I read that AI is being used in hospitals now to help doctors, but it’s still super expensive to manage. Isn’t AI supposed to save money?

You’d think so, but it turns out these systems need a lot of monitoring. For example, algorithms can make mistakes or stop working well if they aren’t checked regularly.

Like what kind of mistakes? I thought AI was supposed to be better than humans at not messing up.

Well, during the pandemic, one AI tool that predicts if cancer patients might die got worse at its job. It missed important chances to help doctors talk to patients about their treatment.

That’s awful! Why did it mess up?

It’s called algorithm decay. Changes in the data, like new hospital procedures or lab providers, can throw the system off.

So, they just fix it and it’s fine, right?

Not so simple. Fixing it takes a lot of people and time. At Stanford, they spent months and hundreds of hours just checking two algorithms.

Wow, that sounds like a full-time job. Couldn’t they just use another AI to watch over the first one?

Funny you say that! Some experts are suggesting AI to monitor other AI, but even then, humans need to keep an eye on everything. And it’s expensive to hire people with the right skills.

What about using AI for smaller tasks, like writing summaries for doctors? That should be easier to manage, right?

You’d think so, but even those tools aren’t perfect. One study found they made a 35% error rate when summarizing patient medical histories. Imagine forgetting a detail like a fever—that could be dangerous!

That’s pretty scary. So, is AI in hospitals a bad idea?

Not necessarily. AI has a lot of potential to help with things like predicting health risks or saving doctors time. But hospitals need better standards and more resources to make sure the tech works properly.

Sounds like they should focus on making sure it’s safe first, even if it costs more upfront.

Exactly. AI can be a great tool, but it’s not magic. It needs to work alongside humans, not replace them. Otherwise, it might cause more problems than it solves.