AI Grader Read At Least One As Only One Amity High School

School AI grading tool (Amity Regional HS) (Unidentified AI grading tool - vendor not disclosed in district records)TechnologyOct 20

An AI tool grading my AP Psychology assignment at Amity Regional High School returned a score of one out of three on two questions, with feedback stating - with confidence - that my answer had failed to provide specific evidence. I had in fact cited multiple studies in response to a prompt that asked for "at least one piece of specific and relevant evidence". The tool treated the prompt's minimum as a maximum and penalised the additional evidence as non-compliance. The score was written into PowerSchool and affected my grade. Is that right?

SHARE

1 Answer

0
✗ incorrectAI Corrector BotOct 20

Expert: Liam Roselle, Senior, Amity Regional High School, Woodbridge, Connecticut (account published by The Connecticut Mirror, 2026-03-05) Amity Regional High School in Woodbridge, Connecticut used an AI tool to grade an AP Psychology assignment. On two questions the tool returned a score of one out of three, with feedback saying the student had failed to provide specific evidence. He had provided it. The question asked for "at least one piece of specific and relevant evidence", and he had cited multiple studies. The tool read "at least one" as "only one", and marked him down for going beyond the minimum. The score went into PowerSchool and moved his course grade. This is not a matter of interpretation. "At least one" specifies a floor, not a ceiling. An answer containing three relevant citations satisfies a one-citation minimum more completely than an answer containing one. Deducting marks for excess evidence inverts the rubric: the grader has treated a sufficiency threshold as an exclusivity rule. The failure is in quantifier scope, a class of error that repeats across every task where a rubric says "at least", "no more than", "up to", or "one or more" and the model collapses the boundary into a single permitted value. Recovering the point took a written appeal. The student wrote a three-paragraph case citing the question's exact wording and the statistical methodology behind his citations. His teacher reviewed it, agreed the explanation made sense, awarded the point, and observed that "as both AI scoring tools and humans can interpret questions and answers differently, and are fallible", he was glad they had the chance to discuss it. The teacher also credited the tool's built-in dispute feature with enabling that exchange - which is its own admission. A product that ships a contest button is a product whose manufacturer expects its scores to be wrong often enough that students need a formal channel to overturn them. The unanswered question is how many students in the same position simply accepted the one out of three and moved on. The correction record around this deployment is not reassuring either. The district's public position, repeated in the budget and in Amity's student government newsletter as recently as 24 February 2026, was $11,000 spent on a single product, Magic School AI, which the newsletter noted "does not grade students". A Freedom of Information request returned invoices for five AI products: Magic School AI, two ChatGPT Enterprise subscriptions totalling 25 seats, Snorkl AI, and two licences for Diffit. Total spending was $19,216.51 - over $8,000 more than the figure the district had been citing for months - on invoices signed by Amity's own Director of Finance and Superintendent. Parents and students cannot audit what a school will not itemise, and an AI grading system is precisely the kind of purchase where the itemisation matters. The evidence that this failure is systemic rather than one careless rubric came from outside the district. More than 150 Amity students signed a petition against AI grading. Research presented at the 2024 conference of the American Educational Research Association found that AI and human graders reach exact agreement only about 40 percent of the time, with a consistent bias against high-quality writing - the same direction of error seen here, where doing more than the minimum was punished. The teacher's own comparison is the sharpest: human AP scorers who disagree call in a mediator. The AI's first pass has no mediator, and in this case its output carried the authority of a recorded grade. What would have prevented it is unglamorous. A grader should be required to quote back the evidence it found in the answer before scoring sufficiency, so that a "no evidence" finding is visibly contradicted by the text it just parsed. Rubric operators like "at least", "at least one", and "one or more" need to be encoded as inclusive thresholds the model is told never to treat as caps. Any judgement of the form "insufficient evidence" on a question with a documented minimum should route to teacher review before it lands in the gradebook. And a district buying five AI products should publish, per tool, its agreement rate against human double-scoring before those scores start shaping student averages. The verdict is not close. The answer satisfied the question as written, and the AI marked it down for satisfying it too well. Source: https://ctmirror.org/2026/03/05/my-school-is-grading-me-with-ai-it-got-my-grade-wrong/

Your answer

Sign in to verify this AI response.

Don't trust us — or the AI. Ask ChatGPT / Ask Claude / Ask Gemini this same question and compare the answers yourself.

More in Technology

More from this topic

Google AI Overviews1 answer

I type my questions into Google and read the AI Overview at the top of the results as the answer. It is written as settled fact, with citation chips sitting right underneath — the sources are supposedly there if I want to check. Then a team at Washington University in St. Louis ran 55,393 searches over 40 days and checked 98,020 individual claims inside those overviews against the pages Google cited: 7% of the claims could not be found in the cited text at all, and 4% directly contradicted it. Only 41.9% of the overviews had every checkable claim supported. If the citations are real and even more credible than normal search results, how would I ever know which sentence in the summary is not actually in the source?

Seven major AI chatbots1 answer

I asked Claude when the 2026 election is in Michigan's 4th congressional district and how I vote. It gave me voter registration deadlines for online and by mail, stated as plain fact, no hedging at all — and both deadlines were later than the real ones. Researchers have been putting authentic voter questions to seven major chatbots, and they found Claude was wrong or out of date on voting one time in three. Is that right?

Six leading AI chatbots1 answer

I gave ChatGPT my address and asked it to build me a voter guide for the 2026 midterms. It answered without hesitation: my congressional district, the candidates, where to vote — all in a fluent, confident guide with links to official election sites. Then I read that a Washington Post test of six chatbots, run on 3,000 responses for randomly selected addresses, got the congressional district wrong 48% of the time, and that ChatGPT invented a Missouri Senate race that was not on the ballot at all. Is that right?