework to two distinct AI use cases -- a business document tagging tool for downstream processing in a large private firm, and a housing resource allocation tool in a public agency -- and conducting six expert validation interviews. Our findings show early evidence that our AI assurance framework is sound and comprehensive, usable across different organizational contexts, and effective at identifying bespoke issues with AI systems.

cs.CY↗

Towards Cognitively-Faithful Decision-Making Models to Improve AI Alignment

Recent AI trends seek to align AI models to learned human-centric objectives, such as personal preferences, utility, or societal values. Using standard preference elicitation methods, researchers and practitioners build models of human decisions and judgments, to which AI models are aligned. However, standard elicitation methods often fail to capture the cognitive processes behind human decision making, such as heuristics or simplifying structured thought patterns. To address this failure, we take an axiomatic approach to learning cognitively faithful decision processes from pairwise comparisons. Building on the literature analyzing cognitive processes that shape human decision-making, we derive a model class in which features are first processed with learned rules, then aggregated via a fixed rule, such as the Bradley-Terry rule, to produce a decision. This structured processing of information ensures that such models are realistic and feasible candidates to represent underlying human decision-making processes. We demonstrate the efficacy of this modeling approach by learning interpretable models of human decision making in a kidney allocation task, and show that our proposed models match or surpass the accuracy of prior models of human pairwise decision-making.

cs.LG↗

Beyond the Single Turn: Reframing Refusals as Dynamic Experiences Embedded in the Context of Mental Health Support Interactions with LLMs

Content Warning: This paper contains participant quotes and discussions related to mental health challenges, emotional distress, and suicidal ideation. Large language models (LLMs) are increasingly used for mental health support, yet the model safeguards -- particularly refusals to engage with sensitive content -- remain poorly understood from the perspectives of users and mental health professionals (MHPs) and have been reported to cause real-world harms. This paper presents findings from a sequential mixed-methods study examining how LLM refusals are experienced and interpreted in mental health support interactions. Through surveys (N=53) and in-depth interviews (N=16) with individuals using LLMs for mental health support and MHPs, we reveal that refusals are not isolated, single-turn system behaviors but rather constitute dynamic, multi-phase experiences: pre-refusal expectation formation, refusal triggering and encounter, refusal message framing, resource referral provision, and post-refusal outcomes. We contribute a multi-phase framework for evaluating refusals beyond binary policy compliance accuracy and design recommendations for future refusal mechanisms. These findings suggest that understanding LLM refusals requires moving beyond single-turn interactions toward recognizing them as holistic experiences embedded within users' support-seeking trajectories and the broader LLM design pipeline.

cs.HC↗

Evaluating AI-Generated Images of Cultural Artifacts with Community-Informed Rubrics

Measurement is essential to improving AI performance and mitigating harms for marginalized groups. As generative AI systems are rapidly deployed across geographies and contexts, AI measurement practices must be designed to support repeatable, automatable application across different models, datasets, and evaluation settings. But the drive to automate measurement can be in tension with the ability for measurement instruments to capture the expertise and perspectives of communities impacted by AI. Recent work advocates for breaking measurement into several key stages: first moving from an abstract concept to be measured into a precise, "systematized" concept; next operationalizing the systematized concept into a concrete measurement instrument; and finally applying the measurement instrument on data to produce measurements. This opens up an opportunity to concentrate community engagement in the systematization phase before operationalizing and applying measurement instruments. In this paper, we explore how to involve communities in systematizing the concept of "cultural appropriateness" in text-to-image models' representation of culturally significant artifacts through case studies with three communities: blind and low vision individuals residing in the UK, residents of Kerala, and residents of Tamil Nadu. Our systematized concepts reflect community members' lived experiences interacting with each artifact and how they want their material culture to be depicted, demonstrating the value of community involvement in defining valid measures. We explore how these systematized concepts can be operationalized into automated measurement instruments that could be applied using a multimodal LLM-as-a-judge approach and challenges that remain. We reflect on the benefits and limitations of such approaches.

cs.CY↗

What People See (and Miss) About Generative AI Risks: Perceptions of Failures, Risks, and Who Should Address Them

Despite growing concerns about the risks of Generative AI (GenAI), there is limited understanding of public perceptions of these risks and their associated failure modes -- defined as recurring patterns of sociotechnical breakdown across the GenAI lifecycle that contribute to risks of real-world harm. To address this gap, we present a survey instrument, validated with eight subject matter experts and deployed on a sample of 960 U.S.-based participants, to assess awareness and perceptions of GenAI's failure modes, their associated risks, and stakeholder responsibilities to address them. To support realism and content validity, our instrument is structured around scenarios grounded in publicly reported incidents and a taxonomy of GenAI's failure modes. Findings suggest that our instrument is (1) effective for assessing risk awareness and perceptions in a way that is grounded in people's current contexts of use, yet is extensible to new contexts that will inevitably arise; and (2) potentially useful for informing the design of AI literacy tools and interventions. We argue for AI literacy and governance approaches that align with how people encounter and reason about GenAI in everyday life.

cs.HC↗

Disclosure or Marketing? Analyzing the Efficacy of Vendor Self-reports for Vetting Public-sector AI

Documentation-based disclosure has become a central governance strategy for responsible AI, particularly in public-sector procurement. Tools such as model cards, datasheets, and AI FactSheets are increasingly expected to support accountability, risk assessment, and informed decision-making across organizational boundaries. Yet there is limited empirical evidence about how these artifacts are produced, interpreted, and used in practice. In this paper, we present a qualitative study of the GovAI Coalition FactSheet, a widely adopted transparency document designed to support AI procurement and governance in government contexts. Drawing on semi-structured interviews with vendors and public-sector practitioners, alongside a systematic analysis of completed FactSheets, we examine how FactSheets are used, what information they surface, and where they fall short. We find that FactSheets are asked to serve multiple and conflicting purposes simultaneously: showcasing vendor offerings, supporting evaluation and due diligence, and facilitating early-stage dialogue between vendors and agencies. These competing expectations, combined with the structural constraints of voluntary and public self-disclosure, limit the ability of FactSheets to function as standalone evaluation or risk-assessment tools. At the same time, our findings suggest that when understood as relational artifacts used to establish trust, shared understanding, and ongoing dialogue, FactSheets can help create conditions that support more meaningful disclosure and governance over time.

cs.HC↗

Navigating Uncertainties: How GenAI Developers Document Their Models on Open-Source Platforms

Model documentation plays a crucial role in promoting transparency and responsible development of AI systems. With the rise of Generative AI (GenAI), open-source platforms have increasingly become hubs for hosting and distributing these models, prompting platforms like Hugging Face to develop dedicated model documentation guidelines that align with responsible AI principles. Despite these growing efforts, there remains a lack of understanding of how developers document their GenAI models on open-source platforms. Through interviews with 13 GenAI developers active on open-source platforms, we provide empirical insights into their documentation practices and challenges. Our analysis reveals that despite existing resources, developers of GenAI models still face multiple layers of uncertainties in their model documentation: (1) uncertainties about what specific content should be included; (2) uncertainties about how to effectively report key components of their models; and (3) uncertainties in deciding who should take responsibilities for various aspects of model documentation. Based on our findings, we discuss the implications for policymakers, open-source platforms, and the research community to support meaningful, effective and actionable model documentation in the GenAI era, including cultivating better community norms, building robust evaluation infrastructures, and clarifying roles and responsibilities.

cs.HC↗

Moral Change or Noise? On Problems of Aligning AI With Temporally Unstable Human Feedback

Alignment methods in moral domains seek to elicit moral preferences of human stakeholders and incorporate them into AI. This presupposes moral preferences as static targets, but such preferences often evolve over time. Proper alignment of AI to dynamic human preferences should ideally account for "legitimate" changes to moral reasoning, while ignoring changes related to attention deficits, cognitive biases, or other arbitrary factors. However, common AI alignment approaches largely neglect temporal changes in preferences, posing serious challenges to proper alignment, especially in high-stakes applications of AI, e.g., in healthcare domains, where misalignment can jeopardize the trustworthiness of the system and yield serious individual and societal harms. This work investigates the extent to which people's moral preferences change over time, and the impact of such changes on AI alignment. Our study is grounded in the kidney allocation domain, where we elicit responses to pairwise comparisons of hypothetical kidney transplant patients from over 400 participants across 3-5 sessions. We find that, on average, participants change their response to the same scenario presented at different times around 6-20% of the time (exhibiting "response instability"). Additionally, we observe significant shifts in several participants' retrofitted decision-making models over time (capturing "model instability"). The predictive performance of simple AI models decreases as a function of both response and model instability. Moreover, predictive performance diminishes over time, highlighting the importance of accounting for temporal changes in preferences during training. These findings raise fundamental normative and technical challenges relevant to AI alignment, highlighting the need to better understand the object of alignment (what to align to) when user preferences change significantly over time.

cs.HC↗

Modeling the Economic Impacts of AI Openness Regulation

Regulatory frameworks, such as the EU AI Act, encourage openness of general-purpose AI models by offering legal exemptions for "open-source" models. Despite this legislative attention on openness, the definition of open-source foundation models remains ambiguous. This paper models the strategic interactions among the creator of a general-purpose model (the generalist) and the entity that fine-tunes the general-purpose model to a specialized domain or task (the specialist), in response to regulatory requirements on model openness. We present a stylized model of the regulator's choice of an open-source definition to evaluate which AI openness standards will establish appropriate economic incentives for developers. Our results characterize market equilibria -- specifically, upstream model release decisions and downstream fine-tuning efforts -- under various openness regulations and present a range of effective regulatory penalties and open-source thresholds. Overall, we find the model's baseline performance determines when increasing the regulatory penalty vs. the open-source threshold will significantly alter the generalist's release strategy. Our model provides a theoretical foundation for AI governance decisions around openness and enables evaluation and refinement of practical open-source policies.

cs.GT↗

Persona-Augmented Benchmarking: Evaluating LLMs Across Diverse Writing Styles

Current benchmarks for evaluating Large Language Models (LLMs) often do not exhibit enough writing style diversity, with many adhering primarily to standardized conventions. Such benchmarks do not fully capture the rich variety of communication patterns exhibited by humans. Thus, it is possible that LLMs, which are optimized on these benchmarks, may demonstrate brittle performance when faced with "non-standard" input. In this work, we test this hypothesis by rewriting evaluation prompts using persona-based LLM prompting, a low-cost method to emulate diverse writing styles. Our results show that, even with identical semantic content, variations in writing style and prompt formatting significantly impact the estimated performance of the LLM under evaluation. Notably, we identify distinct writing styles that consistently trigger either low or high performance across a range of models and tasks, irrespective of model family, size, and recency. Our work offers a scalable approach to augment existing benchmarks, improving the external validity of the assessments they provide for measuring LLM performance across linguistic variations.

cs.CL↗

From Firewalls to Frontiers: AI Red-Teaming is a Domain-Specific Evolution of Cyber Red-Teaming

A red team simulates adversary attacks to help defenders find effective strategies to defend their systems in a real-world operational setting. As more enterprise systems adopt AI, red-teaming will need to evolve to address the unique vulnerabilities and risks posed by AI systems. We take the position that AI systems can be more effectively red-teamed if AI red-teaming is recognized as a domain-specific evolution of cyber red-teaming. Specifically, we argue that existing Cyber Red Teams who adopt this framing will be able to better evaluate systems with AI components by recognizing that AI poses new risks, has new failure modes to exploit, and often contains unpatchable bugs that re-prioritize disclosure and mitigation strategies. Similarly, adopting a cybersecurity framing will allow existing AI Red Teams to leverage a well-tested structure to emulate realistic adversaries, promote mutual accountability with formal rules of engagement, and provide a pattern to mature the tooling necessary for repeatable, scalable engagements. In these ways, the merging of AI and Cyber Red Teams will create a robust security ecosystem and best position the community to adapt to the rapidly changing threat landscape.

cs.LG↗

Why Do Decision Makers (Not) Use AI? A Cross-Domain Analysis of Factors Impacting AI Adoption

Growing excitement around deploying AI across various domains calls for a careful assessment of how human decision-makers interact with AI-powered systems. In particular, it is essential to understand when decision-makers voluntarily choose to consult AI tools, which we term decision-maker adoption. We interviewed experts across four domains -- medicine, law, journalism, and the public sector -- to explore current AI use cases and perceptions of adoption. From these interviews, we identify key factors that shape decision-maker adoption of AI tools: the decision-maker's background, perceptions of the AI, consequences for the decision-maker, and perceived implications for other stakeholders. We translate these factors into an AI adoption sheet to analyze how decision-makers approach adoption choices through comparative, cross-domain case studies, highlighting how our factors help explain inter-domain differences in adoption. Our findings offer practical guidance for supporting the responsible and context-aware deployment of AI by better accounting for the decision-maker's perspective.

cs.HC↗

The "Who", "What", and "How" of Responsible AI Governance: A Systematic Review and Meta-Analysis of (Actor, Stage)-Specific Tools

The implementation of responsible AI in an organization is inherently complex due to the involvement of multiple stakeholders, each with their unique set of goals and responsibilities across the entire AI lifecycle. These responsibilities are often ambiguously defined and assigned, leading to confusion, miscommunication, and inefficiencies. Even when responsibilities are clearly defined and assigned to specific roles, the corresponding AI actors lack effective tools to support their execution. Toward closing these gaps, we present a systematic review and comprehensive meta-analysis of the current state of responsible AI tools, focusing on their alignment with specific stakeholder roles and their responsibilities in various AI lifecycle stages. We categorize over 220 tools according to AI actors and stages they address. Our findings reveal significant imbalances across the stakeholder roles and lifecycle stages addressed. The vast majority of available tools have been created to support AI designers and developers specifically during data-centric and statistical modeling stages while neglecting other roles such as institutional leadership, deployers, end-users, and impacted communities, and stages such as value proposition and deployment. The uneven distribution we describe here highlights critical gaps that currently exist in responsible AI governance research and practice. Our analysis reveals that despite the myriad of frameworks and tools for responsible AI, it remains unclear \emph{who} within an organization and \emph{when} in the AI lifecycle a tool applies. Furthermore, existing tools are rarely validated, leaving critical gaps in their usability and effectiveness. These gaps provide a starting point for researchers and practitioners to create more effective and holistic approaches to responsible AI development and governance.

cs.CY↗