Evidence has emerged that one of the candidates for the New Rochelle Board of Education may have used artificial intelligence (AI) to respond to questions about racial disparities posed by New RoAR News last week. 

The candidate, Jessica Klein, told New RoAR News, “I stand by my answers to your questions. The thoughts and content presented are entirely mine through research and studies I conducted in the past.”  Although she was asked directly whether she had used AI to generate her answers, she did not directly deny having done so.

BOE Candidate Jessica Klein

Klein is a teacher whose website claims “deep roots in public education and 25 years of teaching experience in Westchester.”  According to her website, she attended public schools, but her LinkedIn profile shows that most of her teaching experience has been at the Westchester Day School, a private school in Mamroneck.  

Several readers expressed concerns about Klein’s response to the second question posed by New RoAR News, concerning proposals to address persistent racial disparities in achievement, opportunities, and discipline in the New Rochelle public schools.  These readers suggested that Klein’s answer might have been written by an artificial intelligence program because of its generic tone and content.

To test that suggestion, New RoAR News submitted all five candidates’ answers to Questions 1 and 2 to GPTZero, a free online AI detector application frequently used in academic settings to identify text written by artificial intelligence software.

While there are many AI detector software packages available and their quality and accuracy vary widely, GPTZero is generally recognized as one of the most reliable of these packages, with a conservative algorithm that tends to attribute writing to a human author when results are ambiguous.

Of the ten writing samples submitted to GPTZero, seven were classified by the software as “entirely human” with a high level of confidence, and one was classified as “uncertain” but “likely human.”  Both of the responses submitted by Klein were classified as “AI generated” with a high level of confidence.

GPTZero analyses of Jessica Klein’s answers to New RoAR News’s Questions 1 and 2

AI detectors use a variety of methods to identify text generated by artificial intelligence.  GPTZero uses two of the most common criteria, “perplexity” (unpredictability) and “burstiness” (variation in sentence length and structure) and compares these measures to known samples of human and machine-generated text. 

A recent review of seven “best” AI detectors reported that GPTZero “showed high sensitivity, sometimes detecting AI influence even in heavily human-edited texts” and produced “occasional false positives in fully human-written texts, but these were minimal.”

A formal study of GPTZero in a medical journal in 2023 found that the software correctly identified 90% of human-written scientific text samples as human-written and 65% of AI-generated passages as machine-written. 

Given the 10% failure rate of GPTZero in identifying human-written text, it is impossible to state definitively that Klein did not write her own responses to the questions from New RoAR News.  However, the AI detector’s high confidence that both her responses were AI-generated and that 7 out of the 8 responses from other candidates were human-written, as well as Klein’s failure to definitively deny the allegation, raise serious concerns.

The GPTZero analyses of the other candidates’ responses are reprinted below:

Question 1

Rosa

Question 2

Myriam

Elana

Rosa

Keith