高中英语每日时文欣赏00002 AI Health Tools

Recently, several tech companies have launched AI health assistants for the public. Microsoft released Copilot Health, where users can link their health records and ask questions. Amazon also opened its Health AI tool to everyone. Before that, OpenAI had already released ChatGPT Health, and Anthropic’s Claude can also access health information with permission. This shows that AI for public health is becoming a trend.

The demand for such tools is clear. The current medical system cannot provide timely health advice to everyone. Research shows that today’s AI models can give safe and useful suggestions. However, experts stress that these tools need strict review by independent researchers before being widely released.

In a high-risk field like health, letting companies test their own products may not be wise, as they could miss important problems. Outside experts can help find these issues. Researchers say that while AI models may be ready for use, the “evidence must be strong enough” to support it.

Developers believe now is the time to release these products because the technology is advanced enough to give good health advice. Another key reason is high user demand. Data shows that people are already using AI tools in large numbers for health questions. This reflects the difficulty many face in getting medical help.

In a perfect case, health AI could help users decide if they need to see a doctor, which may reduce pressure on hospitals. But a recent study raised concerns: ChatGPT Health sometimes suggested too much care for minor problems but missed real emergencies. This shows why outside testing is so important.

Although these products often include warnings saying they are not for diagnosis or treatment, users may ignore them and still ask for health plans.

Companies say they test the chatbots to make sure they are mostly safe. OpenAI created a test called HealthBench to measure how well models perform. Still, such tests have limits. One study found that even if a model can correctly identify an illness, users without medical knowledge might only get it right one-third of the time when using the AI, because they may not know how to ask the right questions or understand the answers.

Experts say the best evaluation would involve real people testing the tools before release. Google recently did such a study. Its medical chatbot AMIE did well, but Google is not releasing it yet, citing safety and fairness concerns.

The real key is “third-party” review. No matter how much a company tests its own product, it’s hard to fully trust the results. Third-party reviews are more fair, and if many different groups are involved, they can cover more ground.

Researchers are working to build better testing systems, though it takes time and money. No one expects health AI to be perfect before release—even doctors make mistakes. For people with little access to medical care, an AI that is always available, even if it sometimes makes small errors, could still be better than nothing, as long as the errors are not serious. But with the information we have now, it’s still unclear whether today’s public health AI tools truly do more good than harm.

1. 阅读理解题