A Two-Phase Stability Study of LLM Judges and Bar Council Examiners on Thai Bar-Exam Free-Form Essays
2026-05-25 • Computation and Language
Computation and LanguageComputers and Society
AI summaryⓘ
The authors studied how different human experts and AI language models (LLMs) score legal essays for the Thai bar exam. They found that while humans sometimes disagree on tricky answers, the AI judges mostly agreed with the majority human opinion and never matched the minority human view. This means the AI tends to pick one common interpretation rather than reflecting the full range of human disagreement. The authors also showed that selecting AI judges based on how well they agree with humans may reinforce this bias.
legal essay evaluationinter-rater reliabilityLLM judge panelThai bar examinationgrading rubrichuman disagreementAI alignmentCronbach's alphabenchmarkingstatutory citation
Authors
Pawitsapak Akarajaradwong, Wuttikrai Lertprasertphakorn, Chompakorn Chaksangchaichot, Sarana Nutanong
Abstract
Free-form legal essay evaluation in NLP treats expert inter-rater stability as a single ceiling number, and treats LLM-judge agreement with that ceiling as evidence of judge stability. We test both assumptions on the Thai bar examination through an identical-inputs protocol: three Bar Council-trained examiners (A, B, C) and a 26-LLM judge panel score the same 15 cross-graded answers from the same four inputs (question, official Bar Council grading regulation, gold answer, candidate answer). The headline finding is asymmetric. On 10 of 15 cells where the rubric prescribes both axes, all 29 raters converge in a tight band: panel agreement is universal. On the remaining 5 cells where the rubric does not prescribe how to grade a correct final answer that omits a decisive statutory citation, the human panel splits between two coherent readings (B/C majority at the upper rubric band, score $6$--$8$; A minority at the lower band, score $1$--$2$). The LLM judge population does not split symmetrically: 22 of 26 LLMs score in or near B/C's contested band, 3 sit in the regulation-silent middle gap, and only 1 (GPT-5.4 Nano) approaches A's band without consistently scoring within it. \emph{Zero LLMs in our 26-judge panel reproduce the minority human reading on the contested cells.} The B/C-direction cluster spans every model size, vendor, and price tier we tested. An instrumented three-LLM anchor sub-panel (Claude 4.6 Opus, Gemini 3.1 Pro, GPT-5.4 Pro) carries determinism probes, input ablations, and bootstrap CIs, and reaches anchor panel $α= 0.77$ on the 15 cells against human-panel $α= 0.36$. The high LLM-panel $α$ reflects systematic convergence on the majority reading rather than balanced reproduction of both readings; a benchmark that selects its LLM judge by maximising agreement with a human reference panel will inherit this asymmetry by construction.