Executive summary

AI-assisted grading can reduce repetitive work and improve consistency when rubrics are well defined and teachers retain final authority. This whitepaper describes a staged rollout: pilot design, rubric alignment, human-in-the-loop review, and safeguards for academic integrity.

1. When AI assistance adds value

Long written assignments, structured rubrics, and large cohorts are strong candidates. AI is weaker where holistic judgment, creativity, or nuanced disciplinary norms dominate without clear criteria.

2. Rubric design

Break criteria into observable behaviors. Each level should include exemplar language. Ambiguous rubrics produce inconsistent AI suggestions and undermine teacher trust.

3. Human-in-the-loop

Teachers should always confirm or edit marks and feedback before release to students. The platform should log AI suggestions separately from final grades for audit purposes.

4. Integrity and bias

Monitor for disparate impact across cohorts, document known limitations, and train staff on when to override automated suggestions. Pair AI tools with existing plagiarism and misconduct policies.

5. Rollout checklist

• Select one department for a 6–8 week pilot
• Align rubrics with learning outcomes
• Train staff on override workflows
• Collect teacher and student feedback
• Expand only after measurable time savings without quality complaints

Full version to be published after pilot results are available.