South Korea's first regional AI essay-grading pilot, run by Gyeonggi Province from July 2025, drew real teacher-trust survey data and documented errors — blank answers scored, essays over-rewarded for length — plus a public backlash after a promotional video mocking teachers forc
64.6% of 48 surveyed
Teachers who used Hi-Learning
29.4% of 17 users
Users who felt it reduced grading workload
Details
Maturity
Pilot
Promoter
Gyeonggi Provincial Office of Education; Gyeonggi Education Research Institute
Period
July 2025–present
Keywords
education, assessment, academic integrity, AI grading
Context
In July 2025 the Gyeonggi Provincial Office of Education launched 'Hi-Learning', South Korea's first regional AI grading pilot for descriptive/short-answer questions in Korean language, social studies and science.
Results
A Gyeonggi Education Research Institute survey at Jeongcheon Middle School found 64.6% of 48 teachers (31) had used Hi-Learning; only 29.4% of the 17 users agreed it reduced grading workload. Documented errors included blank answer sheets receiving scores and essays over-scored for length alone.
Conclusions
A November 2025 promotional video mocking teachers triggered union backlash and a public apology from Superintendent Yim Tae-hee; despite unresolved reliability issues, the Ministry of Education has signalled interest in wider adoption.
Implementation
Indicative cost
Medium (€50k–€500k)
Time to results
Medium (1–3 years)
Staffing & skills
Gyeonggi Provincial Office of Education, Gyeonggi Education Research Institute
Conditions for success
replacing vague rubric language with quantified criteria, which the research institute found substantially improved accuracy
Common failure modes
blank answer sheets scored
essays over-rewarded for length alone
under-scoring of concise but correct answers
Commonly funded by
National / regional programmes
Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.
Do you run this practice?
Claim it —
verified implementers get a public contact pathway and can propose corrections.
Data sources
Where this practice's information was retrieved from, and when.
