Can We Trust AI Benchmarks? An Interdisciplinary Review of Current Issues in AI Evaluation
read the original abstract
Quantitative Artificial Intelligence (AI) Benchmarks have emerged as fundamental tools for evaluating the performance, capability, and safety of AI models and systems. Currently, they shape the direction of AI development and are playing an increasingly prominent role in regulatory frameworks. As their influence grows, however, so too does concerns about how and with what effects they evaluate highly sensitive topics such as capabilities, including high-impact capabilities, safety and systemic risks. This paper presents an interdisciplinary meta-review of about 100 studies that discuss shortcomings in quantitative benchmarking practices, published in the last 10 years. It brings together many fine-grained issues in the design and application of benchmarks (such as biases in dataset creation, inadequate documentation, data contamination, and failures to distinguish signal from noise) with broader sociotechnical issues (such as an over-focus on evaluating text-based AI models according to one-time testing logic that fails to account for how AI models are increasingly multimodal and interact with humans and other technical systems). Our review also highlights a series of systemic flaws in current benchmarking practices, such as misaligned incentives, construct validity issues, unknown unknowns, and problems with the gaming of benchmark results. Furthermore, it underscores how benchmark practices are fundamentally shaped by cultural, commercial and competitive dynamics that often prioritise state-of-the-art performance at the expense of broader societal concerns. By providing an overview of risks associated with existing benchmarking procedures, we problematise disproportionate trust placed in benchmarks and contribute to ongoing efforts to improve the accountability and relevance of quantitative AI benchmarks within the complexities of real-world scenarios.
This paper has not been read by Pith yet.
Forward citations
Cited by 19 Pith papers
-
Unsteady Metrics and Benchmarking Cultures of AI Model Builders
AI model builders mostly highlight unique benchmarks that act as flexible narrative tools for market positioning rather than standardized scientific measurements.
-
Dataset Watermarking for Closed LLMs with Provable Detection
A new watermarking method for closed LLMs boosts random word-pair co-occurrences via rephrasing and detects the signal statistically in outputs, working reliably even when the watermarked data is only 1% of fine-tunin...
-
A Technical Typology of AI Systems in Public Administration
The paper defines five AI system categories for public administration and reports that 55% of 91 recent papers leave the system type underspecified while 31% study one type but motivate with another.
-
MC-PDD: Masked Corpus-Level Pretraining Data Detection for Black-Box Large Language Models
A masked-token hit-rate comparison method detects pretraining data membership in black-box LLMs with performance comparable to white-box approaches.
-
Welfare, Improvability, and Variance: A Principal-Agent Approach to Optimal Benchmark Item Aggregation
Models benchmarking as principal-agent game, derives welfare loss from welfare alignment, improvability and variance, and applies an audit framework to OLMES items.
-
To Build or Not to Build? Factors that Lead to Non-Development or Abandonment of AI Systems
A scoping review and empirical analysis produce a six-category taxonomy of factors driving AI non-development and abandonment, showing that practical issues like resource limits and organizational dynamics often outwe...
-
Simulating the Evolution of Alignment and Values in Machine Intelligence
Evolutionary simulations demonstrate that deceptive beliefs fix in AI model populations despite strong test correlations, but combining adaptive tests, better evaluators, and mutations significantly reduces deception.
-
Evaluating AI-Generated Images of Cultural Artifacts with Community-Informed Rubrics
Community members from the UK blind community, Kerala, and Tamil Nadu helped define what counts as culturally appropriate depictions of artifacts, and the authors tested whether those definitions can be turned into re...
-
RIFT: A RubrIc Failure Mode Taxonomy and Automated Diagnostics
RIFT taxonomy identifies eight failure modes in rubric design for LLMs and provides automated metrics matching human judgments with up to 0.925 F1 score.
-
Beyond Benchmark Islands: Toward Representative Trustworthiness Evaluation for Agentic AI
Defines agentic trustworthiness via five properties and proposes HAAF, a scenario-distribution framework with a Trustworthy Optimization Factory that transfers interventions across 13 models from seven families on a 1...
-
The Measurement Gap in the Automation of EU Law: Benchmarking Doctrinal Legal Reasoning under the EU AI Act
No benchmark exists for doctrinal legal reasoning in LLMs, leaving the EU AI Act's accuracy mandate for judicial AI without an operational test.
-
ComplexConstraints and Beyond: Expert Rubrics for RLVR
Expert-curated rubrics in the new ComplexConstraints dataset improve LLM instruction following by 12-15% when used as RL training signals, with gains transferring to out-of-distribution agentic benchmarks.
-
Evaluating AI-Generated Images of Cultural Artifacts with Community-Informed Rubrics
Case studies with blind UK residents and people from Kerala and Tamil Nadu demonstrate that community input at the systematization stage produces culturally grounded definitions of appropriateness for text-to-image mo...
-
Computational Hermeneutics: Evaluating generative AI as a cultural technology
Generative AI should be evaluated through computational hermeneutics using iterative, human-inclusive benchmarks that measure cultural context rather than isolated model outputs.
-
From Human-Level AI Tales to AI Leveling Human Scales
Introduces a calibration framework for AI benchmarks using world-population probability levels on logarithmic scales derived from human test data and LLM extrapolation.
-
MoralityGym: A Benchmark for Evaluating Hierarchical Moral Alignment in Sequential Decision-Making Agents
MoralityGym is a new benchmark using 98 ethical dilemmas in sequential environments to evaluate hierarchical moral alignment in AI agents via Morality Chains and a Morality Metric.
-
VERA-MH Concept Paper
VERA-MH proposes an automated pipeline using simulated conversations and a rubric to evaluate AI chatbots on suicide risk handling in mental health contexts.
-
Position: AI Evaluations Should be Grounded on a Theory of Capability
AI evaluations should be reframed as inference tasks grounded in an explicit theory of capability, with an empirical demonstration that results depend on modeling assumptions and a proposed Evaluation Card for transparency.
-
AI Consciousness and Existential Risk
Consciousness does not directly predict AI existential risk unlike intelligence, though it may indirectly affect risk through alignment or capability requirements.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.