HOT TAKE: AI should write Letters of Recommendations
During my time as a high school senior, I noticed how much work counselors take on during the college application cycle. In my graduating class, there were over 700 students, which also means over 700 letters of recommendation. The purpose of this letter is to show college admissions officers something that the transcript, activities list, and test scores can’t: a unique account of the student from a person who knows them well. Given the number of seniors and the number of counselors writing these recommendations, getting to really know each senior and writing a thoughtful and unique letter of recommendation is an extremely difficult, if not impossible, task.
I then asked myself, "Why aren’t high school counselors using AI more extensively to write these letters of recommendation, and why is there such a bad stigma against it?” Although there are privacy and legal concerns with AI (ex. FERPA), I believe most counselors don’t use AI more extensively in their writing process because of authenticity, or lack thereof. This letter should capture who a student is, and current AI tools can’t do this effectively. To admissions officers, it is not the act of using AI to write a recommendation that causes red flags; it is the generic and often bland output that comes from poor use of AI (the exact problem with templates as well).
However, counselors were never meant to spend all of their time on the computer; they were meant to be one of the people who know the students best! If counselors spend 1 hour talking with students for every 4 hours of writing, with my project, I aim to make that 4 hours of talking for every 1 hour of writing.
With a few of my friends in college, we are creating software that helps solve the exact problem stated above. We want to integrate AI into the process of writing college letters of recommendation to help counselors have the time to get to know their students better while also being able to write unique and authentic pieces of writing.
Our software is a two-step process: reading and writing. Given the student source material (meeting transcript, brag sheet, etc), our software generates provocations in the margins of these documents. These are annotations that are meant to provoke ideas and thoughts while you read, not to summarize. As you read, you can reply to these provocations, highlight, and leave short memos: anything that comes to mind. The next step is the writing process. Our software compiles all of your input into a first draft, where the process of provocations, input, and drafting can continue. What we do and what other AI tools have failed at is to have every sentence in this piece of generated writing be 100% from your memos and thoughts, not the AI, all while saving the writer the time of physically typing every word. At core, our philosophy is that AI is a good provocateur and also good at clean, logical, and persuasive writing. Letting AI do this lets humans, counselors, do what they are good at: finding a unique story or perspective for a student. Throughout this process, the LLMs used would be pre-approved by the school and would be completely secure models that never release student information.