Info Don't use AI (like ChatGPT) for planning a dive

[doctormike]

OK, this is completely random, but the next email that I got after the one with the above response was this one:

 

[lowwall]

Not sure that's true. The advantage of big data approaches and machine learning is that sometimes you come up with associations that would have never happened by human beings looking at numbers and applying standard mathematical techniques.

But we are getting pretty far from my area of expertise, so maybe I'm wrong.

Big data approaches are standard statistical analyses. The difference is that that are applied over a larger set of variables. AI is something different, the only thing they have in common is the requirement to crunch through large amounts of data.

Incidentally, big data throws out the scientific method model of creating hypotheses and then testing them in favor of checking everything and popping up whatever correlations meet a given statistical threshold. In science, we call this "data mining" and it is not a positive. For one thing, it means you cannot know if there is a causative relationship between the two variables. That's why big data is driven by businesses. They don't especially care why the correlation exists. If the cost is low enough, they are happy to take a flyer on whatever tumbles out of the analyses.

Your journal article gives some hints at the problem of using big data approaches in medicine. The first is: What are you actually seeing? Again, correlation is not causation. If you can somehow overcome that, the next question is what can you do with the info? It's certainly not useful on an individual level. At best, there are some public policy implications.

 

[OTF]

Big data approaches are standard statistical analyses. The difference is that that are applied over a larger set of variables. AI is something different, the only thing they have in common is the requirement to crunch through large amounts of data.

Incidentally, big data throws out the scientific method model of creating hypotheses and then testing them in favor of checking everything and popping up whatever correlations meet a given statistical threshold. In science, we call this "data mining" and it is not a positive. For one thing, it means you cannot know if there is a causative relationship between the two variables. That's why big data is driven by businesses. They don't especially care why the correlation exists. If the cost is low enough, they are happy to take a flyer on whatever tumbles out of the analyses.

Your journal article gives some hints at the problem of using big data approaches in medicine. The first is: What are you actually seeing? Again, correlation is not causation. If you can someone overcome that, the next question is what can you do with the info? It's certainly not useful on an individual level. At best, there are some public policy implications.

Well said. There can be value to "data mining" but it has to be approached carefully and scientifically.

One of my very favorite things on the internet ever: Spurious Correlations

 

[Ranga]

Well said. There can be value to "data mining" but it has to be approached carefully and scientifically.

One of my very favorite things on the internet ever: Spurious Correlations

I was expecting that link to be this one.