Can We Trust Item Response Theory for AI Evaluation?

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Han Jiang, Sunbeom Kwon, Jinwen Luo, Ziang Xiao, Susu Zhang
• Venue: arXiv: cs.AI
• Published: 2026-07-16
• Collected: 2026-07-17
• URL: http://arxiv.org/abs/2607.15190v1
• PDF: https://arxiv.org/pdf/2607.15190v1
• DOI: UNKNOWN

Plain-Language Brief

The paper explores the reliability of using Item Response Theory (IRT) for evaluating Artificial Intelligence (AI) systems in benchmarks. The authors find that standard IRT estimation tools may not be reliable in large AI benchmark settings, especially when dealing with skewed or clustered capability distributions.

What the Document Says

• AI benchmarks increasingly use item-level statistical models like IRT for model evaluation.
• Benchmark data often deviates from the data regime of human testing.
• Simulation results show classical estimators can become infeasible and scalable estimators may produce unreliable results in large AI benchmarks.
• The study identifies when latent trait models are reliable and what sample sizes and diagnostics are needed for trustworthy use.

Key Concepts Mentioned

• AI
• Item Response Theory (IRT)
• AI Benchmarking
• Latent Trait Models

Claims Made by the Document

• Classical estimators can become infeasible in large AI benchmark settings.
• Scalable estimators may produce unreliable results with small or non-normally distributed model sets.

Evidence Used

• Results from simulation conditions across 18,000 simulations comparing four estimation tools.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

AI benchmarks increasingly leverage item-level statistical models, particularly item response theory (IRT), to estimate model capabilities, rank systems, select informative examples, and diagnose benchmark quality. However, AI benchmark data often departs from the data regime of human testing, for which standard IRT estimation tools were originally developed: benchmarks typically involve fewer evaluated models, far more items, and capability distributions that may be skewed, clustered, or multimodal. We examine how these regime mismatches challenge the reliability of IRT modeling for AI evaluation. Using item parameters and capability distributions derived from six widely used LLM benchmarks, we simulate response matrices under three common IRT models and compare four estimation tools used in recent benchmark studies: marginal maximum likelihood, Markov chain Monte Carlo, variational inference, and a neural pseudo-Siamese estimator. Across 18,000 simulation conditions, we systematically evaluate computational feasibility, scalability, and the reliability of IRT inferences about model rankings, predicted performance, and item characteristics. Results show that classical estimators can become infeasible in large benchmark settings, whereas scalable estimators can produce unreliable item-level and ranking inferences with small or nonnormally distributed model sets. This study identifies when latent trait models reliably support or risk distorting AI benchmarking claims, and what sample sizes and diagnostics are needed for trustworthy use.