This House would ban the use of AI in K-12 grading.

edu

Algorithmic essay grading has been piloted in three U.S. states. Results are mixed: faster turnaround, more consistent within-grader scoring, but the systems inherit and amplify their training-data biases.

Background

Utah and Ohio piloted AI essay grading on state writing assessments in 2022-2023. The systems (ETS's e-rater, MI Write) match human inter-rater reliability around 0.85-0.92 for the kind of structured writing on standardized tests but underperform on creative or argumentative prompts. The 2024 Texas STAAR controversy, where students discovered they could trick the AI grader with keyword stuffing, exposed the kind of failure mode standardized testing has always had at scale, now executed faster. Human grading has its own consistency problems; the comparison is not between AI and a perfect grader.

Government opens with

Children's assessment data is too sensitive to outsource to a black box that schools cannot audit.

Opposition responds with

Human graders are not consistent, not fast, and not unbiased either. The AI tradeoff has to be measured, not assumed.