AI in cancer diagnosis: Enhancing accuracy and training the next generation of doctors

AI’s deployment in medical diagnosis often takes two distinct forms: as a tool that assists professionals in real-time decision-making and as a tutor that aids in training and skill development. The study’s experimental design explored the effectiveness of these roles by dividing participants into groups that received AI assistance during training, practice, or both.

Artificial Intelligence (AI) has transformed medical diagnostics, particularly in high-stakes areas such as cancer detection. As AI systems become increasingly integrated into clinical workflows, understanding how they complement human expertise is crucial. A recent study titled "Tool or Tutor? Experimental Evidence from AI Deployment in Cancer Diagnosis" by Vivianna Fang He (UCL), Sihan Li, and Phanish Puranam (INSEAD) explores the dual role of AI in medical education and practice.

The study, conducted as a field experiment with 334 medical students specializing in lung cancer diagnosis, examines AI’s impact when deployed both as a real-time diagnostic assistant and as a training aid. The findings highlight AI’s potential to enhance both immediate diagnostic accuracy and long-term professional learning, offering valuable insights for the future of AI-human collaboration in healthcare.

The dual role of AI: Tool and tutor

AI’s deployment in medical diagnosis often takes two distinct forms: as a tool that assists professionals in real-time decision-making and as a tutor that aids in training and skill development. The study’s experimental design explored the effectiveness of these roles by dividing participants into groups that received AI assistance during training, practice, or both. The results showed that AI assistance improved diagnostic accuracy in both training and practice settings, but the most significant performance gains were observed when AI was used in both capacities simultaneously.

Participants who received AI-assisted training developed a better understanding of cancer diagnostics, improving their ability to identify malignancies even when AI assistance was removed. Those who only used AI during real-time diagnosis without prior AI-integrated training benefitted from immediate decision-making support but did not show lasting improvements in diagnostic ability. This suggests that AI, when used as both a tool and a tutor, enhances both short-term accuracy and long-term expertise.

Impact of AI-assisted training on diagnostic performance

Medical students participating in the study were evaluated on their ability to diagnose lung cancer from high-resolution CT scans. AI-generated insights, such as nodule annotations, malignancy probability, and quantified diagnostic metrics, were provided to some groups during training and to others during practice. The study found that students trained with AI-enhanced case studies performed significantly better in subsequent diagnostic tasks, even when AI assistance was removed. This indicates that AI-based training enhances cognitive learning by reinforcing decision-making patterns that align with expert reasoning.

Conversely, students who relied on AI exclusively during diagnosis without AI-integrated training showed improvement only when AI assistance was available. This highlights the risk of over-reliance on AI, where clinicians may struggle to diagnose accurately without AI support. The study underscores the importance of AI-based education in building independent diagnostic skills, ensuring that AI serves as an enhancement rather than a crutch.

Balancing human expertise with AI assistance

While AI significantly improves diagnostic accuracy, the study emphasizes that human expertise remains irreplaceable in medical decision-making. AI assistance provides powerful real-time insights, but human oversight is necessary to interpret ambiguous cases, contextualize patient histories, and exercise clinical judgment. Additionally, AI’s effectiveness depends on user proficiency—those who received AI-integrated training were better equipped to interpret AI outputs, whereas those without prior AI exposure were more prone to misinterpretations.

These findings have significant implications for medical education and AI deployment in clinical practice. Integrating AI-based training into medical curricula can prepare future professionals to collaborate effectively with AI, improving both individual competencies and overall diagnostic accuracy. Moreover, structured AI adoption policies must ensure that AI tools are used to support, rather than replace, human expertise, maintaining a balance between technological advancement and clinical intuition.

Future directions and ethical considerations

The study highlights the need for continued research on optimizing AI integration in medical training and practice. Future studies should explore AI’s impact across different specializations, assess its influence on experienced practitioners, and evaluate its long-term effects on diagnostic accuracy. Additionally, ethical considerations such as AI transparency, potential biases in training data, and regulatory compliance must be addressed to ensure responsible AI deployment in healthcare.

Ultimately, this research demonstrates that AI has the potential to revolutionize cancer diagnosis by serving both as a real-time assistant and as an educational tool. By strategically incorporating AI into both training and practice, the medical field can enhance diagnostic accuracy, improve clinician proficiency, and create a more efficient, technology-supported healthcare system.