Canfield ’28: Admissions is wrong about AI checkers

Thank God ChatGPT was only released 11 months before I applied to Brown. At the time, its writing skills were nowhere near where they are today, and I was not tempted to worsen my college admissions essays with its sloppy prose. I cannot say the same for today’s high school seniors

I pity those who are wrestling with the moral quandary of using artificial intelligence in their college applications. Many may embrace the tool purely out of fear that their peers will use the technology to their advantage. Others point out that wealthier students have access to paid college counselors and admissions coaches who edit and maybe even write their essays for them. Could AI be the great equalizer in college admissions?

I certainly don’t have all of the answers to these questions. But I do know that AI strips the very humanity from our college essays that they are meant to communicate. The irony in outsourcing a personal reflection to a machine is obvious, but it will not stop students from taking this route. Brown claims that it prohibits all AI use in admissions materials. But last week, The Herald reported that Brown’s admissions department does not use third-party AI checkers when evaluating applications. This leaves the University with little recourse to protect the integrity of its admissions. By holding off on using AI detection to absolutely avoid mistakenly rejecting an applicant for using the technology, the University fails to enforce its own policy and indicates a lack of understanding of how important it is that we do not fill the undergraduate student body with a cohort of cheaters.

Let’s do some math. As of July, the best AI detection software on the market, Pangram, boasts a false positive rate of 0.0041% — a University of Chicago study also found that a previous version of Pangram achieved near-zero error rates. Say 30% of college applicants use AI to assist with their admissions essays, as found in one 2024 report, then out of the 249,950 essays 100 words or longer from students who applied to Brown’s class of 2030, about seven of them would get falsely flagged as being AI-generated. Only seven.

The University would likely argue that they wouldn’t want to risk rejecting even one student for a false positive. “We would never want to assume that someone who just prefers to use em dashes in their writing copied the content of some large language model,” Associate Provost for Enrollment and Dean of Undergraduate Admission Logan Powell recentlytold The Herald. But the potential consequence of not using an AI checker to screen admissions essays vastly outweighs the unfortunate byproduct of rejecting seven potentially qualified applicants.

While it is unfortunate that these seven hypothetical students would be unfairly punished, this random bad luck happens in admissions all the time. Admissions officers spend very little time on each application, perhaps as little as eight minutes by some accounts. Even the most seasoned officer is bound to make more than seven mistakes when reading thousands of applications. We allow human error in college admissions; why not near-zero machine error? 

The reality is that if we do not use AI checkers, Brown will admit students who are unqualified and morally unfit to attend the University. Writing an admissions essay with AI outsources an applicant’s critical thinking to a machine. Applicants will not have the opportunity for personal reflection and to demonstrate their ability to think through and write a compelling argument of their own — a skill that is critical for a liberal arts education. Brown’s first institutional value is the pursuit of knowledge and understanding “by educating students duly qualified for discharging the offices of life with usefulness and reputation.” I fear what will come of our school if we cannot determine the integrity of an increasingly large number of our applicants. A generation from now, will students admitted off the labor of a machine still be qualified to discharge the offices of life with usefulness and reputation? 

Critics of AI detection software argue that it is unreliable and inaccurate. A study from 2023 found that “available detection tools are neither accurate nor reliable.” Another 2023 study showed that detectors had higher false-positive rates for text written by non-native English speakers, a troubling reality for a University in which international students comprise 14% of undergraduates. Screenshots of AI detectors flagging the U.S. Constitution for being AI-generated have gone viral on social media — did James Madison have early access to Claude? But just as AI has improved rapidly in the last few years, these detectors have also made progress. A more recent 2025 study found that AI detectors “effectively distinguished AI-generated content from human-written texts.” 

If Brown is not using every resource at its disposal to protect its admissions process, then it is failing. There have always been applicants who use outside resources to write their essays for them, and admissions officers have to rely on their experience and intuition to judge if students’ words are truly their own, a less-than-ideal solution. Unfortunately, no software exists to detect essays written by a paid admissions advisor. But it does exist for AI, and it is foolish not to use it.

College admissions is imperfect — that does not mean that we should give up on it. 

Ethan Canfield ’28 can be reached at ethan_canfield@brown.edu. Please send responses to this column to letters@browndailyherald.com and other opinions to opinions@browndailyherald.com

Source: www.browndailyherald.com