Artificial intelligence in melanoma diagnosis: Three scenarios, shifts in competencies, need for regulation, and reconciling dissent between humans and AI

Abstract

Tools based on machine learning (so-called artificial intelligence, AI) are increasingly being developed to diagnose malignant melanoma in dermatology. This contribution discusses (1) three scenarios for the use of AI in different medical settings, (2) shifts in competencies from dermatologists to non-specialists and empowered patients, (3) regulatory frameworks to ensure safety and effectiveness and their consequences for AI tools, and (4) cognitive dissonance and potential delegation of human decision-making to AI. We conclude that AI systems should not replace human medical expertise but play a supporting role. We identify needs for regulation and provide recommendations for action to help all (human) actors navigate safely through the choppy waters of this emerging market. Potential dilemmas arise when AI tools provide diagnoses that conflict with human medical expertise. Reconciling these conflicts will be a major challenge.

Keywords

• melanoma
• diagnosis
• artificial intelligence
• patient-doctor relationship
• diagnostic accuracy