The complex ethics of applying ChatGPT and language model artificial intelligence in dermatology

To the Editor: We read with interest Beltrami and Grant-Kels' article, "Consulting ChatGPT: Ethical dilemmas in language model artificial intelligence (AI)," a timely and evolving topic.1Beltrami E.J. Grant-Kels J.M. Consulting ChatGPT: ethical dilemmas in language model artificial intelligence.J Am Acad Dermatol. 2023; ([published online ahead of print, 2023 Mar 10])Abstract Full Text Full Text PDF Scopus (11) Google Scholar This generative language technology may support precise and timely dermatologic diagnoses, but we must weigh ethical considerations. In addition to the ethical principles listed in this article, conversations about AI-generated diagnosis must address bias, informed consent, privacy and data security, and accountability. Regarding bias, the authors rightly point out that AI algorithms are only as good as the data they are trained on; if the data are biased, the algorithm will perpetuate or compound such bias, potentially leading to errors in diagnosis, especially for vulnerable populations.2Daneshjou R. Smith M.P. Sun M.D. Rotemberg V. Zou J. Lack of transparency and potential bias in artificial intelligence data sets and algorithms: a scoping review.JAMA Dermatol. 2021; 157: 1362-1369Crossref PubMed Scopus (70) Google Scholar,3Kamiran F. Calders T. Data preprocessing techniques for classification without discrimination.Knowl Inf Syst. 2012; 33: 1-33Crossref Scopus (588) Google Scholar Ensuring diverse training data sets should be a shared responsibility between AI companies and dermatologists. Dermatologists must collaborate with AI developers to ensure varied and representative data for training AI algorithms. To ensure a balanced dataset, oversampling techniques should be employed for underrepresented populations (eg, oversampling skin of color in melanoma models given lower prevalence).3Kamiran F. Calders T. Data preprocessing techniques for classification without discrimination.Knowl Inf Syst. 2012; 33: 1-33Crossref Scopus (588) Google Scholar However, oversampling must be performed correctly to avoid introducing further bias. In healthcare and other industries, respecting patients' autonomy and preferences regarding the extent of AI use in their care is essential. In some cases, AI may be a core component of a diagnostic process or treatment plan, while in others, it may be utilized for supporting tasks that indirectly influence patient care. While obtaining informed consent for all AI-related applications may not be feasible, disclosing AI use in patient intake can maintain trust in physician-patient relationships without disrupting workflow efficiency. Although avoiding AI altogether in healthcare might be challenging, patients' autonomy and preferences must be respected whenever possible. Open dialogue about pros and cons can also help promote trust. If patients do wish to opt out, this impacts the quality of the AI and its dataset, emphasizing the importance of representative datasets. Privacy and data security are paramount, given the possibility of patient data being used to train algorithms and the risk of data breaches. Patient health information must be securely retained, and policies must proactively anticipate possible violations to respond swiftly if needed. Lastly, accountability refers to who bears the liability for harmful results if AI technology has supported the diagnostic process. Establishing clear lines of responsibility and liability is necessary to protect patients and hold caregivers accountable for errors or malpractice. By setting practice standards and guidelines, we can create a framework that can help shape decisions made by courts and malpractice insurers. Dermatologists, ethicists, and other specialists must collaborate to set clear guidelines and practices. Patient autonomy and privacy must be respected, and stakeholders must be informed of technology limitations. The future remains promising with AI support of clinical care, but we must introduce these tools cautiously to safeguard autonomy, privacy, and safety. In medicine, AI may be best suited as augmented intelligence under human supervision and direction rather than autonomous replacement of physicians – we need not perpetuate a false dichotomy of man vs machine.4Kovarik C. Lee I. Ko J. Ad Hoc task force on augmented intelligence. Commentary: position statement on augmented intelligence (AuI).J Am Acad Dermatol. 2019; 81: 998-1000Abstract Full Text Full Text PDF PubMed Scopus (19) Google Scholar,5Nelson C. Kovarik C. Barbieri J. Forget 'Man vs. Machine.' When Doctors Compete With Artificial Intelligence, Patients Lose. Washington Post, 2018Google Scholar None disclosed. Dermatology in the wake of an AI revolution: Who gets a say?Journal of the American Academy of DermatologyVol. 89Issue 4PreviewTo the Editor: We thank Ferreira and Lipoff for their response, which expands on salient ethical implications of artificial intelligence (AI) generated diagnoses in dermatology. Their insightful discussion raises additional ethical questions regarding patient autonomy, informed consent, and legal accountability as we look to a future in which AI and medicine coalesce. Full-Text PDF