Who's Grading You on Coursera? The Shift from Human Peers to AI (www.classcentral.com)

Coursera is quietly shifting assessment from human peer review toward AI-driven grading: peer-reviewed tasks have fallen from ~39% of courses in 2023 to about 25% now, only 10% of new courses added peer grading last year, and many legacy peer reviews are now optional. AI grading, introduced in late 2024, delivers near-instant results and reduces wait times and some inconsistency; the author’s experiments showed AI reliably failed minimal or empty submissions and allows learners to request manual reassessment. However, longstanding peer-review problems—variable quality, gaming, plagiarism, and delayed feedback—aren’t fully solved, and it remains unclear how well AI systems detect generic, plagiarized, or AI-generated responses. For the AI/ML community this shift is significant: it creates a large-scale, production use case demanding robust, explainable automated evaluation of open-ended work. Key technical needs include reliable rubric alignment, adversarial robustness to gaming and AI-generated content, plagiarism detection at scale, fair/transparent scoring, and human-in-the-loop dispute workflows. There are trade-offs between preserving credential credibility (effortful assessment) and scalability/cost, so researchers and engineers should focus on hybrid systems that combine automated grading speed with targeted human oversight, explainability for appeals, and continuous evaluation of model performance on subjective assignments.