NATO summit in Madrid gives you clues on AI models.

These days it is unlikely that your wallet will be stolen in Madrid.

Basically because there is a policeman on every corner.

Even for people like me, who vowed years ago not to watch the news, it’s hard to ignore that there is a NATO mega-summit whose goal is to scare the East.


Or rather to tell it that we are not scared.

Which is not true.

Because we are.


Well, the fact is that this is the way the world works.

The balance between the national and the global.


We are all aware that a single world state in which there were no regional singularities would be gray, dictatorial and inhuman.

And at the same time we have all seen that alliances, union and the abolition of borders make us better.

Balance is tricky.


It’s the same with clinical predictive algorithms and geography.

They have to work at a local level, with a specific population.

And at the same time they have to be usable anywhere in the world.


Look, a couple of years ago, in the middle of COVID, a team from China published in Nature a model made with machine learning that was very good at guessing which cases were going to become severe.
 

And after using it, three groups replicated, from Amsterdam, Paris and New York, saying that they had tried it and it didn’t work for their populations.

So the Chinese team replicated again saying that they had RE-ENGINEERED the model with the data from these places and now it did work with those Western populations.

– There you have it. Try it out and see what nice Areas Under the Curve it is!


Well.

We can call this whatever you want.

Transportability. Calibration. Transfer learning. External validation.

Every time someone rejects a name I invent another one.

Or I take it from the BMJ. Or from JAMA. It’s all there.


If you want to do a predictive algorithm with one population, then you need to transport it to another geography. 

And that has a very precise methodology and technique. With its N and its metrics. It’s not difficult, but you have to be exact and follow the steps well. And if you do it, then it is valid for any place in the world. If you want to see how it is, we can continue here.

Want to use it?:

Start with your proposed AI + RWE use case:

This is the first step for AI + RWE: