What Are the Barriers to Widespread Use of Artificial Intelligence in Oncology?

Why Does This Matter?

Artificial intelligence (AI) has emerged in recent years as a promising technology in many areas of oncology, including diagnosis, treatment planning, and patient monitoring. Despite the excitement, however, serious problems remain in integrating AI into clinical practice. This article examines the main factors limiting wider adoption of AI in oncology, including insufficient accuracy, limited data diversity, ethical and environmental concerns, regulatory gaps, and more.

1. How Accurate Is Artificial Intelligence? Hallucinations and Misleading Information

One of oncologists' biggest concerns is the high “hallucination rate” of AI systems, meaning that they generate information that is incorrect or not based on facts.

Important Note: In some studies, the rate at which AI chatbots produce incorrect information has exceeded 20-30%. This poses a serious risk, particularly in oncology, where treatment protocols change rapidly.

This is particularly dangerous in oncology:

AI tools may misrepresent the literature or cite papers that have not undergone peer review. Without proper verification, this can lead to suboptimal clinical recommendations.

2. Bias in Datasets: Not Suitable for Everyone

AI systems are only as good as the data on which they are trained. However, these datasets predominantly represent White individuals from higher socioeconomic groups. This can cause systems to fail in patients from different racial, ethnic, and geographical backgrounds.

A Fact Worth Emphasizing: AI systems not trained on minority populations may have substantially lower diagnostic accuracy in these groups, potentially deepening health inequalities.

3. Does It Fit Clinical Practice? The Challenge of Integration Into Workflows

Many clinical settings are not ready to integrate AI systems.

Clinical environments are busy and complex. Most clinics lack sufficient technical infrastructure to implement these systems.

Many AI models also behave like “black boxes,” offering no explanation of the reasoning behind their decisions. This makes it difficult for physicians to explain those decisions to patients.

4. Patient Privacy and Regulatory Gaps

Protecting patient privacy is particularly important in cancer care because the data include genetic information. Yet many AI tools require extensive sharing of patient data.

It also remains unclear who bears legal responsibility for adverse outcomes arising from AI-supported decisions. This makes many healthcare institutions cautious about AI.

5. Environmental and Technical Costs: Hidden Burdens

AI systems run in massive data centers that consume millions of liters of water annually, energy, and scarce elements such as lithium and cobalt. This increases their carbon footprint.

Striking Data: The annual carbon emissions of large AI models can approach those of small cities.

Because these systems require continuous updates, they also increase the workload of hospital IT departments. Institutions with limited resources may be unable to sustain them.

6. How Reliable Is AI From the Patient's Perspective?

Patients using AI-supported information tools also risk being misled. Data presented without context, such as average survival rates, may be misleading for an individual patient.

For patients with low health literacy, AI tools may create confusion and lead to inappropriate treatment decisions.

7. Proposed Solutions: More Ethical, Safer, and More Accessible AI

Despite these problems, how can AI be used safely and effectively in oncology?

Clinical Validation Systems

AI recommendations should be reviewed by oncologists in real time.

Continuously updated validation systems should be implemented.

Educational Programs

AI literacy should become part of medical school curricula and continuing education.

Organizations such as ASCO should develop oncology-specific AI training modules.

Multidisciplinary Teams

AI systems should be developed not only by engineers but also in collaboration with oncologists, ethicists, and patients.

Data Security and Transparency

The composition of AI training datasets and details of system performance should be shared transparently with physicians.

Environmental Impact Assessment

Environmental impact analyses should be conducted for each new AI system, and their findings should inform health policies.

Conclusion: AI Holds Promise in Oncology, but Barriers Must Be Overcome

AI systems should simplify oncology practice rather than make it more complicated. They should use validated data, be energy-efficient, and undergo continuous updates. Above all, they should remain centered on people.

Artificial intelligence could revolutionize oncology, but achieving that potential requires a foundation of accuracy, fairness, transparency, and sustainability.