Speaker
Description
The 2025 Gilbert W. Beebe Symposium, convened by the U.S. National Academies’ Nuclear and Radiation Studies Board, examined the expanding role of artificial intelligence (AI) and machine learning (ML) across radiation therapy, medical diagnostics, and occupational health and safety. While technical advances were central, the symposium was framed around a broader epistemological question: how knowledge is generated, validated, and trusted when algorithmic systems increasingly influence radiation-related decision-making.
Across sessions, participants evaluated AI applications in adaptive radiotherapy, quantitative imaging, exposure modeling, digital twins, multimodal integration, and radiation risk assessment. For the international dosimetry community, these discussions are directly aligned with current practice: algorithmic methods are entering workflows for dose reconstruction, individualized estimation, and large-scale exposure modeling. Data readiness emerged as a recurring theme, including metadata standards, harmonization across imaging repositories, and curated datasets to support model generalization. Verification, validation, and uncertainty quantification were emphasized as prerequisites for deploying AI systems when dose estimates inform medical, occupational, or regulatory decisions.
Tensions between performance and interpretability were examined. Algorithmic systems may demonstrate improved predictive capability relative to conventional methods, yet lack mechanistic transparency. Participants considered when AI should advise, extend, or potentially supplant professional judgment, and what levels of uncertainty are acceptable in radiation applications, particularly in scenarios involving clinical decisions, life-course exposure assessment, and policy-relevant risk evaluation.
The symposium underscored that integration of AI into radiation sciences represents a shift in evidentiary standards and professional accountability. For this international dosimetry community, the transition is already underway: algorithmic systems are influencing how exposure is reconstructed, risk inferred, uncertainty communicated, and radiation-related decisions justified across clinical, occupational, environmental, and regulatory domains. Establishing shared expectations for validation, transparency, uncertainty characterization, and governance will determine how these tools are incorporated into professional standards worldwide.