Can AI Mitigate Bias in Writing Letters of Recommendation?

Generative artificial intelligence in graduate medical education

Generative artificial intelligence (GenAI) is rapidly transforming various sectors, including healthcare and education. This paper explores the potential opportunities and risks of GenAI in graduate medical education (GME). We review the existing literature and provide commentary on how GenAI could impact GME, including five key areas of opportunity: electronic health record (EHR) workload reduction, clinical simulation, individualized education, research and analytics support, and clinical decision support. We then discuss significant risks, including inaccuracy and overreliance on AI-generated content, challenges to authenticity and academic integrity, potential biases in AI outputs, and privacy concerns. As GenAI technology matures, it will likely come to have an important role in the future of GME, but its integration should be guided by a thorough understanding of both its benefits and limitations.

Artificial Intelligence Compared to Manual Selection of Prospective Surgical Residents

BACKGROUND

Artificial Intelligence (AI) in the selection of residency program applicants is a new tool that is gaining traction, with the aim of screening high numbers of applicants while introducing objectivity and mitigating bias in a traditionally subjective process. This study aims to compare applicants screened by an AI software to a single Program Director (PD) for interview selection.

METHODS

A single PD at an ACGME-accredited, academic general surgery program screened applicants. A parallel screen by AI software, programmed by the same PD, was conducted on the same pool of applicants. Weighted preferences were assigned in the following order: personal statement, research, medical school rankings, letters of recommendation, personal qualities, board scores, graduate degree, geographic preference, past experiences, program signal, honor society membership, and multilingualism. Statistical analyses were conducted by chi-square, ANOVA, and independent two-sided t-tests.

RESULTS

Out of 1235 applications, 144 applications were PD-selected and 150 AI-selected (294 top applications). Twenty applications (7.3%) were both PD and AI selected for a total analysis cohort of 274 prospective residents. We performed two analyses: 1) PD-selected vs. AI-selected vs. Both and 2) PD-selected vs. AI-selected with the overlapping applicants censored. For the first analysis, AI selected significantly: more White/Hispanic applicants (p < 0.001), less signals (p < 0.001), more AOA honors society (p = 0.016), and more publications (p < 0.001). When censoring overlapping PD and AI selection, AI selected significantly: more White/Hispanic applicants (p < 0.001), less signals (p < 0.001), more US medical graduates (p = 0.027), less applicants needing visa sponsorship (p = 0.01), younger applicants (p = 0.024), higher USMLE Step 2 CK scores (p < 0.001), and more publications (p < 0.001).

CONCLUSIONS

There was only a 7% overlap between PD-selected and AI-selected applicants for interview screening in the same applicant pool. Despite the same PD educating the AI software, the 2 application pools differed significantly. In its present state, AI may be utilized as a tool in resident application selection but should not completely replace human review. We recommend careful analysis of the performance of each AI model in the respective environment of each institution applying it, as it may alter the group of interviewees.

Letters of Reference in the Current Era

ABSTRACT

Letters of reference (LORs) are a common component of the application process for residency training programs. With the United States Medical Licensing Examination Step 1 transitioning to pass/fail grading and with the increasing use of holistic review, the potential role of LORs is rising in importance. Among some key benefits are the ability to provide a broader and more holistic view of applicants, which can include highlighting elements of experiences or skills that could be missed in their application, as well as providing a third-party assessment of the applicant external to their rotation experiences. However, LORs also face issues, including variation in quality, challenges with comparability, and risk of bias. In this article, the authors discuss the unique benefits, limitations, and best practice recommendations for LORs in academic medicine. The authors also discuss future directions, including the role of artificial intelligence, unblinded, and co-created LORs.

Brain versus bot: Distinguishing letters of recommendation authored by humans compared with artificial intelligence

Objectives

Letters of recommendation (LORs) are essential within academic medicine, affecting a number of important decisions regarding advancement, yet these letters take significant amounts of time and labor to prepare. The use of generative artificial intelligence (AI) tools, such as ChatGPT, are gaining popularity for a variety of academic writing tasks and offer an innovative solution to relieve the burden of letter writing. It is yet to be determined if ChatGPT could aid in crafting LORs, particularly in high-stakes contexts like faculty promotion. To determine the feasibility of this process and whether there is a significant difference between AI and human-authored letters, we conducted a study aimed at determining whether academic physicians can distinguish between the two.

Methods

A quasi-experimental study was conducted using a single-blind design. Academic physicians with experience in reviewing LORs were presented with LORs for promotion to associate professor, written by either humans or AI. Participants reviewed LORs and identified the authorship. Statistical analysis was performed to determine accuracy in distinguishing between human and AI-authored LORs. Additionally, the perceived quality and persuasiveness of the LORs were compared based on suspected and actual authorship.

Results

A total of 32 participants completed letter review. The mean accuracy of distinguishing between human- versus AI-authored LORs was 59.4%. The reviewer's certainty and time spent deliberating did not significantly impact accuracy. LORs suspected to be human-authored were rated more favorably in terms of quality and persuasiveness. A difference in gender-biased language was observed in our letters: human-authored letters contained significantly more female-associated words, while the majority of AI-authored letters tended to use more male-associated words.

Conclusions

Participants were unable to reliably differentiate between human- and AI-authored LORs for promotion. AI may be able to generate LORs and relieve the burden of letter writing for academicians. New strategies, policies, and guidelines are needed to balance the benefits of AI while preserving integrity and fairness in academic promotion decisions.