Student use of artificial intelligence (AI) in K-12 education and its implications seem like the elephant in the room in The Phoenix’s reporting on the use of SAT/ACT scores in admissions decisions. Evidence suggests that middle and high school students are leaning heavily on generative AI at alarmingly high rates, often to replace complex thinking — to give them ideas for papers, to outline papers, to write the papers, to analyze articles, to do their math and science homework. Even worse, another body of research has shown that this kind of cognitive offload not only deprives people of the opportunity to build skills and knowledge but can actually reduce cognitive capability in general.
We haven’t felt the effects of these trends strongly at Swarthmore yet. The incoming first-year class is the first to have had easy access to ChatGPT or other large language models during the entirety of their junior and senior years in high school. But every year going forward, the incoming class will have had even more time leaning on AI to replace actual thinking during their very formative educational years. The administration seems to want to lock this test-optional policy in place for five more years. By that time, we’ll have an entering first-year class that has had access to ever more powerful AI tools since the sixth grade.
What does this have to do with standardized testing? The future looks like one in which there will be a huge number of high school seniors across the country who look great on paper — GPAs close to 4.0, crisply written essays, etc. — who have never read a book cover to cover, never come up with a thesis for a paper or composed one on their own, rarely done their own math homework, and rarely read or analyzed an article without dropping it into an AI assistant first. And then, of course, there will be the huge number of students who also look great on paper and actually have put in the work — coming up with their own ideas, writing their own papers, puzzling through complex and difficult texts, burnishing their critical thinking and analytic skills, and generally doing all the things future students have always done at the high school level that prepare them to succeed at a place like Swarthmore. The fundamental problem is that these two populations will be increasingly difficult to differentiate in the admissions process.
Standardized test scores, despite flaws, are one of the most important data points Swarthmore and other institutions will have to differentiate those populations. Of course, they shouldn’t be leaned on exclusively and instead taken as one data point among many. But they will have huge value because — unlike grades or essays — they cannot be gamed or generated via AI. In my opinion, it is extremely shortsighted for the administration not just to abandon that data point (which it effectively does when test scores are optional, since only high scores are submitted) but to decide to do so for five years, which is an eternity in a world being rapidly transformed by AI.
Sam Handlin ’00 is an associate professor of political science at Swarthmore College.
