Instead of AI, Try Poison Control

In Japan, an older man, unwilling to wait for botanical advice, grilled up some mushrooms he’d foraged after AI told him they were edible. The vomiting started within 30 minutes.

In Washington state, a 60-year-old man elected to remove table salt from his diet with the help of ChatGPT. He was later admitted to the hospital with bromide poisoning after he was told that replacing sodium chloride (table salt) with sodium bromide was a good choice.

In California, a 19 year old died after combining kratom and Xanax, a combination that ChatGPT assured him would “smooth out” his high and help him feel less nauseated. In reality, it slowed down his breathing, starving his brain of oxygen.

The problem of people counting on large language models for medical advice is widespread, crossing geographic regions and age groups.

Chris Holstege, MD, is the director for the UVA Health’s Blue Ridge Poison Center. “We do know that the internet and AI can give inaccurate, and at times dangerous, information depending on the sources pulled,” Holstege said.

Free Help Is Already Available

Since opening in 1978, the poison center has helped more than 1 million callers. More than 90% of the time, they’re able to help people get the help they need at home, saving callers from traveling to a healthcare professional or having to pay for a hospital visit.

Unlike AI-tools, the poison center is staffed by trained medical providers, who hold themselves to high standards in making sure the guidance they give callers is accurate.

Illegal Drug Concerns

One of the reasons people cite for using AI instead of calling the poison center is the perception of anonymity. But that really is just perception.

While many AI chatbots store conversation logs and send that data back to their parent corporation, the Blue Ridge Poison Control is confidential.  They do not want illegal substance use to be a deterrent for calling. The important thing is getting you the help you need.

Actually Personal Advice

One of the dangers of AI, and why people listen to it when they might ignore an article, is that it feels like it’s speaking directly to them. This makes them think the advice is personal. It even changes its language preferences to feel like a peer.

But poison centers can provide medical advice based on you. That can feel invasive, especially when you’ve been told it’s confidential, but sharing your age, weight, and other medical conditions are crucial to providing accurate medical advice. And it should probably be concerning that AI doesn’t request that info before responding.

What’s Poison Depends on You

Like a lot of adult women, I take an iron supplement. My doctor has recommended it after looking at my blood tests, and I know I feel better when I take it. For me, it’s safe.

If my 3-year-old were to take it, that’s a whole different story.

One of the most common causes of pediatric poisonings is taking a medication intended for, and safe for, an adult in the household. While some AI models are good about asking if you’re asking for a child or an adult, it’s not consistent or reliable.

For Toxic Substances, It’s Good to Be Sure

So, when it comes to who you should call when you’re worried about poisoning, hopefully the answer is clear. But just in case, here’s a table that may help.

Many people are finding instances where AI has been useful for healthcare, and many of our researchers at UVA Health use big data models and machine learning to accelerate lifesaving discoveries.

But when it comes to the possible ingestion of toxic substances, we’ve already got a great, free resource that can offer medically accurate information confidentially.