Anthropic CEO: 'Curing Cancer is Most Effective Way to Secure Public Trust in AI'

Anthropic CEO Dario Amodei stated, "The most certain way to increase public trust in AI is to cure cancer." But expectations are not without their downsides. As more patients ask ChatGPT about symptoms or test results before visiting a doctor and bring those answers to the hospital, the burden on physicians is increasing. There are also concerns that AI may not fully understand the medical context or provide incorrect information.

Amidst this, cases were introduced last year showing the potential of AI in diagnosing rare diseases. With rare diseases, it is difficult for a single doctor to experience enough cases in a lifetime, and symptoms often overlap with other conditions, making it not uncommon for patients to take more than five years to receive a diagnosis.

A prime example is the son of Rachel Hincken, Oliver. Doctors could not find the cause for his delayed language and walking development, as well as his short stature. They explained only that he would improve over time. Last year, when Hincken input her son's photo into 'Face2Gene,' an AI tool for medical professionals, the AI suggested the possibility of TRPS (Trichorhinophalangeal Syndrome), a rare genetic disease. This was confirmed through genetic testing and clinical consultation. When it was revealed that Hincken herself suffered from the same condition, she said, "Oh my god, we finally found the answer."

Simo, who is also a patient with Postural Orthostatic Tachycardia Syndrome (POTS), saw symptoms begin after the birth of her daughter in 2015 but it took nine months to get a diagnosis. She recalled, "I was a typical case, but the doctors were so focused on the difficulties of the pregnancy period that they didn't notice the symptoms." Now, she uses ChatGPT to analyze her entire genome sequencing results and even compiles a list of tests to discuss with her doctor.

Research results back this up. A study published last year in JAMA tested the diagnostic capabilities of AI chatbots on 90 complex rare disease cases that were already diagnosed. The two chatbots showed accuracy of 13% and 10% respectively, while the accuracy of doctors reviewing the patient's medical records was 5.6%. However, as the study used clinical summaries of already diagnosed cases, there is a difference from actual clinical settings.

What researchers and doctors are focusing on is not that AI diagnoses all diseases better than humans, but that it finds connections that are difficult for humans to find through experience. It shows strength in tasks such as connecting medical images, medical papers, and case reports, or comparing patterns of multiple diseases at once. Dr. Xiao Feng, the clinical geneticist who confirmed the Hincken family's diagnosis, said, "AI searches a much wider area much faster than we do, but human work is still needed to synthesize the data."

A case emerged at the Mayo Clinic involving 77-year-old Mike Bush. Visiting the hospital with shortness of breath and chest tightness, the medical team suspected pneumonia or heart failure. However, AI analyzed the electrocardiogram and suggested a 98% possibility of cardiac amyloidosis, which was confirmed through additional imaging tests. Michael Ames, a specialist nurse who had never diagnosed this disease before, was greatly impressed. Bush also said, "If it weren't for AI, I might have been treated for a different disease."

What these cases show is not a simple story that AI is superior to doctors. With common diseases, doctors can accumulate experience, but with rare diseases, it is difficult for individual doctors to secure sufficient clinical experience. It is at this very point that AI can broadly compare cases from papers, case studies, medical images, and test results.

It has been pointed out that the real bottleneck in discussions on the medical application of AI is data. In the case of rare diseases, data itself such as patient records, tissue samples, and medical images is scarce, and existing materials are often not properly digitized. To find patterns, data must be accumulated beforehand. In the U.S., work is underway to digitize old pathology data from hospitals and academic institutions, and Dr. Matthew Hanna is participating in the development of 'ScanVan,' a mobile container for data extraction.

CEO Amodei explained that he means promises alone that AI will be of great help to humanity are insufficient, and it must show results that actually change people's lives. Former CEO Simo agreed with Amodei's outlook, stating that AI will ultimately be able to "cure every disease."