AI in Dermatology: Discussing Downsides to the Adoption of Artificial Intelligence

One of the fastest-evolving topics in today’s world is artificial intelligence (AI) and its practical application in fields such as medicine.

Discussions of the use of AI tools and of machine learning advancements have been present in the broader cultural conversation for decades.1 Yet when OpenAI launched its ChatGPT model in late 2022, the rapid and widespread adoption of AI into everyday life became, in a very small timeframe, a reality for those with Internet access.2

While conversations about AI’s use in the medical field are not new, machine learning’s practical implementation in dermatology specifically is a comparatively new topic. Using clinical and dermatoscopic data and images for AI models to help diagnose such disorders as skin cancer, psoriasis, and atopic dermatitis has become more widely studied.

Yet, despite its apparent potential in medicine and other fields, there is an increasing skepticism of AI among many in the US.3,4 According to polls, more Americans believe AI will worsen problem-solving skills as opposed to improving them. To many, concerns over its use in fields such as medicine boil down to fears regarding overreliance on such tools and regarding the impersonality of AI.

While many believe AI will be beneficial for clinicians in fields such as dermatology, with benefits to patients looking for melanoma and other skin disorders being touted, the topic of concern over reliance on AI in dermatology is not often highlighted. If an objective view of the costs versus benefits is to be obtained, then an exploration of concerns over machine learning may be necessary.

In this installment of our This Year in Medicine feature series, AI downsides are explored with 3 experts, all of whom were interviewed about their views and their own research on the topic. The HCPLive team spoke with Vinod Easwaran Nambudiri, MD, MBA, MPH, of Brigham and Women's Hospital, Lia E. Gracey Maniar, MD, PhD, an assistant professor of Medicine at the Department of Internal Medicine of Dell Medical School, and Renata Block, MMS, PA-C, a physician associate (PA) and instructor at Rush University.

This feature explores differing views on AI in dermatology, with a particular focus on its effects on training dermatologists and other clinicians, shedding light on an important, evolving topic in the medical field.

The Benefits of AI in Dermatology

As this technology promises new efficiencies and diagnostic support for clinicians, AI’s rapid integration into the field of dermatology has left clinicians and educators scrambling to keep up. The benefits of its use are becoming widely recognized, however, and there is no denying the potential of AI in dermatologic care.5,6

“I think AI, or artificial intelligence, is such an important topic because we have to understand what it is and what it does,” Block explained.7 “I think if you become savvy with understanding the process of how it's developed and what it's intended to do, then you're going to know, ‘This is going to not replace me, but it's going to help augment my decision-making process in the clinic.’”

Generative AI and advanced data synthesis techniques were described in 1 study as enabling the formation of high-quality, representative, and diverse training data for AI models in dermatology.8 The benefits of large foundational models’ reduction of the necessity of massive datasets was also cited. Improvements in diagnostic accuracy are also among the most important long-term goals with AI’s implementation.

In a previous feature by HCPLive, Harald Kittler, MD, a professor of dermatology for the Medical University of Vienna, had discussed his team’s findings published in Nature Medicine on the accuracy of AI results which had been incorporated into human decision-making criteria.9 Kittler noted his team’s data suggested a 12% improvement in accurate skin cancer diagnoses made by dermatologists.

Additional research has come out demonstrating a variety of benefits. In a study, the improvements in the quality of skin disease photographs, resulting from an AI decision support tool’s use, along with a machine learning algorithm, were highlighted.10 A systematic review led investigators to acknowledge AI’s accuracy in identifying and classifying pigmented lesions in patients with skin of color.11

However, despite the wide variety of positive data acknowledged by AI and machine learning proponents, there remain cracks in the facade.

Clinician Views on AI Accuracy and Helpfulness

Dermatologists’ views on such technological advancements have also not yet been widely explored. In 1 study, investigators surveyed dermatologists and found 77.3% of those asked agreed that AI would be a positive influence on dermatology.12 Additionally, 79.8% of survey respondents agreed that AI should be a part of medical training. AI in medical training was explored in a JAMA opinion piece authored in part by Nambudiri and Gracey Maniar.13

“I'm cautiously optimistic that this could help in certain aspects for dermatology residency training,” Nambudiri said in an interview. “I think certainly there's always something that we could improve upon if there is an opportunity to really tailor curriculum, since all residents are not the same, some may struggle in certain areas, and to be able to identify and potentially tailor curricula to help support that resident.”

Gracey Maniar echoed Nambudiri’s cautious optimism on the technology’s implementation.

“I'm cautiously optimistic that this could help in certain aspects for dermatology residency training,” Gracey Maniar explained. “I think, certainly, there's always something that we could improve upon. If there is an opportunity to really tailor the curriculum, since not all residents are the same, and some may struggle in certain areas.”

In her interview with HCPLive at the Society of Dermatology Physician Associates (SDPA) Fall Conference in Texas, Renata Block was slightly more optimistic in her discussion of AI’s use in dermatology.7 She echoed the views of many clinicians who view the technology’s potential relatively optimistically, describing it as essentially another tool helping to augment work.

“I think with anything new, there's always hesitancy because we don't understand it,” Block said. “That's why I thought it was such an important topic to discuss, because I want my colleagues to understand AI, understand how to approach AI, and then feel comfortable utilizing it and honing in on the importance that it's not there to replace us. It's there to augment our practice. I throw out that I think artificial is kind of the wrong terminology to use. It's really augmented intelligence.”

Biased Datasets, Hallucinations, and Patient Security

The downsides of AI technology’s use in medicine and in dermatology are becoming increasingly acknowledged. For example, despite the aforementioned systematic review’s positive findings, the study’s investigators also highlighted significant discrepancies in the number of models developed in populations with skin of color. This was particularly true of patients with Fitzpatrick type IV-VI.

“In our current medical education paradigm or platform, we know that for many years, the representation of a diversity of skin tones in dermatology educational materials has been lacking,” Nambudiri noted. “...I think it's incumbent upon us as physicians, as dermatologists, as the clinicians that take the best care of skin, hair, nails, etc, for us to be representing a wide swath of the clinical spectrum. Every skin tone, every hair type, etc., really needs to be built into these learning data sets such that the AI platforms or tools that we're using can help us in the care of all of our patients.”

AI does not know what it does not know, Nambudiri explained, and dermatology clinicians must build accessible and wide-ranging resources to enhance diagnostic abilities. Issues related to diversity in clinical research are 1 of many problems in this space, with many concerned about the accuracy of image detection and classification without diverse datasets.

“I think it's a really important area to highlight, because the hard part is, is with what we want our tools to be representative of the patients that we see, and right now, they're not,” Gracey Maniar expressed in her interview. “And some of the tools aren't even disclosing how they're trained, either proprietary data sets, or they don't break down by skin tone how their data was trained. So, it's a little bit of a black box, and it's pretty opaque for us to even be able to evaluate how their tools will perform for all populations.”

Additionally, other limitations have been highlighted in recent studies. These include image quality variations, liability concerns, interpretability concerns, acceptance among patients, liability, multidisciplinary coordination challenges, data bias risks, and the potential loss of clinical skills.14,15

In her presentation on the topic of AI featured at the 2025 SDPA Fall Conference, Block pointed out another key issue observed among clinicians and everyday users of models such as ChatGPT: AI ‘hallucinations.’

“We are training these models to detect cancer by using images or predicting outcomes based on historical data, but this is very expensive and very time-consuming,” Block said. “This is a great way to integrate AI into any tool, really, because there's also unsupervised, and that's where we can get in a lot of trouble. Because the algorithm finds patterns on its own. There are no predicted outcomes. And applications include grouping patients with similar symptoms or genetic markers. However, this is very difficult to validate, which can be a huge issue.”

Many machine learning models face the issue of hallucinations. This phrase refers to models’ generation of content not based on real or existing data but rather produced by a model's extrapolation or creative interpretation of training data.16 This has also been experienced among ChatGPT users who, when feeding questions into the machine learning model, occasionally receive incorrect answers.

“[There's] a lot of bias that can happen, and even hallucinations; they're going to start making things up,” Block said. “That is why it's important that you read what the AI does to make sure that it's accurate. But the other issue is transparency and, of course, privacy and security, which can be easily compromised.”

In her interview on the topic, Gracey Maniar stressed her view on the need for a healthy amount of skepticism regarding this technology and its use in dermatology. She expressed the continued value of textbooks, primary sources, and verification of information ascertained via large language models (LLMs) or generative AI tools like ChatGPT.

“Part of that is getting a foundation on what questions we should be asking,” Gracey Maniar explained. “Understanding just because you're typing in a question, for instance, into Gemini, that what you're getting spit back may not be the real truth, and that you need to actually verify with primary sources.”

The Erosion of Core Clinical Skills and Human Contact

One of the most central concerns related to AI’s widespread implementation in medicine is the erosion of skills among regular users. There are concerns over machine learning’s impact on the ability to learn and navigate the complexities of medical training. This topic was explored in Gracey Maniar and Nambudiri’s JAMA paper, wherein challenges in dermatology residency were highlighted.

Dermatology residents’ scholarly work is already being affected by the use of LLMs to provide background research, to help write drafts of manuscripts and abstracts, and to develop slides for educational presentations.13 Such models have also been implemented to generate simulated questions and prepare medical trainees for their standardized examinations.

“Documentation pervades our lives every day in healthcare, and I think it is just as big a source of burnout,” Gracey Maniar said. “...I think it could actually help with the documentation burden and the production pressure that a lot of us feel having to get through so many notes. On the other hand, one of the items we brought up for concern was that we still need residents to learn how to write good notes.”

Gracey Maniar and Nambudiri both stressed the importance of the thought process and learning resulting from taking strong notes during residency training.

“I think part of it is not just getting words on paper, on what we did that day, but it's also that writing can help you think through a differential diagnosis,” Gracey Maniar added. “There's a little bit of an art of medicine, and how you're communicating with other specialists, with the primary care physician who may be reading your note. So, we want to make sure that our residents still retain or learn and develop those skills.”

In his interview, Nambudiri noted the importance of dermatology residents requiring the ability to override AI and other technologies when they are incorrect. He pointed out an example of overreliance on technology and its impact.

“An example that happened to me recently is I had a patient for whom I ordered a radiology test, a CAT scan, and the interpretation was generated through the assistance of dictation software,” Nambudiri explained. “It’s a very first generation of sort of speech-to-text processing tool…In one part of the report, it made a statement appeared to be inconsistent with something later on in the final overall impression. At one point, it said something had changed. At the end, it said it was unchanged. Those are two completely contradictory things that cannot be true both at once.”

Nambudiri described having had to message the radiologist who read the study. This, he expressed, is a classic example of overreliance upon technology in medicine. Additionally, both clinicians pointed out the importance of the human element to medicine and the impersonality of AI tools and telemedicine practices.

“A lot of the time, what we hear from patients and in the literature is that it's people just want to have that empathy factor, and I don't think that is easily replaceable,” Gracey Maniar said. “For instance, we are trying to get a survey and a qualitative study submitted where we asked a federally qualified health center to get people's attitudes towards AI and healthcare, both from patients at a safety net clinic, the physicians, the non-physician providers, and the staff who serve them. One of the big concerns was losing that human factor.

The connection between the patient and the clinician, Gracey Maniar highlighted, is something concerning many patients with fears regarding technology’s increasing use in medicine.

“I think one of the things that seemed to be at risk was more of the electronic communications right now,” Gracey Maniar explained. “At most institutions, it's still a human who is responding to those granted. At some institutions, they're piloting or actively using some of the chatbots or the automated suggested responses, for instance, to portal messages. Since a lot of physicians are getting inundated with the portal messages that are coming, I think the primary physician input into some of those portal messages could start to be overshadowed a bit.”

Addressing Concerns Over AI in the Future

Despite many of these potential downsides to AI’s implementation, all 3 experts noted their own cautious optimism with regard to many of the technological innovations impacting medicine and dermatology in particular. Looking ahead, each of the speakers noted the need for these tools to have regulation and to be more understood, rather than feared.

During his discussion, Nambudiri still described himself as a proponent of AI utilization for ambient documentation. He highlighted his own use of an AI-based scribe for years, expressing the importance of always considering ways to tweak one’s efficiency with documentation using AI-based technologies.

“One of the things that our own institution is thinking through is how to integrate this into residency training,” Nambudiri said. “One of the models that has become quite popular is sort of a graduated introduction of these AI tools for documentation, meaning that they may not be available, say, day one of your residency training, perhaps, but over the course of a few months. That's really meant to allow this idea of understanding and knowing how a clinical note can be most effectively or impactfully structured as an individual, before you're relying on a non-human tool to be doing it for you.”

Nambudiri stressed what he describes as a tremendous opportunity to learn using such tools. Similar to Block, Nambudiri noted the potential value in dermatology training of using certain tools to augment rather than replace work.

“Actually doing a note yourself, if you have the AI recording also available in parallel, and seeing what would come out differently [is helpful], or using the AI tool to generate a draft of a note, and then you, as a physician, can go in and enter it,” Nambudiri said. “It’s actually a teachable moment for us as faculty to be thinking through what is actually critical, and noting what we probably need to be doing more of as physicians.”

In her session at SDPA, Block highlighted the importance of patient safety with regard to AI tools and their use in the medical field. Nevertheless, Block also expressed her belief in the potential for AI to augment work done by clinicians in a manner that leads to positive outcomes.

“...We're taking the human aspect of it and putting it into machine learning,” Block said in her interview. “We're marrying those two things together to give us an objective view of what we're trying to do. So, it's not making that decision for us; it's augmenting our decision. What I want people to walk away with when they hear my lecture is really getting a foundational understanding of how AI can be integrated into our workflow and how they could utilize it to provide the best patient care.”

Addressing some of the more common downsides of AI’s implementation in clinical and educational settings in dermatology, Block highlighted the importance of clinicians understanding regulatory processes and ensuring they are relying on a tool that is US Food and Drug Administration (FDA)-approved, has gone through rigorous research, and has the necessary guardrails up for patient safety.

Gracey Maniar also tempered her statements on concerns over AI with several positive views related to these technologies and their application in dermatology education.

“I also think [we’re] seeing efficiencies help with day-to-day doldrums and clinic operations that can weigh us down,” Gracey Maniar said. “I think having AI help generate patient education materials, since a lot of this should be geared somewhere between a fourth-grade level of education, that's something that we are not necessarily great at as physicians. I think that is something that could help move that along as well, too.”

However, Gracey Maniar noted the unknowns in this area, highlighting the lack of longer-term data and the lack of understanding over ways diagnostic apps can be helpful or hurtful.

“I'm a little cautiously pessimistic, I would say, as far as scholarly work goes,” Gracey Maniar said. “I'd really like to see our trainees using their own creativity and see the problems that we face day-to-day in clinic, and be able to generate research questions and research projects on their own to address the problems that our field faces. It's so easy to plug that into ChatGPT and say, ‘What should we be studying?’ or, ‘Write an abstract for me.’ I would love to still see that creativity and intentional thought work go into those types of areas by our trainees.”

AI and machine learning models have undoubtedly shifted the paradigm dramatically in the medical field, and dermatology is no exception. As technologies evolve and new methods of problem-solving are developed, the potential for growth and for diminished creativity both remain at the center of the world’s attention.

References:

1. Hirani R, Noruzi K, Etienne M, et al. Artificial Intelligence and Healthcare: A Journey through History, Present Innovations, and Future Possibilities. Life (Basel). 2024 Apr 26;14(5):557. doi: 10.3390/life14050557. PMID: 38792579; PMCID: PMC11122160.

2. Balch D. Artificial Intelligence: The Rise of ChatGPT and Its Implications. Faculty Focus. August 25, 2023. Accessed December 5, 2025. https://www.facultyfocus.com/articles/teaching-with-technology-articles/artificial-intelligence-the-rise-of-chatgpt-and-its-implications/.

3. Stein, JP., Messingschlager, T., Gnambs, T. et al. Attitudes towards AI: measurement and associations with personality. Sci Rep 14, 2909 (2024). https://doi.org/10.1038/s41598-024-53335-2.

4. Kennedy B, Yam E, Fuentes J, et al. How Americans View AI and Its Impact on People and Society. Pew Research Center. September 17, 2025. Accessed December 5, 2025. https://www.pewresearch.org/science/2025/09/17/how-americans-view-ai-and-its-impact-on-people-and-society/

5. Bohr A, Memarzadeh K. The rise of artificial intelligence in healthcare applications. Artificial Intelligence in Healthcare. 2020:25–60. doi: 10.1016/B978-0-12-818438-7.00002-2. Epub 2020 Jun 26. PMCID: PMC7325854.

6. De A, Sarda A, Gupta S, Das S. Use of Artificial Intelligence in Dermatology. Indian J Dermatol. 2020 Sep-Oct;65(5):352-357. doi: 10.4103/ijd.IJD_418_20. PMID: 33165383; PMCID: PMC7640800.

7. Block R. How to Better Use AI and Machine Learning in Dermatology, With Renata Block, MMS, PA-C. HCPLive. November 8, 2025. Accessed December 5, 2025. https://www.hcplive.com/view/how-to-better-use-ai-machine-learning-dermatology-renata-block-mms-pa-c.

8. Sengupta D. Artificial Intelligence in Diagnostic Dermatology: Challenges and the Way Forward. Indian Dermatol Online J. 2023 Oct 17;14(6):782-787. doi: 10.4103/idoj.idoj_462_23. PMID: 38099026; PMCID: PMC10718130.

9. Barata C, Rotemberg V, Kittler H, et al. A reinforcement learning model for AI-based decision support in skin cancer. Nat Med. 2023 Aug;29(8):1941-1946. doi: 10.1038/s41591-023-02475-5. Epub 2023 Jul 27. PMID: 37501017; PMCID: PMC10427421.

10. Vodrahalli K, Ko J, Daneshjou R, et al. Development and Clinical Evaluation of an Artificial Intelligence Support Tool for Improving Telemedicine Photo Quality. JAMA Dermatol. 2023 May 1;159(5):496-503. doi: 10.1001/jamadermatol.2023.0091. PMID: 36920380; PMCID: PMC10018405.

11. Liu Y, Primiero CA, Betz-Stablein B, et al. Artificial Intelligence for the Classification of Pigmented Skin Lesions in Populations with Skin of Color: A Systematic Review. Dermatology. 2023;239(4):499-513. doi: 10.1159/000530225. Epub 2023 Mar 21. PMID: 36944317; PMCID: PMC10407827.

12. Polesie S, Gillstedt M, Kittler H, et al. Attitudes towards artificial intelligence within dermatology: an international online survey. Br J Dermatol. 2020;183(1):159-161. doi:10.1111/bjd.18875.

13. Gracey LE, Nambudiri VE, Maderal AD, et al. Challenges and Opportunities for AI in Dermatology Residency. JAMA Dermatol. 2025;161(11):1103–1104. doi:10.1001/jamadermatol.2025.3366.

14. Gholizadeh, N., Rokni, G.R. and Babaei, M. (2024), Advantages and Disadvantages of Using AI in Dermatology. Dermatological Reviews, 5: e248. https://doi.org/10.1002/der2.248.

15. Fliorent R, Fardman B, Heath C, et al. Artificial intelligence in dermatology: advancements and challenges in skin of color. Int J Dermatol. 2024 Apr;63(4):455-461. doi: 10.1111/ijd.17076. Epub 2024 Mar 6. PMID: 38444331.

16. Hatem R, Simmons B, Thornton JE. A Call to Address AI "Hallucinations" and How Healthcare Professionals Can Mitigate Their Risks. Cureus. 2023 Sep 5;15(9):e44720. doi: 10.7759/cureus.44720. PMID: 37809168; PMCID: PMC10552880.