Causal reasoning

But that's, like, usually where, where it stops, and I was curious to learn, okay then, if correlation does not imply causation, what does? And so I got to read more into discovery cause inference literature.

I aimed to extend the current AI systems with this capability of doing causal reasoning, which for humans is very intuitive.

But it turns out it is incredibly hard to train AI systems to do that because they basically are trained on correlations, so they can observe that there are relationships between variables, but they're unable to distinguish the direction of that relationship.

And the methods that we have that are for, like, this type of causal reasoning, they're quite limited in terms that they make very strong assumptions.

So in statistics, well, math, you have the assumption that your data is identically and independent distributed.

But in the real world, this is not always the case, or we know that this assumption is hard.