Did ChatGPT Really Cure a Dog's Cancer? It's Complicated

A viral story claiming that ChatGPT helped cure a dog's cancer has been widely shared, amplified by figures like OpenAI co-founder Greg Brockman. The narrative centers on Rosie, a seven-year-old Shar Pei, who received an experimental personalized mRNA cancer vaccine. While the treatment is real, the extent of AI's role in its development has been questioned, with human researchers emphasizing their critical contributions.

The Dog's Cancer Journey and AI Assistance

Paul Conyngham, Rosie's owner and an Australian AI consultant, documented his dog's battle with late-stage cancer after vets gave her only months to live. Conyngham stated he used consumer AI tools, starting with ChatGPT, to devise a plan. ChatGPT reportedly suggested genomic sequencing and pointed him towards specific institutions and equipment, including an Illumina machine and researchers like Dr. Martin Smith. Conyngham followed these leads, connecting with the Ramaciotti Centre for Genomics at UNSW, where Rosie's genome was sequenced for approximately $3,000. The resulting 320 gigabytes of data represented Rosie's unique genetic makeup.

Developing the Treatment: Human Expertise at the Forefront

Conyngham focused on the c-KIT protein, a known factor in canine mast cell tumors. He used Google's AlphaFold to model Rosie's c-KIT protein, comparing it to a healthy baseline and identifying mutations. He then searched for existing compounds that could target these proteins, identifying a human cancer drug. However, the critical steps of assembling the mRNA vaccine were performed by human experts. Professor Palli Thordarson, Director of the UNSW RNA Institute, confirmed his team was involved in making the mRNA-LNP for Rosie. Dr. Martin Smith, head of the Ramaciotti Centre, performed the DNA/RNA sequencing. Notably, Conyngham revealed that the final vaccine construct was designed by Grok, with Gemini also contributing significantly, not solely ChatGPT.

Clarifying AI's Actual Role and Limitations

While AI tools like ChatGPT can assist in sifting through scientific literature and identifying potential researchers or equipment, they did not design the cancer therapy. AlphaFold, a protein structure prediction model, provided a protein model with a confidence score of 54.55, which experts noted as low and requiring significant lab validation. Dr. Smith confirmed AlphaFold was not used for the mRNA vaccine design. Professor Thordarson cautioned that the treatment may not have cured Rosie entirely, as some tumors did not respond, and further investigation into tumor mutations is ongoing. The vaccine also required co-administration with a checkpoint inhibitor, a common necessity for personalized cancer vaccines.

Lessons from AI in Medicine

The Rosie case highlights the potential for AI to aid in research but underscores the difference between assistance and direct development. Unlike past AI healthcare initiatives, such as IBM's Watson for Oncology which faced criticism for unsafe recommendations, the Rosie case involved established science and credentialed researchers. However, the narrative risks overshadowing the crucial work of scientists and institutions by overstating AI's capabilities. The episode serves as a reminder that while AI can navigate research literature, it does not replace the expertise and infrastructure required for medical treatment design and production.