Artificial intelligence as a copilot, not autopilot: A framework for responsible integration in pediatric hematology-oncology

Abstract

Artificial intelligence (AI) is transforming pediatric hematology-oncology. Current applications include pattern recognition in blood and marrow images, inference of tumor biology, relapse risk prediction, integration of complex clinical data, and support for writing, learning, and communication. Although often framed as a single technological revolution, these tools differ in purpose, evidentiary basis, and risk profile, yet they share a central consequence: the redistribution of cognitive work and clinical authority. Adoption should follow a “copilot, not autopilot” model in which AI extends perception and expertise while preserving accountable clinical judgment, rigorous formative reasoning in trainees, confidentiality, and equitable access to validated benefits. This demands pediatric- and setting-specific validation, transparent disclosure of AI use, systematic source verification, post-deployment monitoring, and graduated trust proportional to clinical stakes. AI may widen access to knowledge and reduce workload, but its value depends on disciplined verification and the clinician’s authority to override or decline its use.