If Testing Companies Use AI to Grade

Source: Nick Potkalitsky Substack
Author: Nick Potkalitsky
Original source: https://nickpotkalitsky.substack.com/p/if-testing-companies-use-ai-to-grade

Published: 2026-02-18
Source type: essay
Private backup: the full article text is archived in the private repository at archives/articles/nickpotkalitsky-substack-com-if-testing-companies-use-ai-to-grade.source.md. It is not published on the public Quartz site.

Summary

Nick Potkalitsky investigates what it means when standardized tests or districts use “AI” to grade student writing. He distinguishes Ohio’s operational use of discriminative AI, which classifies existing writing under human-validated scoring systems, from generative AI tools like ChatGPT, which produce new text and remain unreliable for grading. The article emphasizes prompt sensitivity, model drift, inconsistent harshness or leniency, and potential bias against English learners when training data is insufficient or unrepresentative. Potkalitsky argues educators should ask precise questions about AI type, training data, validation, human oversight, and which students might be disadvantaged.

Pull quotes

The Wrong AI Question

“What struck me wasn’t the concern, that was reasonable. It was that we were all using “AI” to mean completely different things.”

Not ChatGPT

“Yes, Ohio uses AI to score writing on standardized tests. But it’s not ChatGPT, and it’s not what most people imagine.”

Different AI, Different Risks

“Same AI family, completely different jobs.”

Protecting Students

“The technology isn’t going away. Our obligation is to understand it well enough to protect the students who encounter it.”

Big ideas

Claims

Key evidence and examples

  • Ohio’s system is described as a hybrid human-AI process rather than ChatGPT-style grading.
  • The article contrasts operational scoring systems with generative AI experiments that are sensitive to prompts, model versions, and scoring persona.
  • It cites concerns that English learner essays can be scored lower than native-speaker essays judged equal by human raters.
  • The article calls for documentation, validation, representative data, bias testing, and human oversight before consequential grading use.

Education relevance

Highly relevant for standardized testing, district AI procurement, writing assessment, English learner equity, and responsible AI governance in schools.

Durability note

The policy details may change as testing vendors update their systems, but the core distinction between discriminative scoring systems and generative AI remains a useful guardrail for school conversations about automated assessment.

My notes