When AI takes the first pass, who keeps the final say?
AI use described for first-pass essay marking in South Korea and First Orion's quality-assurance testing puts the human role in setting criteria, review, and responsibility in focus where automation produces the first draft.
Artificial Intelligence··Midday
A first draft for essay marks
South Korea's National Education Commission presented a proposal to use AI in the first pass of marking essay answers at a public hearing in Incheon on August 11. According to Hankyoreh, it was offered as an answer to the fairness objection that has blocked essay questions in large examinations. Lee Min-seok, who leads the education panel of the national AI strategy committee, proposed starting with lower-stakes school assessments and then extending use to higher-stakes examinations such as the CSAT university entrance test. In the described method, teachers and specialists prepare the questions and marking criteria. The model re-marks sample answers so the criteria can be calibrated, then produces a first draft of scores after the exam. A teacher reviews and finalises that draft. Lee said final authority over questions and assessment remains with the teacher, and that time saved could be used for individual feedback and lesson design. The hearing series continues, while a draft national education plan is expected at the end of October.[1]
From script to web interface
A joint post by First Orion and AWS describes a different task: moving quality assurance from script-based test automation to agents built around Amazon Nova Act. First Orion says its branded communications services reach hundreds of millions of calls across carriers in the United States, Canada, the United Kingdom, and Germany. In the architecture the company describes, agents handle testing by reading web interfaces in a way a person would. The stated trigger was engineering shipping faster than the quality-assurance team could test. The source explains how the system was built and the results reported by the company; it does not offer an independent measurement of how testing work was redistributed among teams. The report therefore describes the company's path for changing test automation, not automation taking every testing decision. As in the education example, the question of what standard makes an AI output sufficient remains an operational question separate from the system's technical description.[2]
The line between draft and decision
The two examples do not show people withdrawing completely from work in which AI takes an early step. In the South Korean proposal, teachers and specialists establish criteria, the model prepares an initial score, and a teacher finalises the result. In First Orion's example, the company uses agents to keep up with the speed of quality-assurance testing; the source does not detail the management or review chain through which test results are accepted. That difference matters. The education proposal describes a review role explicitly, while the architecture account for quality assurance does not substitute for a description of review arrangements. That detail also shows that technical transformation does not by itself explain the daily division of work. The sources do not say the two applications are governed by the same rule or carry the same level of risk. Their simpler common point is that the tool producing a first draft need not be the person or institution carrying responsibility for the result. As automation spreads, who sets the criteria, who can correct an output, and who makes the final decision become as consequential as the application itself. These questions also matter when a path for correcting an error is set.[1], [2]
