Artificial intelligence (AI) can be a fast way to look up health information. And when you have new symptoms or feel off, it can seem like an easy, private place to start. But for now, research shows there are real risks in using AI for personal symptom or diagnosis questions. Current research shows that public AI tools can miss key details. They can miss emergencies. Or they can sound too urgent. And they can sound very sure even when they're wrong.
Because AI is a fast-moving area of study, recommendations and evidence about its use will likely continue to change. Today, diagnosing and treating illness is too complex for AI to play a major role. It's a good way to learn about your health. But it's also important to know where AI can fall short. For symptom or diagnosis questions, it's best to work with a healthcare professional who can bring the experience, medical knowledge and context needed to spot where AI may be getting things wrong.
When it comes to AI and health, Paul Y. Takahashi, M.D., M.P.H., a physician at Mayo Clinic, says AI is a good educational tool. "I'm always happy when people are trying to educate themselves about what's going on with their health or what they can do to improve it."
Jamie J. Van Gompel, M.D., a neurosurgeon at Mayo Clinic, agrees. "I think AI is a tool that we all should use to see what it does. Sometimes it comes up with some funny stuff. Sometimes it's accurate. Right now, it's the job of the physician to put that all into context for you," he says.
AI can miss important details and give wrong answers
One problem with using AI to check symptoms is that the answer depends a lot on what you ask and what details you give. Depending on that information, the AI tool may not ask the follow-up questions that really matter. More research is needed, but limited research shows that people often may not give AI enough information to reach the right answer.
Healthcare professionals know which questions to ask. They know what details matter most when trying to make a diagnosis. They also may have your medical records, which can give them more context.
Uploading your medical records to an AI tool may seem like a fix, but it isn't that simple. Public AI tools may come with privacy risks. And your medical records may not give the full picture of what you're feeling and what's going on for you right now.
There also are things a healthcare professional can see during an exam that AI can't. "An exam is a big part of a diagnosis," says Dr. Takahashi. "For example: hip pain. If you just emailed or talked to me about hip pain, I wouldn't be able to tell you much. You really need a physical exam to find out if that hip pain is likely from arthritis, bursitis or a fracture."
Another issue is that AI can sound certain even when it is missing key information, weighing the wrong possibilities or applying general information that doesn't apply to you.
AI can miss emergencies or be too urgent
AI can struggle with telling you how serious your symptoms are. Even public AI tools made for health purposes may not correctly tell you how fast you need care. That can be a serious problem.
One study found that AI missed about half of the emergencies it was tested on. AI also can tell people to get care sooner than they need to.
AI tends to do worse early in the diagnosis process. That's when there may not be enough information yet, and several causes may still be possible. AI may jump to one answer too quickly instead of working through the possibilities.
Do not use AI to make decisions about urgent or severe symptoms. It's safer to ask a healthcare professional for advice.
Test results are not the same as a diagnosis
It can be tempting to enter your test results into an AI tool, especially if you get the results before you've had a chance to talk with a healthcare professional. But that can come with risks too.
AI may comment on one number or one report. But that usually isn't enough to make a diagnosis. "AI may assume a worst-case scenario based on limited information, causing you anxiety, when that's not accurate," says Dr. Takahashi.
A healthcare professional looks at test results along with your symptoms, health history, exam and other findings for a more complete picture.
A better way to use AI for symptoms
When it comes to using AI to look into symptoms, consider sticking to more general uses such as:
- Learning general information. Ask questions, such as, "What are common causes of hip pain?"
- Understanding medical terms. Consider questions such as, "What does 'inflammation' mean?" Or ask, "What are tendonitis and bursitis?"
- Preparing for medical appointments. Ask AI to help you make a list of questions before your healthcare visit. For example, "What questions should I ask my doctor about these symptoms?"
- Organize your thoughts. Use AI to help you get organized. For example, you might prompt AI: "Help me make a symptom timeline to bring to my next appointment."
"AI can provide a lot of good general information," says Dr. Takahashi.
Dr. Van Gompel adds, "It also can offer general education about an existing diagnosis and what potential treatment options might be."
But if you're trying to figure out what's wrong, today's AI tools aren't a good substitute for a healthcare professional. An AI tool likely won't have all the details it needs to make a diagnosis. And because of that, it may confidently mislead you or point you toward the wrong diagnosis.
AI systems also are not able to understand your personal goals, values or circumstances the way a health professional can. "A healthcare professional can help guide you in making decisions based on what's most important to you and what works well in your life," says Dr. Van Gompel.
