Do AI Chatbots Provide Reliable Cancer Information? Yes and No

Cancer causes approximately 10 million deaths worldwide each year. Access to accurate, reliable information about treatment options is critical for patients and healthcare professionals. Artificial intelligence (AI) chatbots are becoming increasingly popular sources of health information. However, the quality of their medical advice has not yet been rigorously evaluated. JAMA Oncology has published two new studies offering important data on this issue. Their findings matter greatly to cancer patients and healthcare professionals.

AI Chatbots Usually Provide Accurate Information

Quality and Sources

The first study examined the most frequently searched cancer-related Google queries and reported that AI chatbots generally provided consumers with accurate information.

The researchers examined:

• ChatGPT-3.5,
• Perplexity,
• Chatsonic, and
• Bing AI,

They analyzed the quality of responses from these four different AI chatbots.

Three of the four cited reliable sources such as the American Cancer Society, Mayo Clinic, and the Centers for Disease Control and Prevention.

The Problem of Complexity

The study also noted that the information's complexity could limit its usefulness. Responses were generally at a college reading level, which may be difficult for the general public to understand.

But Treatment Recommendations Have Shortcomings

Problems with Guideline Alignment

On the other hand, cancer treatment recommendations were evaluated by the second study, which found that AI chatbots often failed to provide advice consistent with national guidelines. They particularly fell short for breast, prostate, and lung cancer.

Researchers created 104 different question prompts and compared the resulting treatment strategies with the 2021 National Comprehensive Cancer Network (NCCN) guidelines.

Misleading Advice

The study found that about a third of the chatbots (34%) had at least one treatment recommendation inconsistent with national guidelines or accepted medical practice. Moreover, about 13% of chatbot treatment recommendations were absent from existing guidelines and considered “unreal” or “hallucinated.” These misleading recommendations often involved local treatment, targeted therapy, or immunotherapy for advanced cancer.

Future Hopes and Recommendations

The Potential of Artificial Intelligence

Despite these limitations, there is optimism about AI's future in medicine. Algorithms can be trained on data from leading medical systems worldwide and updated with the latest guidance, such as NCCN guidelines.

The Need for Careful Monitoring

These algorithms must be carefully monitored as they are integrated into health systems. They can provide incorrect or misleading information, with potentially serious consequences for patients.

Conclusion

Although AI chatbots generally provide accurate cancer information, they appear unreliable for treatment recommendations. This is an important warning for patients and healthcare professionals, and more research and regulation are needed for AI to realize its full medical potential.