Colleges are using AI in admissions. How can they do it right?

Some colleges receive tens of thousands of applications every year. Admissions officers are turning to Artificial Intelligence to help with the screening. What that could mean for who gets into college and how.

Rick Clark, executive director of Strategic Student Access at Georgia Tech.

Taylor Swaak, senior reporter at the Chronicle of Higher Education, covering technology and innovation.

Emily Pacheco, assistant director of admission at Loyola University Chicago.

Phil Komarny, chief innovation officer at Maryville University.

Huy Nguyen, chief education and career development advisor at Intelligent, an online resource for higher education planning.

Part I

MEGHNA CHAKRABARTI: Regular listeners to On Point, and I truly hope you are one of them. Know that we are keenly interested in understanding how artificial intelligence is transforming everything about our lives in ways visible and invisible. You might have recently heard our rebroadcast about artificial intelligence and universal basic income, for example.

Or remember our weeklong series about how AI is changing health care. Today we've got another one that should be of interest to more than a million Americans and their families every year, at least. And that is AI and college admissions. And we'll start with Rick Clark. He's the Executive Director of Strategic Student Access at Georgia Tech, and he joins us from Atlanta.

Rick, welcome to On Point.

RICK CLARK: Thanks so much for having me. Great to be with you.

CHAKRABARTI: So before we talk about AI as a potential tool for college admissions, I'd love to actually get your help in understanding more deeply how college admissions works right now. So first of all, I understand that Georgia Tech received, what, more than 50,000 applications recently?

CLARK: This last year we were right at 60,000 total for the first-year class.

CHAKRABARTI: 60,000! And how many students were admitted?

CLARK: So we admit about 15%. We're a public institution, so that number looks different for in state versus out of state. Our admit rate in state more like one out of every three. But then that out of state admit rate is continuing to decline closer to about 10% now.

CHAKRABARTI: I see. Okay. Good point about in state versus out of state. An important factor for Georgia Tech. But nevertheless, the point is that there's an overwhelming number of students, much more, or in orders of magnitude more than the spots that Georgia Tech has.

Okay and how many admissions officers or temporary hires do you have to go through those 60,000 applications?

CLARK: Yeah, so our admission full time staff is about 35 right now. And then we do bring in seasonals. In recent years, as we've continued to see a pretty sizable growth in applications. Now about 60 that are coming in seasonally to assist in reading. And that is to allow us to get into a more holistic review. Because if you rewind, really just to 2020, the total applications were about 40,000 and then in 2016, it was 30,000.

So we've had 100% increase in the last eight years and then a 50% increase in the last four years. And it's just necessary to bring in additional folks to assist with that.

CHAKRABARTI: But we're still talking, let's say, approximately a hundred folks for 60,000 applications. So each person is managing, on average, many hundred, many hundreds, I should say.

So how does it work? This is going to sound like a totally gauche question, but I've always been fascinated by this. Does a particular reader just get a stack of applications and there's like various deadlines that they have to meet in order to say the person's grades are X and their extracurriculars are Y, what happens?

CLARK: Yeah. And really that is essentially what happens and if you go back, I think about my early days in admission at Wake Forest University, and those were physical files that you literally would take home or into your office. And exactly like you just said, this is the caseload you're responsible for.

There's the deadline by which you need to have fully read that application and made a recommendation on whether or not the student should be admitted, deferred or denied. And then, of course, there's layers that come after that, in terms of looking through with multiple readers or having a second person look back over, usually somebody a little more senior to do that.

Over time, it's evolved a bit. Obviously all of that's electronic now, so nobody's, thankfully, taking all this stuff home with them, but then also, you'll do simultaneous reading at a lot of schools now, where literally two people are reading the same application at the same time, maybe they're reading the full application, or maybe, let's say you and I are partners, you would be reading the academic side, a student's GPA and course choices and SATs, let's say. And I'm looking at their essays and their activities and their teacher recommendations, and then you might be making then an academic decision, is the student qualified or competitive, and I'm doing the same, outside the classroom, is this a fit for our institution? And so lining those decisions up either allows a student to move one direction or another for final decision making.

CHAKRABARTI: Okay. This is so fascinating to me, so forgive me, if I want to get in more into the nitty gritty. So on the academic side, for example, when we say reading an application, you're reading grade letters, essentially, or test scores.

What does that actually mean? Is there a formula? Because my understanding is that schools have their own sort of proprietary, let's call it algorithm, for example.

CLARK: That's right.

CHAKRABARTI: That they convert grades into some sort of meaningful number when it comes to an admissions profile.

CLARK: Yes, exactly.

And you hit the nail on the head, not only with what's currently happening, but in our conversation around where AI, I think can come in. Because every school, when they talk about, let's say, a student's GPA, it's not always what the student sees on their transcript. For example, there are some schools that will calculate out nonacademic courses, so a student might have taken physical education, or an art class, that particular institution does not consider part of their academic evaluation, and so they calculate that out, or increasingly you're seeing some schools say, in the ninth grade, a student is 14.

What value does that really have on predicting their success at our institution? And so you're seeing more schools calculate out the ninth grade. Other schools will say, all right, an AP class, we're going to give that half an extra point, an IB class or a dual enrollment class. And so some schools are putting weight on the grades and other schools are not.

And so you're right, it's proprietary, it varies from school to school, and that's a big part of the academic review. The other piece is many schools, especially more selective schools, will typically have a rubric where they're evaluating a student's rigor of curriculum, so it's not just a number, let's say a 3.8, arbitrarily. But it's also that student had choice at the particular school they went to now, did they really push and stretch themselves or maybe not so much. And so they're sometimes assigning an additional number, a quantitative number to the rigor of a student's curriculum. Again, that's currently happening, but that also is something you can train AI on going forward.

CHAKRABARTI: I'm going to get to the AI part in just a second, because of course this is the topic at hand. When I hear about okay, there's judgments that even have to be made when it comes to evaluating transcripts then, right? Because everyone wants to know, is it better to get a B in an AP class or an A in a non-AP class in that same grade and subject?

CLARK: That's right. And of course, the obnoxious admission answer is make an A in the AP class, that's tongue in cheek, and it's only certainly true for certain schools and a pretty small minority of schools, really, around the country. But yes, that's right. Schools create their own rubrics to evaluate what they want to put priority on.

Hopefully, that's based on what they see in their current student population.

CHAKRABARTI: We have a lot of, by the way, I just want to jump in here. Forgive me. We have a lot of listeners in Georgia and they're dying to know your actual answer to that question for Georgia Tech.

CLARK: Oh, they want to know my answer.

CHAKRABARTI: ... Oh, yeah.

CLARK: You should take a AP. Yeah. The truth is that as things have gotten more and more competitive for us, we are expecting students to take a very rigorous high school curriculum and do well in those courses. Yes, we admit students who make B's and AP courses. But increasingly, if a student is taking just the sort of main level classes, even if they're making A's, that's typically not going to be competitive in the applicant pool.

Students are stretching and pushing themselves. And so we encourage students to take that more rigorous curriculum.

CHAKRABARTI: I suppose Georgia Tech being what it is, that shouldn't come as any surprise to anyone hearing this. It's a great university. But there you go, Georgia listeners, you heard it straight from Rick Clark.

Okay, so again, sticking with the academic side here, it makes total sense to me why a powerful tool like AI could come in and do essentially those calculations in the blink of an eye and how that would free up person power and time for an admissions office. Just gimme a sense of a com comparison, like how much time would it take for a human or to do that to complete the academic side of the rubric for an applicant?

CLARK: Yeah. Several minutes for sure, because you're, first of all, you're getting to the transcript to begin with. You're having to determine in that school what is the grading scale, and of course, that varies from one school to the next. You're having to look at the profile of the high school, which every school has, that shows what was available to the student, what could they have possibly taken.

And then making again, those determinations within your particular institutions' calculation. So again, a human is going in and literally oftentimes calculating, typing in numbers and into a calculator effectively, online calculator there also, and Georgia Tech's a great example of this. We put a lot of priority on math, and so we have our humans at this point go into a transcript and say, what was the highest-level math the student took?

Let's say that is BC calculus. They go to a drop-down menu, they click on it, they scroll down, they select BC calculus. Because for us, we do put priority on that, and it's predictive of success, schools around the country are doing the exact same thing. So a human right now is taking several minutes, at least to calculate GPAs, evaluate rigor of curriculum and then make a recommendation.

And in many institutions, that's kind of it. Anyway, they are using formulaic admission decisions, and so AI has huge potential, but even in a holistic review right now, a lot of our seasonal folks are doing that type of work, and I do feel like this is where we're going to be able to free them up for high level tasks going forward.

CHAKRABARTI: And even though people might hear, Oh it's just a few minutes for each application. We're talking over 60,000 applications, right? So this is like hundreds of hours. We've got to take a quick break in just a second. So is Georgia Tech right now using AI for that kind of thing?

CLARK: So we are currently waiting for our CRM slate to roll out some of these opportunities, and that is coming in the year ahead.

But that is a big part of, at least one of the primary areas where we're about to start implementing AI.

Part II

CHAKRABARTI: We are talking about the growing use of AI in college admissions. And why? It's a sheer numbers issue. For example, in the 2023 cycle of admissions, 1.24 million Americans, distinct first year applicants, using the Common App, applied to colleges.

So that's actually not even all of the college applications or applicants in 2023, but at least those 1.24 million people submitted more than seven million applications to colleges and universities around this country. Ranging from Georgia Tech, as we're hearing from Rick Clark, who had 60,000 applications to sort through this year. Over at UCLA in California, the most applied to university in the country, its admissions office received more than 173,000 applications for the fall of 2024, including 146,000 first year applications.

So with that volume of applications, which includes lots of subjective analysis, which we'll talk about in a second, but a lot of just straight up quantitative analysis, you can see how AI could be a potentially powerful tool for admissions offices. So here's Phil Komarny. He's the chief innovation officer at Maryville University in St. Louis. Maryville no longer actually looks at essays from their applicants, but Komarny says that AI has still been saving the admissions staff a ton of time.

KOMARNY: I believe in giving machines inhumane work and giving humans really humane work to do. So parsing through a transcript, looking, adding up all the science grades to get to some kind of evaluation, to get a grade to put into our information system takes at least 10 to 12 minutes and 15 minutes for transcript sometimes, for an admissions counselor.

We've got that down to seconds, because all the manual processing, that ingestion, classifications been done by a model. It even exports it into a spreadsheet that they're used to seeing.

CHAKRABARTI: So that's Phil Komarny at Maryville University in St. Louis. Rick Clark is with us. He's Executive Director of Strategic Student Access at Georgia Tech.

And Rick, I have to admit, I was surprised to hear that these processes weren't automated already at colleges and universities. But nevertheless, no matter if it's a human doing it. Or AI doing it in a blink of an eye, even on the academic side, there are judgments that have to be made, right?

Like you said, is this the most rigorous course selection that a student could have had? Or how does a particular course at said high school compare to another one at a different high school? There are judgments that have to be made in every rubric, as you put it. How are those, do you know, or do you have any concerns about how the AI is asked to make, the AI tools are asked to make those same judgments?

CLARK: No it's quite similar to the way that admission officers are currently trained. And that is by taking a rubric and then looking at the school profile in order to determine what was available at that particular high school, and then what was the student's choice on their way through their academic journey. Yeah, as you mentioned, and as I believe Phil also alluded to here, the nice thing about AI is that's what it does well. The ability to upload information and have scanning and pulling in order to manifest a result is what AI and some of these models do extremely well. Freeing up the human that he just referenced, that 10 to 12 minutes.

To instead of doing all of that work, checking all of that work, and this is where, of course, and I'm sure in your prior conversations around AI, there's a lot of paranoia of, is this going to eliminate jobs, and what is this going to look like going forward? I think when we talk about holistic admission, this makes it even more human, and it makes it even more whole person, or holistic in that, instead of a human, I would say, wasting their time calculating GPAs or selecting the highest math, Instead, what they're doing is saying, this is what the model shows at this high school and this student's transcripts, here's how they would evaluate that rigor of curriculum.

Do I agree? And it's just so much of the way that we're already incorporating AI into our regular lives. I'm taking ways to navigate my way through Boston. Do I agree this is the best way? Sometimes yes and sometimes no, but a human's making those real time confirmation decisions or discretionary decisions that may veer from what the AI suggests.

CHAKRABARTI: That's a really good analogy. It makes it quite clear that it doesn't sound like at any point in this process, at least at Georgia Tech, that the human judgment part is going to be removed. But is it, does it work such that Georgia Tech is given its rubric and the judgments that it makes about, as I said, like comparing courses, to the company that's developing the tool for your platform, and that's how the AI then is programmed for Georgia Tech?

To put it in a very sort of awkward way.

CLARK: Yeah, that's right. Schools are going to have to decide if they want to utilize some of the AI options that are already being rolled out within their particular CRM. So for a lot of the better known, highly ranked and more nationally branded schools, many are on a CRM called Slate.

I would say Salesforce is another primary one that you see in the marketplace. And in both of those, there will be the option, because it's already there, for the school profile matched up with a student's transcript, and all you'll have to do is upload the rubric that you're currently using externally, in order to sync all of those up, in order to bring about a result, right?

And some schools are going to decide, no they're not comfortable with that. So at Georgia Tech, we have something called the Center for 21st Century Universities, and Georgia Tech also has the largest online master's in computer science. They're already doing this outside of a CRM, where they're effectively looking at historical data and in student performance. But then matching that up with the student's transcript and rubric and profile, in order to make these kinds of determinations. So that's going to be an institution-by-institution determination. At Georgia Tech right now, we're trying to determine on the undergrad side, is it going to be best to utilize what Slate is rolling out or to go with Center for 21st Century University's kind of proprietary solution.

CHAKRABARTI: And CRM, meaning Customer Relationship Management System, for those people who aren't fluent in the Salesforce type language out there. So Rick, hang on here for just a second. Because we wanted to get a sense as to how broadly or widespread the use of AI has become in college admissions.

So we turned to Intelligent.com. It's an online resource for higher education planning. And last September, so about a year ago, Intelligent surveyed almost 400 high school and college education professionals who have deep knowledge of their school's admissions processes.

And at the time, about half of the respondents already said that their schools were using AI in the admissions process, and that percentage is expected to grow dramatically by the end of this year.

HUY NGUYEN: It's not surprising to me that actually eight and ten institutions are looking at this because the admissions process for students is quite intense, right?

CHAKRABARTI: That's Huy Nguyen, Chief Education and Career Development Advisor at Intelligent.

NGUYEN: Some of the larger schools are taking in, tens of thousands of applicants each year throughout the application process, and so it does create a tremendous amount of data and workload. And so I think that even for colleges, they're looking at, how do we increase productivity?

How do we look at more students that are applying? And how do we go through this data in order to find the right students that are going to fit well within our organization?

CHAKRABARTI: As we've talked about with Rick, admissions offices see AI as a powerful tool to help sort through all those applications, and the data they produce to find answers to the kinds of questions that we talked about.

In addition, some schools are using AI for more subjective reviews. Like summarizing personal essays and recommendation letters. As Huy reports, platforms like Student Select can scan essays and interview notes, and churn out summaries on things like personality insights, skills, and other non-cognitive traits.

Keep that in mind, because I will return to it later. It is so interesting. Now, clearly, any use of AI brings important ethical questions along with it. And in college admissions, Huy says higher ed officials have ethical concerns, but they also see AI as a way to make the admissions process itself more ethical.

On that point, they see AI as being able to put checks on human subjectivity and bias. The technology could actually make the admissions process more fair.

NGUYEN: If you think about the way that a lot of schools are doing this, it's usually just one person or maybe one person that moves that person onto a panel and then they review the applicants as well.

So let's not make any mistake that there is human bias and human error, as well. That's just part of being a human. And I think that by using AI, a lot of these organizations are actually thinking that it's a little bit more equitable. Because they're able to review more essays and more subjective things like letters of recommendation, and giving more people a chance to actually be reviewed by more than just one person.

CHAKRABARTI: In fact, Intelligent survey found that 90% of respondents said they believed AI will somewhat or very likely help with reducing bias in the admissions process. On the other hand, every task an AI tool is asked to do carries ethical risk, such as how far or how many AI tools should be used to assess the whole student, or exactly what is the tool told to look for.

When summarizing desired characteristics from a student essay, recall in the recent Supreme Court case that ended affirmative action, documents revealed that Harvard University ranked applicants on five criteria, academics, extracurriculars, athletics, recommendations, and personality.

And on that one, court documents showed that Harvard consistently ranked Asian students lowest in personality traits, such as likability. Now we don't know the tools they used, and humans were making these evaluations, and we don't know how the ranking system produced such a result. We just know that it happened.

So with AI, understanding that process, as schools use it more and more, understanding how AI is doing that process, and its outcomes will become even more opaque. Nevertheless, a surprising number of education officials in the Intelligent.com survey said that they foresee AI tools eventually being used to make the ultimate decision, whether a student is admitted to a school or not.

NGUYEN: Yeah, I think it's going to depend on the phase that the applicant is in, right? So obviously, if they don't meet the initial criteria, and AI automatically flags them as not meeting the academic criteria, I would say that yes, a large percentage of institutions are looking at that, having AI make that initial and final decision to move somebody on. When we surveyed these institutional leaders as well, about 87% of them said that is going to either sometimes or always make that final decision.

CHAKRABARTI: Again, how are those decisions going to be made? Will a tool be told to cut students in a first round based on grades only? What about the promise we continually hear from institutions of higher education all over this country, that they consider the whole student, and their background and their circumstances when thinking about who to admit to their freshman class, which is why I think the Intelligent.com survey produced a very interesting conflict. It found that the majority of surveyed schools already using AI say they allow it to have the final say on applicants in certain points in the process. But at the same time, 66% of admissions professionals are already concerned about the ethical issues surrounding the tools that they're using.

Ultimately, Huy says that the use of AI in college admissions opens up a new chapter for how prospective students should think about their applications.

NGUYEN: Just like how you might work with a counselor in order to come up with strategic ways to stand out as an applicant. I think understanding how these models work and building an application, and speaking to that in some of your subjective materials that you're putting in, really is going to be advantageous for students and parents.

CHAKRABARTI: That's Huy Nguyen, Chief Education and Career Development Advisor at Intelligent.com, an online site for higher education and planning. Now, Rick, I know you probably have a lot to say about that and I will get to you, but I want to introduce Taylor Swaak.

She's Senior Reporter at the Chronicle of Higher Education who covers technology and innovation. Taylor, welcome to On Point.

TAYLOR SWAAK: Hi, Meghna. Hi, Rick. Pleasure to be with you guys.

CHAKRABARTI: The things that we heard Huy say about the rapid spread and growth of the use of AI in college admissions, along with the same rapid growth of the ethical questions around it, does that track with the reporting that you've been doing, Taylor?

SWAAK: I would say it does. And to an extent, I feel like it even emphasizes how quickly this is moving. Because I was focusing on AI admissions last admission cycle. And at that point, using AI as a an evaluator of sorts, that was something that for most admissions folks that I spoke with, wasn't even on the table yet.

We were seeing primarily this focus, this has come up before. On using AI tools to tackle that so called administrative drudgery. So maybe scraping transcripts and getting that information into a database or crafting messaging campaigns. And more commonly in identifying prospective students.

So you know, using AI tool to compare applicants' data against a historical data set and then placing those students into groups, indicating their likelihood of attending. So those two focuses were the ones that I was hearing most just last admissions cycle. So listening now and hearing about using AI from an evaluation perspective, and colleges starting to consider the potential there, we're already seeing huge gains in just considerably a matter of months.

CHAKRABARTI: I'm going to play devil's advocate here for against the people who might be concerned with the use of AI in college admissions. All the questions that are raised with AI about, is it going to produce a bias, how do you know what likability is or isn't, aren't those just the same questions we already have with human evaluators?

So are the ethical concerns potentially overblown? Taylor?

SWAAK: That's a great question. I feel like there's a trickiness to AI. So many of these tools are black boxes. We don't know entirely how they're arriving at the answers they're getting to. And a big thing that folks have told me as well, especially when we're talking about high stakes scenarios for use, is that AI can't be held accountable.

So on top of being a black box, there's really no accountability, where if something goes wrong or a student feels like they have been wronged, an AI tool reviewed their application and made a decision based on that. And I think that there are some legal implications, as well, and accountability is just an ongoing question that I think folks are grappling with as well.

Part III

CHAKRABARTI: So Rick, let me turn back to you. Just walk me through for a second some of your responses to the ethical questions that we raised in the last segment, such as, how far should admissions offices allow AI to go to make decisions about whether or not to admit certain students?

CLARK: I think it'd be interesting to know more about some of the surveys that went out and what the admit rates and selectivity of certain schools were, because, again, right now around the country, the average admit rate for four year schools is 65% to 67%.

Meaning, there's a lot of schools that are making formulaic decisions. They are using numbers, essentially, to make their decisions, and in those places, it makes a ton of sense for turning over, effectively, your decision making to AI and then just checking it, right? When I hear big numbers and percentages, part of me thinks yes, for a pretty decent slice of American higher education, that makes a lot of sense.

CHAKRABARTI: Rick, you still there?

CLARK: ... An opportunity recently to talk about AI with a Senate help committee, and I was speaking with my friend Jim Rawlins, who's the VP for enrollment at University of California, San Diego, where they get, over 100,000 applications. And his big thing was, listen, we're not going to experiment with AI on live 17 year olds, but what we are going to do is run A/B testing.

And for instance, right now, when admission officers and seasonal folks come in, they do training and norming. They look back at prior years and applications. And then that same group of readers will score to see if they're in line or out of line with what the rubric would say, and what the consistency would be across the staff.

Truth is, though, because of the volume, as you described, and what I would call is an unforgiving calendar of admission. Meaning these deadlines come quick and students procrastinate and they flood in at the same time. It's rare to get much of a chance to do consistent re-norming during the process.

That's where I think AI has a lot of potential, because it's able to --

CHAKRABARTI: Rick's line keeps dropping in and out, which is really unfortunate. We're going to work to get him back here. Taylor, let me turn back to you and let me ask you first about something that you reported on, and then we'll return to Rick's point about really, in a sense, for a lot of college admissions, this is a numbers game.

But I earlier had mentioned a platform called Student Select, and in your reporting, you found that when Student Select scans personal statements or letters of recommendation, it scores applicants on those non cognitive traits, like positive attitude. Did the developers of Student Select, I don't know if you were able to talk with them directly, did they tell you exactly how the tool evaluates positive attitude?

SWAAK: I'm really glad that you brought this back up, because the interesting thing about that particular use case, and I think it speaks to a broader point that I should have mentioned earlier, I had heard about that kind of capability through Student Select, but I wasn't able to get a list of college partners at any colleges that I did speak with, said that they didn't use it in that way.

And it's likely that folks do, but I do think that particular instance that I came across and mentioned in the piece is, I think, a reality check, as we're seeing the capabilities for AI develop, that capability doesn't mean implantation. That what AI is able to do is progressing so rapidly.

And there are theoretical ways that it can be used, but that doesn't necessarily mean that's how colleges are using it. And I know in my particular experience, and again, even this school year, we could see things changing. And even when I spoke with folks a few months ago, they said, Hey, talk to me in a year.

We might be more open to this as we get more comfortable with AI and the technology. But for right now, using it to review personal statements, and scanning essays and giving personality traits. Something that feels like a very human process. Folks seem a little bit more reticent to just jump into that.

And another point I want to make sure is really clear for listeners. When we use the phrasing of AI making decisions, I do think we run the risk of anthropomorphizing a little bit. I had a great conversation with an expert a few months back, when I was really dipping my feet into AI. And they told me, remember, AI isn't thinking it's mathing.

It's taking data that it's been trained on, lied over an algorithm, and it's making predictive calculations on what a right answer would be, based on what it's trained on. So it's not human. And that's why I think when we're talking about really high stakes areas, whether that's admissions, financial aid, counseling, folks aren't immediately rushing to using it in this way.

But as we're hearing, and I would be really curious to track this in the upcoming admission cycle, maybe that's something we'll start seeing. But in my personal experience, even looking at that specific vendor, I wasn't having any luck finding use cases. And I think that does say something.

CHAKRABARTI: Your dogedness notwithstanding, I wonder if people are just reluctant to say. From what I heard you say, Student Select wouldn't give you any colleges or names. But no one develops a tool in the complete absence of a potential market for these things.

Because as you probably know, Taylor, there's these kinds of AI tools to analyze, let's say, cover letters are already deeply in use in the world of work. And it's gotten to the point where you can just Google it and you can find websites that say, use these sets of words, maybe even in this order in your cover letter, to make it through that first AI evaluation.

It does make me wonder whether or not such a thing is inevitable in college admissions, for the simple fact of what we keep coming back to, that the number of human readers is so far outstripped by the number of applications they have to deal with.

SWAAK: Yeah, no, I'm sure it's alluring, I've read about just how burnt out a lot of folks in admissions are, and you were just talking with Rick at the start of the call, thousands upon thousands of applications.

And the AI, the capabilities do exist, so I can imagine that it's appealing as a concept, I really do think a lot of folks are balancing that appeal and that efficiency with, some folks that I spoke with in admissions, they really do take pride in their work. And they do see the admissions process as a process that really does need to maintain that personal element, both from a moral standpoint, from a public perception standpoint.

They're like, we students, we know that many of them put in hard work on these applications and we want them to know that we're looking at them and not in an AI tool. And so I definitely think moving forward, there's going to be these checkpoints and these conflicts about efficiency over retaining that personalized human element. For sure.

CHAKRABARTI: Yeah, because it made me wonder. I was just theorizing in my own head. Ultimately, this all comes back to humans, right? Because as Rick was confirming with us, the colleges and universities in using any form of an AI tool in the admissions process has to communicate to the developers, Here are our rubrics.

Here are our values. Here's how we ask the human beings to do it. Now program the AI to do that even faster. So judgment is, human judgment still at the basis of all of this, but AI is getting extremely like terrifyingly good now. But I do wonder if we're at the place where say, I don't know, writing from Ernest Hemingway was submitted as a theoretical student essay, would an AI just judge it in a certain way being like, this isn't very, quote unquote, sophisticated writing.

Probably not a good fit for the university, but a human might read it and be like, actually, this is truly brilliant, in part because of its seeming lack of of sentence complexity. Who knows? But as your reporting shows and you've said, I don't know if we have time to fully find the answers to that question because of the rapid spread of the use of AI in college admissions.

SWAAK: And I know colleges are trying to wrap their arms around use both in admissions and outside of admissions by establishing guidelines, at the very least, basic truths of how it can and can't be used. And then encouraging departments to create their own. And for the most part, sure, there may be a couple of bad apples that just see using an AI tool in quote, decision maker is super efficient, save me a lot of time, the right to go. But I think for the most part, people don't, like I said before, do care about the work and the perception of that work. And they don't want to risk perpetuating bias and discrimination.

And so I do feel like a lot of folks really are approaching this with care and thoughtfulness and trying to set up some guardrails and shared understandings of how to use these tools. So I think that more so than just jumping in two feet first, especially like I was saying, when we're talking about functions that are so high stakes, like admissions.

CHAKRABARTI: I think we have Rick Clark back on the line. You there, Rick?

CLARK: I am here. Yes.

CHAKRABARTI: Sorry about those technical difficulties that we were having. We need some folks from Georgia Tech to fix our system here.

CLARK: (LAUGHS)

CHAKRABARTI: But Rick, so I just wanted to get your thoughts on again, all the ethical questions that we've been trying to navigate through on the use of AI in college admissions and how far should it go?

Go ahead.

CLARK: Yeah. And I'm not sure how much of the prior response you heard, one of the things that I wonder about that survey that was discussed is the admit rate for some of those institutions. Because to me, it makes a lot of sense that a school that's already utilizing formulaic decisions would feel very comfortable, in many ways, turning over that preliminary decision to AI. That's going to run numbers and render a decision, simply to be checked by a human, but on the holistic side, for the more selective institutions around the country.

I talked with a good friend. I testified at a Senate help committee around AI recently and talked with my friend, Jim Rawlins, who's the VP at University of California, San Diego. They receive over 100,000 applications.

CHAKRABARTI: We did hear this part where he said he's not going to run experiments on live 17 year olds, but they'll do A/B testing.

CLARK: Exactly. Yeah, that's correct. I do think that's where a lot of potential is. Again, seeing this norming side of things and checking bias in real time, and there's so much potential there, I think, for the implementation of AI. But ultimately where I think people should feel comfortable, where I think that people can rest a little bit assured when they're, as you said earlier, paying money to apply, and putting in all these hours and having worked for four years in high school. Is that we, admission directors and deans and VPs around the country, we're ultimately judged for the results.

And when you're talking about presenting a class to your board of trustees, to your alumni, to your president, you're going to make darn sure that it is accomplishing the institutional priorities of your particular goals every year. And one of the things that's nice about AI, I think, also, and where these models are going is you're able to incorporate more of how students are doing on your campus, particularly from individual high schools, and bring that back into the review process.

And to this point, that's only been done really, very nominally, and I think this again is really what it's all about. Is every school has their own institutional priorities and goals, and that's what these directors and deans are responsible for, and they want those students to come to campus and to succeed.

And so the more historic information that you can incorporate, in order to A) check bias and B) achieve those goals. That's really, ultimately what college admission is all about.

CHAKRABARTI: So Taylor, this was a really interesting aspect of the potential use of AI in college admissions that I hadn't thought of, but you've reported on it, that here's another, here's a tool where, as Rick was saying, admissions officers can get a better sense of how students from various parts of the country are doing.

They can ask the AI tools different questions to see, should we strengthen relationships with certain high schools where we might have only admitted like two to three kids, but all those two, three kids are doing well. So can you just talk about that a little bit? Because development, let's say, of an incoming college class or at least incoming set of applicants is also a major part of what these admissions offices do.

SWAAK: Oh, absolutely. I would say that was probably the primary one that I heard. And it's actually one that goes beyond two years ago. Predictive analytics is something that college admissions offices, many of them have been doing for years. And yeah, I think a large part of it is tied to, I feel like I'm reading, hearing about all the time.

This looming enrollment cliff, colleges are always looking for new students or wanting to get a better sense of which students, potential applicants might attend. So that they can better distribute and funnel resources and time and energy. And, yeah, for that particular piece you're referring to, I remember the VP of enrollment I was speaking to was like, look, I need to tell my budget team like six to eight months before we have the final list of students finalized, about how many students we can expect for the next year, so that they can make the budget.

And they want a number that isn't just being pulled randomly out of a fishbowl, I think is how he was saying it. He was like, we want to have some sense of an informed estimate. And the use of AI algorithms to, like I was saying before, compare applicants data against a historical data set, which is current and former students, and then grouping them into a likelihood of attending.

He said that was really helpful, both in providing that estimate, but also like I was saying before, when you're an office that has limited time and resources and energy, that group that's on the fence, maybe that's where you're putting a little bit more of that time and energy.

CHAKRABARTI: Rick, I have 10 seconds left.

So I have a quick yes, no for you. Do you recommend students, potential applicants ask universities now if they're using AI in the admissions process, and if so, how?

CLARK: I do. I think that students should feel comfortable asking any question that they have, that is their responsibility and it's important for them to understand how decisions are being made and how they're being reviewed.