Physical sciences are primarily experimental endeavors. Theorists are there to provide tools and understanding, usually in retrospect, and their work is secondary to the people who actually work in the lab trying things out.
I say this as a *theoretical* physicist
So you're left with this cartoon of the AI... numerically simulating experiments? Which is nonsense, at least when it comes to quantum mechanics.
These are problems with exponentially large computational spaces, there are physical limits on computation that prevent you from numerically calculating the answers you need.
@yarrriv Yeah, it's nonsense. Even specialists of Machine Learning are aware of the danger of trying to extrapolate too much from limited data, it's called overfitting.
I know EY previously claimed that a superintelligent should be able to derive General Relativity from watching an apple fall from a tree so he's certainly not afraid of overfitting.
@yarrriv There was an xkcd which went something like this
Good: We realised our data was bad, so we threw it all away and started over.
Bad: We realised our data was bad, so we included more variables in our statistical analysis to reach more nuanced conclusions.
Very bad: Our data was bad, so we decided to correct it using AI techniques...
yarrriv
Diego Beghin
RedFishBlueFish