The Promise and Peril of AI Consultation

I saw a story today about AI nurses doing consultations. There’s a lot to like about this, and a lot to dislike.

In the like column, an info-rich AI nurse could be a big improvement. Consider how absolutely terrible almost every single human medical practitioner is. It has to be one of the worst professions when it comes to knowing your subject matter. Health pros are almost always myopically narrow, and uninterested and uncurious about their own field. They run through a set of tests taught to them at some medical school based on biases and bad science, and then tell you to take a pill that undoubtedly makes your overall health worse.

When you’ve spent lots of hours researching health issues on your own, and connecting with others doing the same, you find tons of new ideas and insights and cures. But the biggest problem is that there it just too much information, too many similar but not exact symptoms, too many remedies to test, too many disparate places this info exists, and much of it in other languages. Imagine an AI nurse able to absorb and sift and sort and pattern match to make recommendations in an unbiased way. “Scanning hundreds of thousands of forums and papers and websites, it seems these are three likely candidates for what you have, and these appear to be the top three treatments with the highest correlation with success.” It could be massive.

In the dislike column, the parameters on any AI system are put there by humans. And so far, these parameters have led to fairly ridiculous results on any matters of disagreement or dissension. Imagine the kind of stupid, tyranny-approved, unscientific answers a mainstream medical AI bot would be programmed to give you. “Take your prescriptions” is probably the gist of it.

Then again, I don’t know if it could be worse than the status quo.