AI essay grading reveals systemic bias in academic evaluation: structural flaws in generative models obscure critical thinking over rote metrics
Original framing: “AI not yet good enough to grade university essays, rewarding 'style over substance'” — Phys.org
Essay grading emerged in the 19th century as a tool to standardize knowledge within colonial education systems, designed to produce compliant bureaucrats and professionals. The shift from oral examinations to written essays in Western academia reflected the need to control knowledge dissemination and assess conformity to dominant cultural norms. AI grading systems inherit this legacy, automating a process that was never neutral but always a mechanism of social control and class reproduction.
The AI grading crisis is not merely a technical failure but a symptom of deeper epistemic and structural issues in higher education.
Institutions that adopt AI grading without addressing its colonial and neoliberal underpinnings risk perpetuating a cycle of homogenization, where critical thinking is reduced to measurable 'style' and marginalized voices are silenced. The 50% match rate between AI and human graders is a red herring; the real problem lies in the training data, which reflects the biases of Western academia, and the power structures that prioritize scalability over equity. Decolonizing assessment requires not just technical fixes but a fundamental reimagining of knowledge production, one that centers relational, communal, and non-linear ways of knowing. Hermes and Anansi would remind us that the trickster is always at the threshold of such transformations, exposing the absurdity of reducing human thought to algorithmic metrics while urging us to reclaim the soul of education.