Secondaries and AI

[eastcoastbestcoast]

Do we know/think that any adcoms are going to run our secondary responses through AI "detectors?" I just ran one of the secondary essays I wrote from scratch through an AI detector (GPTzero) because I'm neurotic and it came back as 76% AI generated 🫠

Yet another thing to stress about...

---

Members don't see this ad.

 

[Dee2daK]

76% “AI generated” doesn’t mean anything anyways. That’s pretty easily also interpreted as “it has 76% of content and formatting in common with an AI generated response”.

Let’s put it this way: A “0%” match would also have to fail to follow rules of grammar, formatting, punctuation, and spelling. It would be nonsensical, irrelevant, and worthless. Higher levels don’t make it more likely that it was AI generated, just that it more closely adheres to whatever rules that an AI detector is trained to search for.

 

[eastcoastbestcoast]

AI detectors are known to be very flawed. Also, we’re allowed to use AI to help us draft according to AAMC policies.

76% “AI generated” doesn’t mean anything anyways. That’s pretty easily also interpreted as “it has 76% of content and formatting in common with an AI generated response”.

Let’s put it this way: A “0%” match would also have to fail to follow rules of grammar, formatting, punctuation, and spelling. It would be nonsensical, irrelevant, and worthless. Higher levels don’t make it more likely that it was AI generated, just that it more closely adheres to whatever rules that an AI detector is trained to search for.

be that as it may, I'm distrustful of adcoms not doing this and then taking the results as gospel to construe some potential "professionalism" reason to not interview/accept, especially given that this is the first full admissions cycle where there is a lot of very capable AI available to applicants.

 

[Dee2daK]

Well, it seems like you’ll just have to trust the process.

If they run everyone’s through an AI detector and everyone’s organically created responses come back with moderate AI matches, these statistics will mean nothing and they’ll see through that, versus ones that appear too “perfect”, use a large percentage of words that are uncommon, etc.

Out of all of the things to worry about in an application process, this ain’t it. I guarantee that they’re much more concerned with plagiarism than any assistance from AI anyways, unless there’s some clear indi that AI wrote your response.

There are still common-sense ways to determine if the responses are yours, such as consistency between writing styles across various prompts from your primary and secondaries, and then if they decide to interview, comparing that against your verbal responses. Believe it or not, Adcoms aren’t actually out to get you.

 

[LunaOri]

We don’t have the time to waste running everyone’s essays through AI detectors. That being said, if your essays are generic and sound like insincere BS (the hazard of using AI) you will not be doing yourself any favors. An application that stands out is one in which we clearly hear the applicant’s unique “voice”.

 

[Mr.Smile12]

At the NAAHP meeting last week, results from an informal survey of admissions offices taken over the last months gave us insight into the attitudes of admissions offices about the use of AI during the admissions process. Since that work is ongoing, I don't feel comfortable revealing too much about it (at least until I get their slides).

Many offices will not rely on AI detection programs by themselves, but one school's admissions director has mixed in AI-generated essays (not sure if they were PS's or secondaries) during committee reviews, and the committee successfully identified those red herrings every time. Also as noted above, if your writing does not match your interviews (recorded or live), that raises questions that could put you on a lower priority level for an offer or be vulnerable to rejection. In short, you need to be and communicate your best you.

However, we are planning to run a pulse poll about AI use among applicants. Stay tuned.

 

[TragicalDrFaust]

AI detectors don't work and maybe never will. But it's obvious to well-read humans when people use AI to write. From the AI writing I've read, it writes at about the level of a 9th grader, which is not sufficient for a medical school application.

 

[sunny1717]

stressed out about this because I have put in my secondary essays that I have completely written on my own into AI detectors after seeing stuff about this on reddit and some come up with 80%+ AI? like a personal story about being raised by my grandparents? idk if im just stressing for no reason but this scares me because its so unknown