arXiv subjects

Allie Griffith

Publications and source records attributed to Allie Griffith.

4 recordsLinked to original sources

arXiv2026

The Doctor Will (Still) See You Now: On the Structural Limits of Agentic AI in Healthcare

Across healthcare, agentic artificial intelligence (AI) systems are increasingly promoted as capable of autonomous action, yet in practice they currently operate under near-total human oversight due to safety, regulatory, and liability constraints that make autonomous clinical reasoning infeasible in high-stakes environments. While market enthusiasm suggests a revolution in healthcare agents, the conceptual assumptions and accountability structures shaping these systems remain underexamined. We present a qualitative study based on interviews with 20 stakeholders, including developers, implementers, and end users. Our analysis identifies three mutually reinforcing tensions: conceptual fragmentation regarding the definition of `agentic'; an autonomy contradiction where commercial promises exceed operational reality; and an evaluation blind spot that prioritizes technical benchmarks over sociotechnical safety. We argue that agentic {AI} functions as a site of contested meaning-making where technical aspirations, commercial incentives, and clinical constraints intersect, carrying material consequences for patient safety and the distribution of blame.

cs.CY↗

arXiv2025

The Measurement Imbalance in Agentic AI Evaluation Undermines Industry Productivity Claims

As industry reports claim agentic AI systems deliver double-digit productivity gains and multi-trillion dollar economic potential, the validity of these claims has become critical for investment decisions, regulatory policy, and responsible technology adoption. However, this paper demonstrates that current evaluation practices for agentic AI systems exhibit a systemic imbalance that calls into question prevailing industry productivity claims. Our systematic review of 84 papers (2023--2025) reveals an evaluation imbalance where technical metrics dominate assessments (83%), while human-centered (30%), safety (53%), and economic assessments (30%) remain peripheral, with only 15% incorporating both technical and human dimensions. This measurement gap creates a fundamental disconnect between benchmark success and deployment value. We present evidence from healthcare, finance, and retail sectors where systems excelling on technical metrics failed in real-world implementation due to unmeasured human, temporal, and contextual factors. Our position is not against agentic AI's potential, but rather that current evaluation frameworks systematically privilege narrow technical metrics while neglecting dimensions critical to real-world success. We propose a balanced four-axis evaluation model and call on the community to lead this paradigm shift because benchmark-driven optimization shapes what we build. By redefining evaluation practices, we can better align industry claims with deployment realities and ensure responsible scaling of agentic systems in high-stakes domains.

cs.CY↗

arXiv2024

More than Marketing? On the Information Value of AI Benchmarks for Practitioners

Public AI benchmark results are widely broadcast by model developers as indicators of model quality within a growing and competitive market. However, these advertised scores do not necessarily reflect the traits of interest to those who will ultimately apply AI models. In this paper, we seek to understand if and how AI benchmarks are used to inform decision-making. Based on the analyses of interviews with 19 individuals who have used, or decided against using, benchmarks in their day-to-day work, we find that across these settings, participants use benchmarks as a signal of relative performance difference between models. However, whether this signal was considered a definitive sign of model superiority, sufficient for downstream decisions, varied. In academia, public benchmarks were generally viewed as suitable measures for capturing research progress. By contrast, in both product and policy, benchmarks -- even those developed internally for specific tasks -- were often found to be inadequate for informing substantive decisions. Of the benchmarks deemed unsatisfactory, respondents reported that their goals were neither well-defined nor reflective of real-world use. Based on the study results, we conclude that effective benchmarks should provide meaningful, real-world evaluations, incorporate domain expertise, and maintain transparency in scope and goals. They must capture diverse, task-relevant capabilities, be challenging enough to avoid quick saturation, and account for trade-offs in model performance rather than relying on a single score. Additionally, proprietary data collection and contamination prevention are critical for producing reliable and actionable results. By adhering to these criteria, benchmarks can move beyond mere marketing tricks into robust evaluative frameworks.

cs.AI↗

arXiv2024

ASTPrompter: Preference-Aligned Automated Language Model Red-Teaming to Generate Low-Perplexity Unsafe Prompts

Existing LLM red-teaming approaches prioritize high attack success rate, often resulting in high-perplexity prompts. This focus overlooks low-perplexity attacks that are more difficult to filter, more likely to arise during benign usage, and more impactful as negative downstream training examples. In response, we introduce ASTPrompter, a single-step optimization method that uses contrastive preference learning to train an attacker to maintain low perplexity while achieving a high attack success rate (ASR). ASTPrompter achieves an attack success rate 5.1 times higher on Llama-8.1B while using inputs that are 2.1 times more likely to occur according to the frozen LLM. Furthermore, our attack transfers to Mistral-7B, Qwen-7B, and TinyLlama in both black- and white-box settings. Lastly, by tuning a single hyperparameter in our method, we discover successful attack prefixes along an efficient frontier between ASR and perplexity, highlighting perplexity as a previously under-considered factor in red-teaming.

cs.CL↗