Estimating the Prevalence of Generative AI Use in Medical School Application Essays
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
Generative artificial intelligence (AI) tools became widely available to the public in November 2022. The extent to which these tools are being used by aspiring medical school applicants during the admissions process is unknown.
Methods
We retrospectively analyzed 6,000 essays submitted to a U.S. medical school in 2021– 2022 (baseline, before wide availability of AI) and in 2023–2024 (test year) to estimate the prevalence of AI use and its relation to other application data. We used GPTZero, a commercially available detection tool, to generate a metric for the likelihood that each essay was human-generated, P human , ranging from 0 (entirely AI) to 1 (entirely human).
Results
Fully human-generated negative controls demonstrated a median P human of 0.93, while AI-generated positive controls demonstrated a median P human of 0.01. Personal Comments essays submitted in the ‘23–‘24 cycle had a median human-generated score of 0.77 (95% confidence interval 0.76–0.78), versus 0.83 (95% CI 0.82–0.85) during the ‘21–‘22 cycle. Approximately 12.3 and 2.7% of essays were evaluated as having P human < 0.5 in the test and baseline year, respectively. Secondary essays demonstrated lower P human than Personal Comments essays, suggesting more AI use. In multivariate analysis, younger age, visa requirement, and higher GPA were significantly associated with lower P human . No differences were observed in gender, MCAT score, undergraduate major, or socioeconomic status. P human was not predictive of admissions outcomes in uni- or multivariate analyses.
Conclusions
An AI detection algorithm estimated significantly increased use of generative AI in 2023-2024 medical school admission applications, as compared to the 2021-2022 baseline. Estimated AI use demonstrated no significant differences in admissions decisions. While these results provide information about the applicant pool as a whole, AI detection is imperfect. We recommended exercising caution before deploying any AI detection tools on individual applications in live admissions cycles.
Description
Medical school applicants increased their use of generative AI to write application essays in the most recent admissions cycle, but this use did not confer an admissions advantage.
