Eric Topol

And the request is, well, we want two trials, randomized trials, independent.

We want to have p-values that are significant.

And we want to know the biological mechanism, ideally with a dose response of the drug.

But there are many drugs, as you review, that have no biological mechanism established.

And even when the tobacco problems were mounting, the actual mechanism of how tobacco use caused cancer wasn't known.

So how important is the biological mechanism, especially now that we're well into the AI world where explainability is demanded?

And so we don't know the mechanism, but we also don't know the mechanism in lots of things in medicine too, like anesthetics and even things as simple as aspirin, how it works and many others.

So how do we deal with this quest for the biological mechanisms?

Yeah, no, I'm so glad you brought that up because when Demis Hassabis and John Jumper won the Nobel Prize, the point I made was maybe there should be an asterisk with AI because they don't know how it works.

I mean, they had all the rich data from the protein data bank.

And they got the transformer model to do it for 200 million protein structure prediction.

But they still, to this day, don't fully understand how the model really was working.

So it reinforces what you're just saying.

And of course, it cuts across so many types of AI.

It's just that we tend to hold different standards in medicine, not realizing that there's lots of lack of explainability for routine medical treatments today.

Now,

One of the things that I found fascinating in your book, because there's different levels of proof, different types of proof, but solid logical systems.

And on page 60 of the book, especially pertinent to the US right now, there is a bit about Kurt Godel.

And what he did there was he basically, there was a question about,

a dictatorship in the U.S.