AI has quickly changed college admissions. Students turn to bots to help write their essays, making it harder for admissions folks to get a sense of a student’s voice and thinking. And some colleges embrace AI as a way to automate the evaluation of admissions essays, leading some to worry about the signal that sends to students. On this episode, Jeff talks to a veteran college enrollment official who is leading an AI experiment at his university, and an admissions consultant who has changed her mind about what she advises applicants about using AI.
On this episode, Jeff talks to a veteran college enrollment official who is leading an AI experiment at his university, and an admissions consultant who has changed her mind about what she advises applicants about using AI.
Related links:
“Should AI score admissions essays?” by Jeff Young in The Chronicle of Higher Education
“I’m a college admissions counselor. I’ve changed my mind about students using ChatGPT,” by Sophie Sajnani in The San Francisco Examiner.
“What will be scarce?” by Alex Imas on his Ghosts of Electricity Substack.
Thanks to this week’s sponsor, Studiosity.
Sponsor:
This episode is brought to you by Studiosity.
Real learning happens through real thinking and real experiences. It's something that takes effort, not shortcuts.
For over 20 years, Studiosity has partnered with educators to protect the thinking, the essential work that the student must do for themselves, not detection, not an us versus them culture in teaching and learning, just the real journey behind every degree that is worth holding.
They're doing it for the institutions that trust their students, for the institutions that back their teachers to teach, not police, and for the institutions that stand behind their degrees.
Keep the learning real with Studiosity.
Get the walkthrough at studiosity.com/learningcurve.
Jeff Young:
Hello, and welcome to Learning Curve, a look at education struggle to adapt to the AI era.
I'm Jeff Young, a longtime education journalist.
The college admissions process has long been stressful for students, and these days the rise of generative AI is adding a new layer of anxiety for students applying to selective colleges. It was already hard to tell how to make your application rise above the others, and how the colleges were deciding who gets each coveted slot.
These days, students applying to college are also wondering whether they should use AI to polish and improve their admissions essays. Will doing that show colleges that they're able to use the latest tools to better make their case, or will the college ding their application if they sense a bot is doing the writing?
And to complicate things, a growing number of colleges are turning to AI themselves, using new AI tools to summarize applicant essays and even score them.
For this episode of Learning Curve, I'm diving into this issue of how AI is changing the college admissions process. And we'll hear from two experts today: a veteran college enrollment official who's leading an AI experiment at his university, and an admissions consultant who has changed her mind about what she advises applicants to do when it comes to AI.
First, I was curious to hear from a selective university that has gone all in with AI in admissions. That institution is Virginia Tech, where, for the first time this admission cycle, AI became a major part of screening admissions applications. And here is a part we'll dig into: the hope is that AI will free up time for human professors and admissions folks to interact more with students to learn all about that.
I connected with Juan Espinosa, Virginia Tech's Vice Provost of Enrollment Management, and I started my conversation with Juan Espinosa by asking him to remind us how different things were just 10 years ago and how things were changing even before generative AI emerged.
Juan Espinoza:
It was a different time for sure when you think of just from a student perspective, of how many applications were being sent out.
I think the average, or you know, 20 years ago was three to five, and if you rewind to that time, you often heard of strategies shared by counselors in which you should have one reach school, one school that you think you were well positioned for, admissions wise, and then the safety school, and so that usually was three. And if there was any sort of, you know, question on where one of those schools, you know, fell into that, then that's where you would add another school, just kind of, you know, make sure you were well represented in that area.
And so that's where you got to that higher end of five schools, kind of what it is. And the other day I spoke to a student, I guess it's been a few months now, and I asked, you know, how many applications do you send out. And they said I sent out 35 applications. And that's not rare to hear anymore, because a lot of colleges, and this is a good thing, have been focusing on trying to streamline their application process, and now you have to make it easier
And you have shared application platforms now that are readily available that make it really easy to send applications, and now you have a newer thing in the last few years, direct admission, where you're getting offered before you even submit an application.
Jeff Young:
Yeah, these students haven't even applied, and they're getting a note saying you're in,
Juan Espinoza:
Yeah, you're in, you just gotta, you know, circle back and give us a little bit more, you know, now you should apply, yeah, and so I think, and it's so, it's so interesting too.
From a student perspective, so my kid went through this process, a couple years ago, she's a sophomore at James Madison University, which is just right up the road from where we are in Blacksburg, and it's just so interesting to hear the anxiety of students as they navigate through this process, because they're hearing from the institution side this is a record year of applications, year after year after year. Even with the whole enrollment clip conversation in the background, we're still, yeah, there's supposed to be
Jeff Young:
Less students, and yeah.
Juan Espinoza:
But we're hearing record application year after year. And it's because students are freaked out, and they think that it's really hard to get into college now, and well.
Jeff Young:
That's what they've been told, yeah.
Juan Espinoza:
And so this could hedge their bets and send more applications out, and it's easier to do that now more than ever before.
And so it's almost a self-fulfilling prophecy in a way, because as there's more applications going out, the offer rates are dropping.
Yield rates are going the other way, right? Because at the end of the day, you can only be at one place in the fall. And with the population declines that we're expecting, and we're starting to see this year, I think yield rates will continue to drop.
Jeff Young:
The yield rate just means how many accepted students ended up attending. In other words, what percent of the students who got a yes ended up showing up on campus?
One issue for colleges in all this is, how do you plan for your class when so many students applied but may not have that much interest in coming, because they have other schools that they're more excited about and might get into.
Juan Espinoza:
There's no normal year in admissions anymore. Ever since the pandemic, it seems like each year comes with its own set of challenges, and so it's just difficult.
And so, going, you know, to your initial question of rewinding, you know, back to when I first started in this field, it's a very different profession in some ways, not entirely, but when I think of kind of those revered names, you know, those admissions leaders that were so well respected across the country when I first entered the profession.
And I was lucky to work with one of them, Dave Krause. It was very heavily built on relationships, and that's not to say relationships don't matter today, and how we do admissions, but you saw kind of a transition from very heavily focused on relationships only to more of a data informed and even in some cases data driven focus in the admissions field, because you had to, with the volume of applications that were coming in, the changes in offer rates, changes in yield rates, all these new variables coming in from a data standpoint, which we honestly didn't have access to several years prior.
And so you had to start paying attention to this. And then it started creeping into how you predicted your class. And then you start having machine learning have a larger presence in some of your predictive modeling.
And so it ended up opening some doors for newer leaders in the admissions profession to have a very heavy data background to help kind of, you know, decipher through all this metadata, which all professions kind of went through that in some ways. How you're going to dig through this data and find relevant patterns that are going to allow you to put your institution in the best strategic position moving forward.
Jeff Young:
You say all professions have this, but I think it seems, it seems fair to say from watching college admissions in the last decade, I mean, enter the quants, like people that were like data interest, because that, because, like, you said, there was suddenly more data, more processing of data. To I want to first back up to the relationships.
Juan Espinoza:
Yeah.
Jeff Young:
Describe that relationship. What do you mean by that? I think some people may not understand which relationships you mean, and how that worked.
Juan Espinoza:
Yeah, when I first entered the profession, if you were going to make goal, you had to meet with students and find them where they were at. Relationships with counselors were critical.
Jeff Young:
So, do you are going to high school counselors?
Juan Espinoza:
High schools, you're doing high school visits, you're standing behind the table, you're at college fairs. That was a huge part, if not the biggest part of the strategy itself, and in many ways it wasn't data informed, a data driven to the level it is today, because it was more of, we had to be, we had to be at as many places as possible, right.
And then, as this more data-informed process came in, colleges started paying more attention to all right, where are we seeing some good responses from different markets when we're visiting those?
And then, with, you know, you kind of brought in that across the entire country, and in some cases across the world.
Then you start strategically doing some more surgical strikes on when you're sending counselors, and then when you start using data segmentation, you're able to look at, ‘hey, we need to build a larger presence, let's say in the Houston area.’ You're using data segmentation in a way of, hey, here's certain neighborhoods within that area that that align with variables that we think will lead to a higher mobility of students willing to go out of state and consider Virginia Tech, and so you just peel additional layer and additional layer until you really find a group that you're like, this is where we need to be.
We're in the old days when I first started, it would be like we need to build our presence in Houston. Let's go visit every high school in Houston and try to get traction. Now it's like we need to visit five schools in Houston, and you know, once they're done, they move on to another city, because the data shows these are the schools that are going to have the most traction in our messaging and in our program offerings, and that's who we need to talk to, and then move on to the next city.
Jeff Young:
I'm not sure the average student applying to college these days understands how much number crunching goes into how the colleges are seeking them out, I noticed that Juan used the term surgical strikes to describe their data use. It's probably not a perfect comparison, but I keep thinking about that baseball book, Moneyball, which outlines how more sophisticated use of statistics by teams like the Oakland A's revolutionized scouting in baseball in a way that has been happening over the years in admissions, and that was before generative AI.
So, for years, so for years before generative AI, colleges were looking for ways to use earlier forms of AI algorithms to predict which applicants who applied would actually accept if given an offer. Meanwhile, there have been other changes since 2018 Virginia Tech revamped its application to try to get to know its applicants better, adding four short essay questions to their application. The questions center around the university's motto, Ut Prosim, which is Latin for that I may serve.
So one question is ‘Share how you contribute to a community that is important to you.’ Another is, ‘Share a time when you were most proud of yourself, either as a role model or when you displayed your leadership’
The answers must be really short, each one is limited to 120 words, just 120 words to answer each of these four questions.
Juan Espinoza:
Give the student an opportunity to tell their story, tell, tell where they were coming from, and you know, talk about some of their experiences, and that was so important, because at that time we only had about 30,000 31,000 applications only, but the pool was very homogeneous, you know. Our average GPA was about a four zero at that time. Well, the average GPA for the applicant pool was about a 4.0 at that time.
We were getting applications from students that thought they could get in, and so we needed an additional lens to help differentiate through that pool, and the non-cognitive attributes allowed us to do that. They were hugely successful. We saw great gains with retention, progression, graduation rates since the introduction of it. We felt that that was really the missing piece in our review process.
Jeff Young:
The downside of this approach is that having these four short essay questions, even though the answers are super short, it's labor-intensive to score. Since they adopted the system, it has taken a massive effort each year.
Juan Espinoza:
We trained about 200 to 300 readers that were representatives from our community to read the essay questions. So this is separate from our admissions committee. We had a whole group trained just to read essay questions, and we created a review platform where they would be able to see the question and the answer that was submitted, but no personally identifying information of any kind.
All right, to minimize bias, and we set that up, and the system we set up required each essay prompt to have two readers, and then if the difference of scoring was more than four points, a third reader would kick in, and then the scores would be average.
Jeff Young:
So there's a lot of human time, lot of human time at an online portal that just they crank through looking apps.
Juan Espinoza:
About 500,000 essay prompts needed to be read, as many application 500,000 counted how many times each essay had to be read in the average third reader that comes out to about 16,000 hours it was something we believed in something that showed was tied to our recent success but it was slowing us down so each passing year as our applications grew from 30,000 to the most recent year.
We're approaching about 60,000 We're about 58,000 applications. That was more and more essays to read.
Jeff Young:
So, when you double your applications, the old system was straining.
Juan Espinoza:
Yeah, it was straining. And so we had to make a decision: are we going to discontinue this because we were getting feedback from our applicant pool that we're taking too long. You're the last school in Virginia to let me know if I'm getting in or not. You're letting us know in March, and other schools are letting us know weeks ahead of time, and so it was putting us at a disadvantage, even though we believed it.
And around that time, one of our readers, our volunteer readers, was a faculty member, his name is Dr. Lewis Hickman. He came out and said, you know what, I got a lot of background in AI, which was this was about four years ago, give or take.
Jeff Young:
So, before ChatGPT was big,
Juan Espinoza:
Yeah, so you know it's kind of a foreign thing for a lot of people. So, I've got some research in AI, I'm a linguist by training, but I got a lot of background in AI. I think I can help with your problem, and try to create a AI companion tool with my research team.
At that time, you know, it was like, all right, let me help this guy out. He needs data, he's trying to do some research, we can partner on this. I honestly had no expectation this would be something we would ever implement, to be quite honest with you.
Jeff Young:
Oh, wow. So you're like, let's help the linguist, yeah,
Juan Espinoza:
Exactly. And so we engage with him. The first year we get results, and they're okay. And it was interesting because he was trying to create something that essentially was replicating a human reader, like, you know how. And there was, there was, he was part of a published article that, that was that was showing a lot of weaknesses in that approach, in that you know if you're training in an AI tool to try to be human, it's really, really difficult.
So the second year he moved to more of a tool that focused on patterns, combinations of words and this and that, you know what leads to a three, which is the highest score. What leads to a zero for the lowest score for each essay?
And we saw a significant increase in the productivity of that of that tool almost immediately. And then it became the cold question of how do you manage bias, right? And that's really tricky when you're dealing with AI, and a huge concern, and so it was quite brilliant. What he did, he moved to an ensemble of using LLMs, so just like when you ask only one person a perspective, then your answer is going to be heavily, you know, dictated by that person.
Well, you ask a group of people, you start limiting any sort of bias, and different perspectives are more varied and stronger for that. He did the same thing with LLM, so instead of disaster one LLM, he created an ensemble to get different perspectives to the point where, so
Jeff Young:
It's like, ask ChatGPT and Claude,
Juan Espinoza:
And so by doing that, you know, you remove the need to do any deep bias process at all, because it varied and showed no subgroups were at a disadvantage, was fantastic. It really simplified the work.
Now, I'll be clear, we don't put this in Chat GPT. These are on-premise LLMs that we have on campus, or we went through the whole division of IT, you know, audit, making sure that everyone felt comfortable. It was great. We weren't working with any vendors, so this all stayed within the Virginia Tech ecosystem from an IT standpoint. And by the third year, it was producing data we could not ignore anymore. It was doing just as well, if not better, than our human readers, when looking at disagreement rates, and it was able to average an essay read in 1.8 seconds versus the two minutes our human readers were taking.
Jeff Young:
So 1.8 seconds versus two minutes. Yeah, it's like the experience we all have when these, these chatbots spit out so fast, you're like, wait, you just wrote my, my paper. Yeah,
Juan Espinoza:
So it just got to the point of we got to do this.
The provost at the time was supportive of it. The president, Dr. Tim Sands, was like, ‘This is exciting.’
I know there's risk being the first in a lot of cases to embrace technology, but I think this is a win-win, and that's when we started realizing the question wasn't, are we going to do this, but how are we going to do it, and I felt strongly, so strongly, that we had to be transparent, you know. I think that there is such mistrust right now in AI that you're only going to contribute to that mistrust if, if you, as an institution, aren't being clear on how you're going to utilize it in the application process, so I know publicly we were essentially the first to announce we were using it, but I question if we truly were the first, if there were others that are already utilizing in some way, but kind of behind,
Jeff Young:
Sure, sure, but you want it, you thought it was important. To say this is what we're doing.
Juan Espinoza:
So important, because you know the admissions offices in general, they kind of have a reputation of being, you know, very vague on how things and decisions are made, you know, it's a lot of behind the curtain type stuff, and I really feel that since we've started working as a team here at admissions with that as a key principle, we try really hard to explain to the public how we make decisions, because it's our goal that expectations are aligned with the decision they ultimately get. If a student is surprised by their decision, in some ways that's a failing on our side and not communicating effectively how we make decisions and what variables are important to us, and in a lot of ways I feel that there's a high level of trust the student and really the family give when they submit an application to a college, that their application will be handled with care, and we need to honor that trust, so that that transparency piece was really important. The second pillar in all of this was making sure that we weren't simply handing this over to AI, that we're going to have a human safeguards throughout the process.
Jeff Young:
What does that look like? What does that look like in this?
Juan Espinoza:
Well, that means in our new system we still have a human reader, and then we have the AI tool, and if there's a discrepancy, and we reduce that discrepancy, which previously was four points to trigger three a third reader, now it's just two points, and that third reader remains a human. Those are the human safeguards there, but then ultimately the admissions decision is still 100% handled by humans on the admissions committee. So here's a subsection of the group that scores imported, and you have that human safeguard in multiple ways. That score is imported for the admissions committee to review, and then that's part of the one of many variables that they're reviewing as part of that, as part of that process.
Jeff Young:
So, the AI is one of the two. The AI is one of, in the old system, two human readers of these essay questions.
Juan Espinoza:
Exactly right. Yeah. Okay. And you know, I think that's so critical. I really think that's critical, because I think right now the perception and the reality for some companies and organizations that are, that are utilizing AI is motivation is efficiency, and it's almost a race of how can we cut costs, and I think that's a race to the bottom, and it's losing sight in what the gain is, and that is as we gain some efficiencies. So, in our case, 8000 human hours are being saved. Question is, what are we going to do with those hours? How would you repurpose that to the students' advantage and betterment from an experience standpoint. I think that cannot be lost in all of this.
Jeff Young:
What do you do with the 8000 saved hours?
Juan Espinoza:
You connect with students, I mean, that's so all the volunteers that we had reading essays were trying to think, how can we make sure that they're still engaged in this process, but maybe they're more available to talk to students.
I tell people all the time during the application process, admissions counselors research show are always top three on who students want to talk to once they're offered admission. We drop to 17. Who do they want to talk to once they're in? They want to talk to students, they want to talk to faculty, and they want to talk to alums.
Jeff Young:
So, what does that look like? Those relationships, because I am interested in this. So, you're saying hopefully you could have that, or are you actually doing it? Those the same human time.
Juan Espinoza:
So this was our first cycle, so we're still in the process of trying to see, you know, how are we going to be able to do this, but yes, we are going to find ways to connect our faculty, because those were the biggest portion of our readers that were reading the essay questions. Try to find new opportunities for them to connect with students, try to integrate that into our visitor experience, try to have more time with them in a student, because we think that one, they showed a capacity to help out, they have the time, they identified that time. We're going to give them something much more exciting to do than just reading essay questions. We're going to connect them with students that are considering Virginia Tech, and I think that's, and we're going to be strategic on when that happens. It's going to be after the student is offered, but we feel that that's going to have a much larger impact in the long term, than sticking to our old system, and just having them read essays only.
Jeff Young:
I think it's worth noting that Virginia Tech still hasn't figured out exactly what that thing is, and I'm curious what they come up with that will work. Busy professors could score admissions essays on their own time. Sneaking in a few minutes here or there in their busy schedules, but interacting with students as Juana Spinosa described might prove hard to work out the logistics for.
Now, how do students feel about this?
What has their reaction been to Virginia Tech bringing in this layer of AI review?
We'll get to that right after the break.
Sponsor:
This episode was brought to you by Studiosity.
Real learning needs a struggle, but for educators, assessment security just means more detective work, and that's not what students came for. Hence, not what the teachers came for. True learning and authorship isn't found in a probability score or a surveillance log.
It's found in the journey.
Studiosity makes assessment security easy by shifting the burden away from every educator and into a proactive institution-wide infrastructure, from Viva-style inputs to clear binary validity reporting, Studiosity is the multilayered assessment security process turning institutional policy into practice.
Just a clear signal to the teacher owned by the student that I did this thinking. De-risk your assessment flow and reduce workload. Give your teachers back the time to teach, not police, and give your students the agency to show their own growth.
Keep the learning real with Studiosity.
Get the walkthrough at studiosity.com/learningcurve. That's studiosity.com/learningcurve.
Now back to the episode.
Okay, before we actually do get back into it, I wanted to remind you about a new newsletter that I started about AI at colleges for The Chronicle of Higher Education.
It's called Jagged Intelligence.
On the latest installment, I looked at how colleges are reacting to the Pope's recent take on AI, rounded up the latest news on which colleges have signed big deals with AI companies, and linked to a new crowdsourced tracker of practical ways that colleges are adopting AI on their campuses.
If you haven't already, I hope you will head over to chronicle.com and click on the word newsletters at the top to sign up.
Okay, let's get back to the thorny issue of AI in college admissions.
Jeff Young:
You know, I have to ask, a lot of times students feel like they're being told all over the place not to use AI, like, because, you know, it's cheating, it's not being human, and these are real concerns. I'm not trying to minimize this, and at the same time, here you are as this esteemed institution using AI to replace part of a human job that professors were doing. So, how do you square that? What you know, what do you say to a student who might be throwing their hands up and be like, "Wow, like I see what you're saying, but you're still like you're still using a robot to do what a human did, and, and you're telling, and you're, and you're advising me not to do that.
Juan Espinoza:
Well, what we do, and I think this is so important as part of the trust piece, we're communicating what they get out of this, and what they get out of this is they get a decision much sooner than ever before. So we were able to shave five weeks out of our review process. So instead of getting decisions out in March, we got them out, I think it was around the third week in January. So that was a direct benefit for them. Number two, we tell them to use it as a brainstorming tool.
Jeff Young:
So you're not saying don't use AI.
Juan Espinoza:
We're not saying that, we're saying our strong recommendation, because how we just explained what we're looking for, the answers are tied to your experience and your characteristics, so I would argue to be at a disadvantage for just throwing this into Chat GPT, and expecting you're gonna get the highest score out of that. That's not going to happen.
But if you want to use it for brainstorming, if you want to use it as a reflection piece, go ahead and do it. We encourage it, because the reality is AI is not going anywhere, and what we're seeing from our employers is they're expecting a certain level of fluency, so we don't want to discourage from it now, are we? I hear lots of different perspectives on what we're doing at Virginia Tech from outsiders. Are we using this to detect AI? No, we're not. We're not interested.
Jeff Young:
Okay, so you don't have some AI detector machine in there.
Juan Espinoza:
I think that's a waste of time because of how quickly this is advancing, whatever you're using to try to detect AI is going to be out of date, you know, in probably a few days, and so it's …
Jeff Young:
So you don't take into account whether it feels human or bot, like the essays,
Juan Espinoza:
What we're taking into account, what are human readers and what the AI tool, which was trained. Trained on several years, seven admission cycles of previous readers, and what patterns led to a higher score. As I explained earlier, what we're hoping to get to is the student's story, and I would argue, as good as AI is right now, it hasn't been able to replicate us, all right. It's really good at lots of things, but being human isn't one of those things yet. All right, I might stand corrected many years from now, but I hope we can stay and so and so. That's where our advice is. It's in your best interest to answer this now. These are short answer essays that you know, give me
Jeff Young:
An example, give me a flavor, I can picture it, I think, but
Juan Espinoza:
So one of the questions is we ask, ‘What are your goals?’ All right, and so you put that in the chat GPT, it, you know, it doesn't know who you are, it's going to spit out the bunch of goals, but is it really what you want to do, and are those goals going to be focused on the long term, because when we ask that question, our hope is the students are going to answer in the long term, that's what ends up getting a higher score versus they're asking, putting out a goal very much in the short term.
All right, okay, okay, so in that process, if you just put that question to Chat GBT, saying, 'Hey, I'm a college application, it's asking me what goals it's not going to be focusing on the long term, necessarily. It's not going to know their story, it's not going to know their personal characteristics. And so I think that's a great example of where, if you use that, you're going to be at a big disadvantage in this process. And we've, you know, we've seen experience of that. We're seeing that students who are actually tying to their experiences on the application and everything else, and their goal is aligned with what they've done and where they're headed, they're going to get a higher score, and so that's an important piece of all this, and that's what keeps us human in a lot of ways too.
Jeff Young:
So far, applicants to Virginia Tech, they seem okay with this use of AI, at least there hasn't been any big pushback, but other colleges that have used AI in admissions have faced some resistance. For instance, the University of North Carolina at Chapel Hill faced some heat last year over the use of AI to score grammar and writing style admissions essays. In that case, rather than talk openly about its AI use, the university seemed to be doing it quietly. In fact, it only came to light after student journalists at the university's student newspaper, The Daily Tar Heel, wrote about the AI use in admissions, and student editors at the paper wrote an editorial last year criticizing the practice of using AI to score grammar and style in essays. Writing quote for admissions officers to even use a tool that frequently misreads and relays inaccuracies to the reader threatens every applicant's chance of admission. Until AI is further developed and reliable, it should hold no significant place in determining students' futures.
These days, there's a growing call for admissions offices across the country to be more transparent about their AI use, and to establish some new norms for the sector. For instance, the National Student Legal Defense Network, that's a nonprofit student advocacy group, it recently issued a list of do's and don'ts of AI in college application evaluation. Transparency was a big part of that, telling colleges they really need to tell students what the rules are. Right now, some colleges do make their stance clear, and others don't.
So, at the moment, all this has applicants wondering what to do here. I asked Juan Espinosa what advice he would give applicants when it comes to using AI in their applications.
Juan Espinoza:
My advice is what I'm giving now: use it as a brainstorming tool, you know. I encourage you to explore this technology, but I will say that the fact that we've engineered these questions only have about 120 words as a response, so that's a short paragraph, you know. Students aren't doing a writing sample, we're not asking them for a 5,000 word essay, and you know, getting them to circle back,.
Jeff Young:
There's not enough words to judge their writing, that's not the goal.
Juan Espinoza:
Yeah, this is about content. This is about the meat of their response. It's not going to be a writing sample. In fact, we do not penalize for grammatical errors. That's not part of this review. If you're gonna..
Jeff Young:
That might surprise some of these applicants.
Juan Espinoza:
Yeah, I know. But for this portion, the essay questions.. we're not really for the whole application, I guess, because this is the main place where you would have any grammatical errors. You're not penalized, you can have the wrong grammar, misspellings. We're focused on the content, and if we can still understand the content and the message that you're trying to answer and deliver to us. That's what we're evaluating, that is a key portion of this, is making sure no one's this at a disadvantage for misspelling or anything else.
Jeff Young:
I have to also ask, I'm sure you've seen this, this argument. There was an MIT economics professor who kind of knows, knows this idea of the AI approach, and he worries that that there's there's could be if a lot of institutions start to go this route that there could be a sameness because these models are trained a certain way and it might make similar similar decisions and and maybe narrow applicant pools accidentally like there might be an unintended consequence of relying on these AI models over time, because they, they could be, you know, like even if you have a couple models that they're not humans.
What do you say to that concern about, you know, kind of homogenizing it through algorithms?
Juan Espinoza:
Yeah, I think that is a very valid concern, signal degradation on a lot of variables we're using right now, I mean, it's going to get tougher, we're already seeing that in certain degree with with grades, when you see so much grade inflation, you're not seeing a bell curve and grade distribution.
Jeff Young:
Right, so in other words, if there's just, if there's not as much distribution on grades, they don't mean as much, they're not sending a signal, and so the same thing with these essays.
Juan Espinoza:
You're exactly right. And I think, as admissions offices, we're going to have to pay attention to that, because we are in a path where our signals that we're utilizing our review process, they're being degraded in how we were going to be able to evaluate that, and that's going to be a challenge, not just for the midterm, I would argue, but in the short term, and it's something we need to pay attention to now.
Jeff Young:
For better or worse, there is a whole cottage industry of admissions consultants and coaches who promise to help applicants improve their chances of getting into a selective college. And I was curious, what these experts are seeing as far as AI, and what they're recommending.
I got on a short Zoom call with Sophie Sajnani, an admissions consultant who's helped hundreds of students apply to colleges, working with students and families from around the country on Zoom sessions. I started by asking Sophie what her initial advice was a year or so ago when students started asking her about whether to use AI.
Sophie Sajnani:
So my initial advice when students asked that was, I advised no, you should not, because I wanted to preserve the creative aspect of the writing process.
Jeff Young:
You recently wrote an op-ed that I saw in the San Francisco Chronicle about how you've had a change of heart recently about AI. So, so what? What happened to bring you to that, and what is it?
Sophie Sajnani:
The real moment where I had that change was actually a phone call. So, between early admissions and regular decision, you know, again, students were starting to use a lot, AI a lot more, and I wanted to make sure that they were protected, because, again, sometimes it's hard to know what's happening on the admissions end, how they're perceiving these essays.
So, I actually spoke to someone high up at a private university in Boston. It was not an Ivy League university, but it's a very highly ranked one. It's very expensive, and on the call, they told me if a student uses AI, so they said something along the lines of, it is what it is, we can't judge them for it, we won't change our decision based on it, and the only way that we would find out is someone externally potentially complaining on the student or writing a tip, but otherwise it is what it is, it's fair game, and that completely changed my perspective.
Jeff Young:
Sophie interpreted that to mean that AI is just part of the landscape now, and that, at least for some colleges, they don't feel like there's a fair and accurate way to detect it.
Sophie Sajnani:
I realize that this needs to be given to students, because it's an advantage in such a competitive admissions process. Every advantage you can have a counselor using AI is beneficial.
Jeff Young:
I asked her what she thinks using AI in a positive way would look like in college admissions.
Sophie Sajnani:
For this, I would go back to again, you know, when I had had that phone call, and when I was thinking really hard about this, I had looked at a quote, actually from a New Yorker writer, and she had talked about writing an essay about losing her mother and the grief she had experienced with it, and she had said for a long time she wasn't able to put this into words, and it wasn't until she had Chat GBT where she was finally able to give voice to all those emotions she had. It helped her write an essay about it.
That being said, she said that all the ideas and all of the big statements, the really good parts of the essay were her own, and so I view AI like that, where those emotions, those original thoughts should come from the student. Just use a AI to put that together, and then also have essay drafts that you can pick from. That's something I spoke to in the. SF op-ed is that once students use AI, they should be thinking critically. They shouldn't just accept the output, they should ask, "Do you have another draft for this? or use their own judgment and use my judgment to help push back on some of the writing. I think that's a very important part of the process too.
Jeff Young:
Now she is giving students different advice than the Virginia Tech official I talked to when it comes to AI.
Sophie Sajnani:
And I still believe that the brainstorming and the outlining process of the essay are the most important. I think that is where the creativity comes, and I don't think that that should be necessarily replaced by AI, which actually goes against what a lot of colleges state a lot of colleges state you should use AI for the brainstorming or the outlining, just not to write that final draft. I'd say the opposite. I'd say that don't use AI for that, that's where the real work happens, the discovery. Use AI to just help you put those final words together. That's my, my take on it.
Jeff Young:
I guess. The, the one thing that I do, you know, I hear folks in the admissions world from the college side worry about is that AI could sort of flatten the thing they're looking for in these, in this whole exercise, which is a glimmer of who this person is. What do you think of that, that concern, as someone who's like working with students? I mean, is that something that now this tool is out there, and maybe everyone's using it, so.. but you know, is could that have this pretty big negative impact on the whole idea of college admissions? Really,
Sophie Sajnani:
I would say to colleges that are thinking about that, about how AI could potentially flatten students' voices or not give them a full picture of the student's identity. I would say then it's on you guys to change the process. This is particularly coming from the essay, that's where students can show these things, so if that's a concern for them, then they need to change on their end. They can have students submit writing samples from their school assignments. They can bring back the essay section of the SAT, which was stopped a few years ago. There's so many things that they could do to change this process. If they're not going to put any of that, any of those changes, and then I think that the students have every right to use these, these tools, and I think that then they should expect that maybe the essay will look a little bit different, and maybe they have to change some of their internal rubric grading skills, or how they evaluate the essays, but I definitely think it's on admissions to figure out how to navigate the essays, and if this is something that they need to change. Potentially,
Jeff Young:
it's worth noting that so far colleges have taken varied stances on how they see AI in admissions. So, I would say applicants should check the website of any college they're applying for and try to find out what the policy is, as this continues to shake out. When I talked with Juan Espinosa at Virginia Tech, who has seen so much rapid change around admissions and AI, I asked him what he thinks has been lost compared to his early days in the field.
Juan Espinoza:
You know, relationships were so big, and there's so much value in that, and I separate your relationships with counselors and relationships from students, right? So, relationships with students, that's a win-win. The more opportunities you have to connect with students, that's the betterment of the institution, and for the student as they're trying to figure out where they're supposed to be.
Jeff Young:
Sure, you mean like the more behind the scenes relationship here?
Juan Espinoza:
Yeah, yeah, absolutely.
But when you're looking at counseling relationships, absolutely critical in making sure back to my point that we're aligning students' expectations with what they're going to get as a decision, and the best way to do that is having transparency with those key stakeholders in that process, but the lines can get blurry sometimes, and you know they're, you know, I would know for a fact sometimes there were some calls from counselors saying, hey, you really need to look at this student because they're amazing, and is it? Are they amazing because they're doing great things? Are they amazing because you have a great relationship with them, and they're your favorite, right? And then you start inserting bias and all of that, and I would argue that's not good for the system as a whole, as a whole.
And so I think as we start entering, and I would argue we're kind of exiting this period where data is king, right?
Or that's the main piece of our decision making as we gain these efficiencies in in hours. I think relationships are going to be so key, and there's an economist. There's a professor of economics from the University of Chicago, Alex E. Moss, who just released an essay that I thought was really interesting, that talked, yes, you got people who think we're all doomed, right, we're all going to be automated and everything else, and you got other people that are declining as a group that said this is all hype, this is all going to fall apart, and then you got a bunch of people in the middle, he was essentially arguing that yes, I think it's pretty clear there's a lot of automation that's going to take place, right, and there are some people on that, you know, on the fringes that would say, you know, the field of economics is going to disappear because scarcity no longer will exist, we're all going to get whatever we want, right?
Jeff Young:
Right, we'll all get a minimum basic income and then it'll be great, yeah.
Juan Espinoza:
And he does a really good job in explaining those.
Yes, scarcities are going to shift, but they're going to still exist this in a different way and in different areas. And one of the examples that they talked about was as we enter a more automated period, and let's try to use our field as an example, so as more things are automated, there's going to be a sameness to your point, not just from the student side, but from the institutional side. When we're going to start, our messaging is going to start sounding very, very similar, and so I would argue that that opens the door, and he explained this brilliantly, where the relational sector is going to be well positioned for this, so you think of occupations that have direct reliance on relationships, so that's health care, education in general falls into that.
So, do the lens of admissions offices applying this, that means as we get more hours, like I use us as an example, we're gonna get 8000 more hours than we had in previous years. How are we going to repurpose that to be well positioned for that relation relational sector. How are we going to make sure that as students are getting more sameness from universities that we're differentiating by being there and having a relationship with them, giving them opportunities to connect with representatives of our university?
So relationships are going to start becoming much more important, and I want to be clear, and I keep repeating myself, that doesn't mean relationships were important as data became more prevalent in this process, sure, but it means that it reemerges as a lot more important as that sameness starts coming from the institutional side, and as students receiving that, that they're going to start paying more attention to who's who. Am I having conversations with who am I going back to that earlier model, and that still exists today, obviously, of us going to high schools and going to college fairs. That interaction becomes so much important.
Campus tours have always been important, right. As for students trying to figure out, you know, if it's going to be a good fit, I think more investment in those areas. So, as admissions offices are able to find some efficiencies in one area, they need to repurpose that, and I think prepare for this transition by making sure that they're investing more on the people side of things, not just on the recruiters, but institutions should be focusing on investing in faculty, making sure that that experience, that that relationship, which is going to be the key in differentiating themselves from others, is going to be what students are going to be paying more attention to.
Jeff Young:
It feels to me like we're gonna see more change in the next couple of years in terms of AI and admissions. Virginia Tech said it's going to be publishing more details on its approach, so that others could replicate it if they want, and plenty of other colleges are doing AI experiments in their own.
As far as admissions, of course, there are all these pressures on applicants and colleges to bring in some form of automation or use these tools, and I guess the big question for me is, Can this moment of AI's arrival give a chance to rethink how the whole admissions process works and make it more human?
This has been Learning Curve.
Each episode we tackle a big question around AI and education.
If you don't already, I hope you will follow Learning Curve wherever you listen, and please take a minute to leave a quick review. I saw a new one the other day. These really help. Thank you so much. This episode was reported and put together by me, Jeff Young.
Our theme music is by Kamaku, with additional music by Blue Dot Sessions. Our episode art was generated with the AI tool Midjourney.
We'll be back in two weeks with a fresh episode.
Talk to you then.