To know the variables required for "intelligence" we need to recreate the universe in which intelligence came to be.
It did take evolution 3.5 billion years to do it after all, so I'm not sure what makes us think we can do it in a fraction of that time.
but there would be a drive to simulate this more efficiently. simulate using a classical model and see if the results match well enough. then the atomic level; then looser and lower-resolution chemical/E&M models, and so on. start compressing the elements being simulated (simplify the environment, prune and rearrange the neurons so long as the thing as a whole still acts the same), and so on. there’s a good chance we can go pretty far before the simplified simulation meaningfully diverges from the true simulation.
why can’t this be flipped? start with a simplified model/simulation, and keep adding details until you reach the point of diminishing returns? it’s a reversal of the same search process as above, but if you subscribe to the first approach (do you?), then is the latter approach truly impossible or just difficult in different ways?