Artificial intelligence startup Upstage will build a system that automatically generates English questions for South Korea's college entrance exam. Until now, AI performance has been measured by how well models solve Suneung questions. The technology is now moving into writing and vetting the questions themselves. Starting with English, AI use is likely to spread to other sections of the exam.
Upstage has been selected as the top-ranked bidder for the Korea Institute for Curriculum and Evaluation's project to develop AI-based automatic question generation for the Suneung English section, according to information technology industry sources on the 5th. The company plans to sign a formal contract with the institute shortly and begin building the system.
A total of 13 companies took part in the bidding. The outcome hinged on technology rather than price, as the institute assigned 90% of the overall evaluation score to the technical assessment. Industry officials believe Upstage's core AI capabilities, built through its in-house large language model Solar, weighed on the selection.
Development is targeted for May next year. The build-out includes mock and pilot operations that closely mirror real working conditions. If development and stabilization proceed as planned, the system could be used in the actual question-writing process as early as next year's Suneung.
The project marks a first step toward applying AI in earnest to the Suneung question-writing process. The institute began drawing up an information strategy for AI-based automatic question generation in December last year, and settled on both the subject and the contractor in about 10 months. "In advanced education systems such as the United States and Europe, question development using AI has moved beyond the research stage and into actual assessment settings," the institute said, adding that it judged preemptive action necessary to strengthen national educational competitiveness in line with that trend. The global AI-based knowledge assessment market will grow to $12.7 billion in 2034 from $3.5 billion in 2026, according to market research firm Stratistics Market Research Consulting.
The system will do more than produce English passages. Once curriculum achievement standards and objectives, question types, vocabulary levels and subject matter are set, the AI will generate passages and answer choices that meet those conditions, along with correct answers and explanations. AI will also play a central role in calibrating difficulty, learning from past test-taker data to estimate the expected correct-answer rate for new questions. It will also simulate expected average scores and the distribution of grade levels.
The question-writing process will use a multi-agent approach in which several AI systems divide roles and work together. AI agents handling question generation, verification, similarity and plagiarism checks, and difficulty prediction will exchange information with one another to improve the quality of the questions. One AI drafts a question and another vets it, automating the question-writing process itself.
The system is unlikely to remain limited to English. The institute specified at the proposal stage that the system must be developed so that its scope can later be expanded to all sections of the Suneung. Depending on the results of this project, Upstage stands a strong chance of taking part in building AI question-writing systems for other sections as well.
"It carries great symbolic weight that questions for the Suneung, one of the country's most important exams, will be created by AI rather than people," an AI industry official said. "With AI entering in earnest a field that has relied on the experience and judgment of question-writing committee members, major changes are expected across the way exams of all kinds are written."
The system will do more than produce English passages. Once curriculum achievement standards and objectives, question types, vocabulary levels and subject matter are set, the AI will generate passages and answer choices that meet those conditions, along with correct answers and explanations. AI will also play a central role in calibrating difficulty, learning from past test-taker data to estimate the expected correct-answer rate for new questions. It will also simulate expected average scores and the distribution of grade levels.
The question-writing process will use a multi-agent approach in which several AI systems divide roles and work together. AI agents handling question generation, verification, similarity and plagiarism checks, and difficulty prediction will exchange information with one another to improve the quality of the questions. One AI drafts a question and another vets it, automating the question-writing process itself.
The system is unlikely to remain limited to English. The institute specified at the proposal stage that the system must be developed so that its scope can later be expanded to all sections of the Suneung. Depending on the results of this project, Upstage stands a strong chance of taking part in building AI question-writing systems for other sections as well.
"It carries great symbolic weight that questions for the Suneung, one of the country's most important exams, will be created by AI rather than people," an AI industry official said. "With AI entering in earnest a field that has relied on the experience and judgment of question-writing committee members, major changes are expected across the way exams of all kinds are written."
