Study: ChatGPT's cancer treatment plan is full of misinformation
CTOnews.com, August 27 (Xinhua)-- OpenAI's chat robot ChatGPT is popular all over the world, but a new study shows that it is far from replacing human experts in a key area. Researchers at Brigham Women's Hospital affiliated to Harvard Medical School in the United States have found that the cancer treatment regimen generated by ChatGPT is full of mistakes.
In the study, published in the Journal of Oncology of the American Medical Association, the researchers presented ChatGPT with treatments for a variety of cancer cases and found that 1/3 of the responses contained incorrect information. The study also points out that ChatGPT tends to mix right and wrong information, so it is difficult to identify which information is accurate.
Dr Danielle Bitterman, co-author of the study, said they were "shocked by the extent to which misinformation is mixed with correct information, which makes it difficult for even experts to detect errors." She added, "large language models are trained to provide very persuasive answers, but they are not designed to provide accurate medical advice. Error rates and instability of answers are key safety issues that need to be addressed in the clinical field."
ChatGPT became famous overnight after its launch in November 2022 and reached 100 million active users two months later. Despite the success of ChatGPT, the prototype artificial intelligence model is still prone to "hallucinations", that is, confidently giving misleading or completely wrong information.
CTOnews.com notes that efforts to integrate artificial intelligence into health care are already under way, mainly to simplify management tasks. Earlier this month, a major study found that using artificial intelligence to screen for breast cancer is safe and could cut the workload of radiologists by nearly half. A computer scientist at Harvard University recently found that the latest version of the model GPT-4 can easily pass the US medical license exam, suggesting that it has better clinical judgment than some doctors.
However, due to accuracy problems with bioformative models such as ChatGPT, they are unlikely to replace doctors in the near future. A study in the Journal of Oncology of the American Medical Association found that 12.5% of ChatGPT's answers were "hallucinations" and were most likely to give the wrong information when asked about local treatment or immunotherapy for advanced diseases.
OpenAI has acknowledged that ChatGPT may be unreliable, and the company's terms of use warn that its model is not intended to provide medical information and should not be used to "provide diagnostic or treatment services for serious medical conditions."
