@inproceedings{gao-etal-2026-llms-judge,
title = "Can {LLM}s Judge Pedagogy? Assessing Conversational {AI} {STEM} Tutoring with {AI}-as-a-Judge",
author = "Gao, Xintian and
Shen, Qian and
Li, Xin",
editor = "Wilson, Joshua and
Ormerod, Christopher and
Beiting-Parrish, Magdalen",
booktitle = "Proceedings of the Artificial Intelligence in Measurement and Education Conference ({AIME}-Con): Full Papers",
month = oct,
year = "2026",
address = "Wyndham Grand Pittsburgh Downtown, Pittsburgh, Pennsylvania, United States",
publisher = "National Council on Measurement in Education (NCME)",
url = "https://aclanthology.org/2026.aimecon-main.69/",
pages = "612--621",
ISBN = "979-8-9983004-0-0",
abstract = "Large language models are increasingly used as automated judges in education, yet their ability to score pedagogical quality in AI tutor responses to K-12 STEM student inquiries remains underexplored. This study evaluates whether two LLM-based scorers, Nemotron-3-Super-120B-A12B and GPT-OSS-120B, can approximate human judgment of single-turn Socratic style responses to STEM inquiries. Using a four-dimension rubric adapted from the CPS-R Questioning and Thinking subscale, we compare human and model ratings. Results show mixed reliability: agreement is stronger for more observable instructional features such as Cognitive Demand, but weaker for more interpretive dimensions, especially Encouraging Metacognition and Differentiation. Chance-corrected reliability is also sensitive to skewed score distributions, as shown by a base-rate effect in Differentiation. A mixed-effects analysis further reveals that the two LLM scorers differ systematically, with larger divergence on elementary-level items. We also observe prompt-adherence failures in generated tutoring responses, where some outputs briefly violate the instruction to avoid direct answers before returning to a Socratic response. Overall, the findings suggest that LLMs can assist with large-scale pedagogical evaluation, but human oversight remains necessary for nuanced instructional assessment and for maintaining Socratic response style."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="gao-etal-2026-llms-judge">
<titleInfo>
<title>Can LLMs Judge Pedagogy? Assessing Conversational AI STEM Tutoring with AI-as-a-Judge</title>
</titleInfo>
<name type="personal">
<namePart type="given">Xintian</namePart>
<namePart type="family">Gao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Qian</namePart>
<namePart type="family">Shen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xin</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the Artificial Intelligence in Measurement and Education Conference (AIME-Con): Full Papers</title>
</titleInfo>
<name type="personal">
<namePart type="given">Joshua</namePart>
<namePart type="family">Wilson</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Christopher</namePart>
<namePart type="family">Ormerod</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Magdalen</namePart>
<namePart type="family">Beiting-Parrish</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>National Council on Measurement in Education (NCME)</publisher>
<place>
<placeTerm type="text">Wyndham Grand Pittsburgh Downtown, Pittsburgh, Pennsylvania, United States</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-9983004-0-0</identifier>
</relatedItem>
<abstract>Large language models are increasingly used as automated judges in education, yet their ability to score pedagogical quality in AI tutor responses to K-12 STEM student inquiries remains underexplored. This study evaluates whether two LLM-based scorers, Nemotron-3-Super-120B-A12B and GPT-OSS-120B, can approximate human judgment of single-turn Socratic style responses to STEM inquiries. Using a four-dimension rubric adapted from the CPS-R Questioning and Thinking subscale, we compare human and model ratings. Results show mixed reliability: agreement is stronger for more observable instructional features such as Cognitive Demand, but weaker for more interpretive dimensions, especially Encouraging Metacognition and Differentiation. Chance-corrected reliability is also sensitive to skewed score distributions, as shown by a base-rate effect in Differentiation. A mixed-effects analysis further reveals that the two LLM scorers differ systematically, with larger divergence on elementary-level items. We also observe prompt-adherence failures in generated tutoring responses, where some outputs briefly violate the instruction to avoid direct answers before returning to a Socratic response. Overall, the findings suggest that LLMs can assist with large-scale pedagogical evaluation, but human oversight remains necessary for nuanced instructional assessment and for maintaining Socratic response style.</abstract>
<identifier type="citekey">gao-etal-2026-llms-judge</identifier>
<location>
<url>https://aclanthology.org/2026.aimecon-main.69/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>612</start>
<end>621</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Can LLMs Judge Pedagogy? Assessing Conversational AI STEM Tutoring with AI-as-a-Judge
%A Gao, Xintian
%A Shen, Qian
%A Li, Xin
%Y Wilson, Joshua
%Y Ormerod, Christopher
%Y Beiting-Parrish, Magdalen
%S Proceedings of the Artificial Intelligence in Measurement and Education Conference (AIME-Con): Full Papers
%D 2026
%8 October
%I National Council on Measurement in Education (NCME)
%C Wyndham Grand Pittsburgh Downtown, Pittsburgh, Pennsylvania, United States
%@ 979-8-9983004-0-0
%F gao-etal-2026-llms-judge
%X Large language models are increasingly used as automated judges in education, yet their ability to score pedagogical quality in AI tutor responses to K-12 STEM student inquiries remains underexplored. This study evaluates whether two LLM-based scorers, Nemotron-3-Super-120B-A12B and GPT-OSS-120B, can approximate human judgment of single-turn Socratic style responses to STEM inquiries. Using a four-dimension rubric adapted from the CPS-R Questioning and Thinking subscale, we compare human and model ratings. Results show mixed reliability: agreement is stronger for more observable instructional features such as Cognitive Demand, but weaker for more interpretive dimensions, especially Encouraging Metacognition and Differentiation. Chance-corrected reliability is also sensitive to skewed score distributions, as shown by a base-rate effect in Differentiation. A mixed-effects analysis further reveals that the two LLM scorers differ systematically, with larger divergence on elementary-level items. We also observe prompt-adherence failures in generated tutoring responses, where some outputs briefly violate the instruction to avoid direct answers before returning to a Socratic response. Overall, the findings suggest that LLMs can assist with large-scale pedagogical evaluation, but human oversight remains necessary for nuanced instructional assessment and for maintaining Socratic response style.
%U https://aclanthology.org/2026.aimecon-main.69/
%P 612-621
Markdown (Informal)
[Can LLMs Judge Pedagogy? Assessing Conversational AI STEM Tutoring with AI-as-a-Judge](https://aclanthology.org/2026.aimecon-main.69/) (Gao et al., AIME-Con 2026)
ACL