A researcher from Salinas is using artificial intelligence to make college admissions more equitable. – Monterey County Weekly

A little over a week ago, AJ Alvero was thinking about Confederate monuments.

Specifically, he read that a monument honoring Confederate General Robert Selden Garnett was removed from the lawn in front of Colton Hall in Monterey. And then he read the monument was replaced with a plaque that still named Garnett as the designer of the state seal of California, but left out his Confederate legacy. Not good enough, Alvero thought.

A few days later, someone tore out the new plaque and left a sign behind, saying Celebrate real heroes. No place of honor for racists.

Alvero, a doctoral student at Stanford University, says hes not the one who did it. I was very strongly toying with the idea but someone beat me to punch, he says.

It wouldnt be the first time that Alvero acted to strip a Confederate name from a public space.

Growing up in Salinas, Alvero often crossed an intersection that was officially known as Confederate Corners. In the wake of the deadly neo-Nazi rally in Charlottesville, Virginia in 2017, he organized a community effort to change the name of the intersection, enlisting the support of the Monterey County Board of Supervisors. He wanted to call it Campesino Corners to honor the areas farmworkers. The board selected the name Springtown.

The intersection is unremarkable in appearance, and the fact it has a name is not very widely known which is why Alvero thought that renaming would be low-hanging fruit in the effort to undo American racism. But he was wrong and did not anticipate the backlash and vitriol against him.

Now, a few years removed, Alvero is still focused on the power of words and language in our public life. But hes leading a more sophisticated and systemic charge on bias. His target is college admissions and his instrument of change is artificial intelligence.

He recently published a groundbreaking peer-reviewed study that argues its possible to combat bias in the admissions process by analyzing the language used in application essays to detect demographic patterns.

Days before the killing of George Floyd on May 25, which triggered a national reckoning on racism, the University of California took a giant step to address stark disparities in college admissions. By a 23-0 vote, the university systems governing board decided to phase out the use of the SAT and ACT in the admissions process because evidence shows that they drive inequity. A few weeks later, the board voted, unanimously again, to support the restoration of affirmative action in California, which had outlawed the practice in 1996 through Proposition 209.

For university admissions officers, these two decisions increased the focus on evaluating personal essays and circumstances of the hundreds of thousands of applicants they screen each year.

The U.S. Supreme Court, while not exactly endorsing affirmative action, has ruled that consideration of race in admissions is constitutional, as part of a highly individualized, holistic review of each applicants file, giving serious consideration to all the ways an applicant might contribute to a diverse educational environment.

A recent and closely watched lawsuit against Harvard University challenged the use of race as a factor in admissions, claiming the university discriminates against Asian American students. Ultimately, a federal judge, Allison Burroughs, rejected the lawsuit.

In her decision, Burroughs wrote that Harvards process of weighing test scores alongside subjective personal essays survives strict scrutiny.

But, she added, the process could be improved: admissions officers should receive training on recognizing implicit bias. Statistical analysis should be used to discover and counter race-related disparities.

The decision enshrined the continued consideration of race while also raising the bar on what admissions officers must do to achieve fairness. Heres where Alveros research comes in.

Alvero, who studies education, sociology, language and data science, teamed up with other Stanford graduate students to explore how a more equitable future for college admissions might be achieved.

In other words, if SAT scores become obsolete, and personal essays become more central, how can the selection process be improved to survive new constitutional challenges?

Like most scholarly research, the starting point of Alveros academic paper is data.

In this case, the data was 283,676 application essays submitted by 93,136 applicants who identified as Latino or Latina. In the first study of its kind, Alveros team used computational analysis to discover patterns across a mass corpus of essays.

By running the essays through relatively simple computer algorithms, the team found, they could correctly predict the gender and income level of an applicant about four out of five times.

In another fascinating finding, the paper showed which words are more likely to be used by different demographic groups male versus female and low-income versus high-income applicants. And the purpose of admissions essays, it turns out, was originally to keep certain students out.

In an interview with the Weekly, Alvero spoke about his findings and what they mean.

AJ Alveros academic career started at San Diego State University but he soon dropped out. He came back to Salinas to lick his wounds and start over. Eventually, Alvero (left) made his way to Stanford University where he is a fourth-year Ph.D. student studying education and data science. (right) Alvero pushed the U.S. Geological Survey to change the name of this intersection in Salinas from Confederate Corners to Springtown.

Weekly: Artificial Intelligence and big data are complicated topics even for people who have grown up with technology. How would you start explaining your research to your grandparents?

Alvero: Lots of studies argue that standardized tests like the SAT and ACT are biased, by race, by social bias, by gender in certain ways, and that we shouldnt use them in college admissions.

So if thats the case, what about the admissions essay? We have this idea that the admissions essay gives students a chance to talk about their true selves. Yet so far, the essays havent been placed under scrutiny, as the test scores have. Thats where my research comes in.

Why would a language performance, like writing an essay, be less biased or more biased than a test score?

How did you get interested in answering that question?

Ive always been interested in language and social issues. I read about the history of college admissions essays, and they were actually designed at Harvard [in the early 20th century] to filter out Jewish applicants. Its pretty incredible. The president of Harvard was one of these old-money Boston families and he decided, We have too many Jewish students on campus.

He realized that all of his old clientele, which are the very wealthy, white Protestant elites of New England those applicants are not passing the entrance exam. So he created the personal statement, introduced extracurricular activities, introduced the letters of rec all these subjective measures to give WASP elite students a chance. It worked. Jewish enrollment was cut in half.

Thats the history. And I thought, well, these essays are still being used widely.

I also reflected on my time as a high school teacher in Miami helping students write these essays. Students from immigrant backgrounds tended to write about certain things. Students who worked with Teach for America teachers, they tended to write about certain things. So I noticed there was a lot of patterning in the types of narratives that students were deploying in their personal statements.

At Stanford, I got really interested in learning about just using text as big data in computational methods of analysis. There have been advances in computational methods to analyze texts.

But in part, it was me being at the right place at the right time.

What do you mean?

The idea of using written texts as a form of data has become a very popular idea at Stanford, you see a lot of researchers in many different departments and fields leveraging text as data.

But the texts in your study are not Wikipedia articles. What you obtained was much more exclusive, even confidential: Nearly 300,000 admissions essays by self-identified Latino applicants. How rare is that and how did you get them?

I dont want to toot my own horn or anything, but its extremely rare. So, to our knowledge in my lab, were the first ones to use these computational methods on a large collection of admissions essays.

What I did was email a couple of admissions officers, and only one of them got back to me. I can only tell you which university off the record.

Deal. I wont reveal more than what it says in your study, that it was a large public university system. Did the data come with strings attached?

They did come with strings attached. My original pitch was, What are the Latinx kids writing about? There are lots of people under this umbrella, I figured I could get a lot of data, and its a category and a group of people that Im very interested in, partly because Im part of that umbrella as a Cuban American.

So, I asked them, Can I get admissions essays written by Latinx students? And they said, Sure, how many do you want? I said, Ill take all of them. Eventually, I learned they had an interest specifically in Latinx student essays in hope of increasing Latinx enrollment.

So they wanted your expertise, meaning that others could have asked and gotten the essays. Youre the one who went ahead and tried.

I think every university wants to get better at reading these essays. And no one wants to be subject to that kind of lawsuit like Harvard. No one wants to face that kind of scrutiny.

You ran the essays through the algorithms and found that they were able to predict something really interesting. Can you tell me about that?

We found that even a relatively simple machine learning algorithm was able to predict the gender of the applicants about 80 percent of the time and whether or not they were above or below the median income, which was a proxy for higher or lower income, about 70-something percent of the time.

Boy Scouts and foreign countries, thats basically what the higher income applicants are writing about (see diagrams, p. 26). And with boys, some of the words most associated with them were hardware, chess, Lego and Rubiks Cube. If we were to survey every single college admissions essay reader in the country and ask, what do you think about if a student wrote about chess? How would you describe that student? They might read chess, Rubiks Cube, hardware, Legos. And they might think, Wow, this is a very intellectual person.

What the data is showing is that are also words that boys are just using much more often than girls. Do our admissions readers realize this? Are they being trained to recognize this? On the flip side, theres makeup and cheerleading. Do people think makeup is also intellectual and very engaging? I dont think so.

I also found it fascinating that girls are talking about being girls, using words like Latina, daughter and female but boys are not bringing up their gender.

Yes, and are our admissions officers being trained for this? Do they even know this? I think the answer is no. Im hoping to connect this research to actually practice in college admission.

An analysis of word frequencies across nearly 300,000 personal essays revealed which words were most characteristic of different demographic groups.The top left were the words favored by female applicants, the top right by male applicants. The bottom left were words used more frequently by higher-income applicants, the bottom right by lower-income applicants. (ELD and ETS are acronyms related to English language learners.) The size of each word reflects the frequency of use.

How would you do it?

Its very complicated and no ones really sure if and how its going to work. But a common practice in college admissions right now is when an application reviewer is looking at test scores and GPAs from an applicant, theyll also have some contextualization.

For example, lets say a kid gets a pretty solid but not fantastic SAT score. How did everyone else at that school do? Maybe that kid didnt get a perfect score but its way better than everyone else did at their school. Then an admissions reviewer could take that into account. That would be the idea: trying to contextualize the essays. Because at the moment, we dont have anything close to that.

What if you took potential insight from the algorithm, and combined that with human insight? Maybe thatll be better. We have got to find out. So thats what Im trying to be the person to find out.

How does this pursuit connect to the fact that you are Latino and were born and raised in Salinas?

I dont want to just straight-up talk smack about Salinas. But there was a lot of prejudice and there were a lot of biases and stereotypes and racism against Mexican people and Central American people.

The way it would work for me was almost like two-factor authentication, like where you first type in your password, but then it needs a code from your phone. Im fair-skinned and have light eyes. And a lot of people will look at me and be like, Oh, yeah, no, youre not Mexican.

But then I start talking, or mention Im Cuban. Ah, you pass the first password. But the second, no ones answering the call. Lots of my friends growing up, they never got past that first step of the password. But for me, I was able to move in and out of the crowd.

Seeing the treatment of Mexican Americans always bothered me so much. And, I always hoped that I could be in a position where I can speak out on it and people would hear me. Im hoping this research can be my first big way to do that.

See the article here:
A researcher from Salinas is using artificial intelligence to make college admissions more equitable. - Monterey County Weekly