AI and the Everything in the Whole Wide World Benchmark

2021/11/26 by Inioluwa Deborah Raji, Emily M. Bender, Raji, Inioluwa Deborah +7 · 6 voices · 41 citations
Computer Science · #Explainable Artificial Intelligence (XAI)

paper · pdf · doi:10.48550/arxiv.2111.15366

Abstract

There is a tendency across different subfields in AI to valorize a small collection of influential benchmarks. These benchmarks operate as stand-ins for a range of anointed common problems that are frequently framed as foundational milestones on the path towards flexible and generalizable AI systems. State-of-the-art performance on these benchmarks is widely understood as indicative of progress towards these long-term goals. In this position paper, we explore the limits of such benchmarks in order to reveal the construct validity issues in their framing as the functionally "general" broad measures of progress they are set up to be.

Cited by

• BenCSSmark: Making the Social Sciences Count in LLM Research
• One Persona, Many Cues, Different Results: How Sociodemographic Cues Impact LLM Personalization
• Selecting Language Models for Social Science: Start Small, Start Open, and Validate
• From Controlled to the Wild: Evaluation of Pentesting Agents for the Real-World
• How Do LLMs Fail In Agentic Scenarios? A Qualitative Analysis of Success and Failure Scenarios of Various LLMs in Agentic Simulations
• Eval Factsheets: A Structured Framework for Documenting AI Evaluations
• Whose Personae? Synthetic Persona Experiments in LLM Research and Pathways to Transparency
• Rethinking AI Evaluation in Education: The TEACH-AI Framework and Benchmark for Generative AI Assistants
• Realist and Pluralist Conceptions of Intelligence and Their Implications on AI Research
• Towards a Standard, Enterprise-Relevant Agentic AI Benchmark: Lessons from 5.5 billion tokens' worth of agentic AI evaluations
• Setting ε is not the Issue in Differential Privacy
• Measuring what Matters: Construct Validity in Large Language Model Benchmarks
• Back to the Communities: A Mixed-Methods and Community-Driven Evaluation of Cultural Sensitivity in Text-to-Image Models
• Scales++: Compute Efficient Evaluation Subset Selection with Cognitive Scales Embeddings
• Why Public Service AI Governance Frameworks Risk Failing in the Age of General-Purpose AI: Lessons from Policing
• Localized Adaptation Reveals Distinct Learning Signatures in Transformers
• ‘Appears to be about’: an evaluation of AI-generated metadata quality for community archives
• When benchmark inferences do not compose: Projectibility in AI evaluation
• EsoLang-Bench: Evaluating Genuine Reasoning in Large Language Models via Esoteric Programming Languages
• Reward Models are Metrics in a Trench Coat
• The Benchmarking Epistemology: Construct Validity for Evaluating Machine Learning Models
• Risk Management for Mitigating Benchmark Failure Modes: BenchRisk
• Multi-modal video data-pipelines for machine learning with minimal human supervision
• What Slows Down FMware Development? An Empirical Study of Developer Challenges and Resolution Times
• ConsistencyAI: A Benchmark to Assess LLMs' Factual Consistency When Responding to Different Demographic Groups
• Benchmarking is Broken -- Don't Let AI be its Own Judge
• Assessing Human Rights Risks in AI: A Framework for Model Evaluation
• Intuitions of Machine Learning Researchers about Transfer Learning for Medical Image Classification
• Towards Ecologically Valid LLM Benchmarks: Understanding and Designing Domain-Centered Evaluations for Journalism Practitioners
• Branching Out: Broadening AI Measurement and Evaluation with Measurement Trees
• NeurIPS should lead scientific consensus on AI policy
• Toward a Theory of Generalizability in LLM Mechanistic Interpretability Research
• What Does Your Benchmark Really Measure? A Framework for Robust Inference of AI Capabilities
• Hallucination‐Free? Assessing the Reliability of Leading <scp>AI</scp> Legal Research Tools
• AI systems and the reproduction of (standard) language ideologies in World Englishes
• The Artificial Intelligence Cognitive Examination: A Survey on the Evolution of Multimodal Evaluation From Recognition to Reasoning
• The Collaboration Paradox: Why Generative AI Requires Both Strategic Intelligence and Operational Stability in Supply Chain Management
• Is GPT-OSS Good? A Comprehensive Evaluation of OpenAI's Latest Open Source Models
• The Cultural Gene of Large Language Models: A Study on the Impact of Cross-Corpus Training on Model Values and Biases
• Benchmark-Driven Selection of AI: Evidence from DeepSeek-R1
• Advancing Data Equity: Practitioner Responsibility and Accountability in NLP Data Practices
• Bench-2-CoP: Can We Trust Benchmarking for EU AI Compliance?

Discussions

• Link here, ran out of room in the last skeet. [bsky, 10 points, 1 comments]
• I feel like I've been effectively yelling about this for years, especially as it relates to general benchmarks (arxiv.org/pdf/2111.15366) and in the medical domain (arxiv.org/pdf/2503.106..., it's gre [bsky, 9 points, 1 comments]
• A nice @techcrunch.com discussion re inadequacy of benchmarks w/ illuminating commentary by Dr. David Widder. Hope author follows up by contacting @rajiinio.bsky.social, @emilymbender.bsky.social @a [bsky, 5 points, 0 comments]
• Heres one about the limits of the benchmarks because of "construct validty": arxiv.org/pdf/2111.15366 [bsky, 1 points, 0 comments]
• "benchmarks couldn't predict the risks… and yet no one raised the alarm. Why?" * Stares in Grover arxiv.org/abs/2111.15366 [bsky, 0 points, 1 comments]
• AI and the Everything in the Whole Wide World Benchmark - This position paper explores the limits of such benchmarks and their validity issues arxiv.org/abs/2111.15366 [bsky, 0 points, 0 comments]

Related

• A Rosetta Stone for AI Benchmarks
• Can We Trust AI Benchmarks? An Interdisciplinary Review of Current Issues in AI Evaluation
• The Geometry of Benchmarks: A New Path Toward AGI
• Unsteady Metrics and Benchmarking Cultures of AI Model Builders
• What Does Your Benchmark Really Measure? A Framework for Robust Inference of AI Capabilities