“Frequently, medical technology outstrips our ability to understand its consequences” - Brandy Schilace
If Hospital Walls Could Speak:
From experiments on enslaved women in the 1600s to denied access to treatment today, healthcare discrimination has disproportionately affected Black communities in the United States for generations. Many of these injustices stem from scientific racism, the belief that Black individuals are biologically different from others— if their individuality is acknowledged at all. Many healthcare workers who upheld this pseudoscience exploited their authority to justify these immoral practices, leaving behind a legacy that continues to affect both the trust and medical experiences of Black patients today.
Consider the Tuskegee Study of Untreated Syphilis in the Negro Male conducted by the US Public Health Service from 1932 to 1972. It targeted non-consenting Black men under the false pretense of being “treated for bad blood” (ZME Science, 2024). In reality, researchers intentionally withheld diagnosis and treatment, even after 1943 when Penicillin became the approved cure for syphilis (Centers for Disease Control and Prevention, 2024). In fact, the study was designed to continue until the men had passed away from the disease. As of 2023, Black men were about 4 times as likely as White men to report having syphilis in the US (Statista, 2025). Although the study ended, its legacy persisted, fueling ongoing disparities and medical mistrust within Black patient populations.
The New Face of an Old Disease:
Although legal protections exist to combat such practices, scientific racism still implicitly (and often, explicitly) influences healthcare today. In response, some have proposed using Artificial Intelligence, utilizing AI’s ability to objectively reason to reduce the potential for clinician biases. A case in point: a 2022 NIH study found that although a greater proportion of Black patients report symptoms of knee osteoarthritis, the condition remains severely underdiagnosed and undertreated in Black communities. This results in worse symptoms and health over time. While many factors contribute to this, bias remains a central one. In fact, most healthcare providers carry racial implicit bias, often overlooking Black patient’s symptoms or attributing them to causes like stress while responding more attentively to the concerns of White ones (NIH, 2022). AI, however, can detect patterns that human diagnostic tools may miss. A 2021 University of Chicago study revealed AI’s ability to explain “pain disparities in about 43% of African American patients who are not diagnosed as having osteoarthritis using KLG [a test used to determine the severity of osteoarthritis] score.”
Despite initial optimism, AI’s promises of alleviating health disparities are far from fulfilled. A 2022 report by The Lancet Digital Health showed that some US AI models can accurately predict a patient's race from their medical scans. As a result, these models become susceptible to inheriting the racial biases that they aim to dispel— a consequence of the data used to train these systems. For example in dermatology, CNNs, AI models that detect skin disorders, are often trained with images of White patients. Consequently, these models make significant errors when diagnosing Black patients, with “approximately half the diagnostic accuracy compared with what their creators originally claimed” (NIH, 2021). This is concerning as Black men have the lowest survival rates for skin conditions such as melanoma: 52% as compared to 75% for White men (American Academy of Dermatology Association, 2023).
Debugging the Bias:
Does this mean that AI has no place in healthcare? Not quite. AI relies on proxies, or substitutes of a variable that is difficult to directly measure. While this can be helpful in healthcare for making quick assessments, biased proxies can make for inaccurate conclusions. For example, because fewer funds are allocated towards the health needs of Black patients, some models assume that they are healthier than their White counterparts (NIH, 2021). This flawed proxy leads to misleading conclusions and undermines the need for healthcare reform.
AI is a powerful tool. The truth, however, is that for it to serve its intended purpose in healthcare, change must occur both in the information used to train AI models and our understanding of the historical biases that it may reflect. An algorithm is only as neutral as the data it is built upon. It can be difficult to detect systemic biases and so it’s important for healthcare providers to educate themselves on the historical basis of medical discrimination and actively evaluate AI claims. There may still be a long way to go with AI, but the future of healthcare does not need to look like the past.
References
Centers for Disease Control and Prevention. (n.d.). The untreated syphilis study at Tuskegee Timeline. Centers for Disease Control and Prevention. https://www.cdc.gov/tuskegee/about/timeline.html
Elflein, J. (2025, February 17). Syphilis rates by race/ethnicity and gender U.S. 2023. Statista. https://www.statista.com/statistics/622887/syphilis-rate-in-the-us-by-ethnicity-and-gender/
Gichoya, J. W., Banerjee, I., Bhimireddy, A. R., Burns, J. L., & Celi, L. A. (n.d.). AI recognition of patient race in medical imaging: A modelling study - the lancet digital health. Digital Health. https://www.thelancet.com/journals/landig/article/PIIS2589-7500(22)00063-2/fulltext
Hall, W. J., Chapman, M. V., Lee, K. M., Merino, Y. M., Thomas, T. W., Payne, B. K., Eng, E., Day, S. H., & Coyne-Beasley, T. (2015, December). Implicit racial/ethnic bias among health care professionals and its influence on health care outcomes: A systematic review. American journal of public health. https://pmc.ncbi.nlm.nih.gov/articles/PMC4638275/#:~:text=Main%20Results
Nachimuthu, S. (n.d.). Reducing racial disparities in knee pain using AI. The University of Chicago Booth School of Business. https://www.chicagobooth.edu/research/center-for-applied-artificial-intelligence/research/knee-pain#:~:text=Reducing%20health%20disparities%20using%20artificial,knee%20pain%20in%20these%20patients
Norori, N., Hu, Q., Aellen, F. M., Faraci, F. D., & Tzovara, A. (2021, October 8). Addressing bias in big data and AI for health care: A call for open science. Patterns (New York, N.Y.). https://pmc.ncbi.nlm.nih.gov/articles/PMC8515002/#bib14
Puiu, T. (2024, August 7). Remembering the tuskegee experiment: When Rural Alabama Black men were intentionally exposed to syphilis with no treatment. ZME Science. https://www.zmescience.com/science/news-science/what-was-tuskegee-experiment/
Saunders, R. (n.d.). Largest study on racial differences in men with melanoma shows men with skin of color have lowest survival rates. American Academy of Dermatology Association. https://www.aad.org/news/melanoma-study-men-skin-of-color-lowest-survival-rates
Wentt, C. L., Farrow, L. D., Everhart, J. S., Spindler, K. P., & Jones, M. H. (2022, September 22). Are there racial disparities in knee symptoms and articular cartilage damage in patients presenting for arthroscopic partial meniscectomy?. JB & JS open access. https://pmc.ncbi.nlm.nih.gov/articles/PMC9489158/