I think it's because we learn biology with a textbook cartoon illustration of the cell with sparse organelles floating in a clear sea of cytoplasm. The packed reality looks more like the halftime bathroom line at the Rose Bowl. Tons of heterogenous proteins and metabolites and mRNA and everything seemingly diffusing to its destination, shouldering past everything else. The sheer combinatorial complexity of the number of neighbors each molecule is imparting force on, combined with the constantly shifting conformational changes of many proteins which change the forces they receive, it's overwhelming. Recall that only recently AlphaFold was able to decently solve protein structures in isolation - add in dozens of shifting neighbors and that's suddenly a dynamic problem.

I am not sure where whole-cell simulation is at the moment since I've been away from the field for about 15 years, but I recall a rather difficult multi-month simulation that was trying to model an "empty" volume of cytoplasm away from the organelles, about 1/50th of total cell volume, and with all proteins and metabolites replaced by hard spheres of varying "stickiness". It was considered a huge success to just get a few of the diffusion rates for various compounds in the right order of magnitude. I mean, if you really want to get the fleeting interactions right you need to be modeling individual water molecules. I know there have been large advances in computing in the intervening years, but this was on I think #20 in the TOP500 at the time. Unlike AlphaGo I don't see any immediate avenue for AI to help with this because unlike protein crystal structures there is no wealth of quality training data for cellular dynamics at the molecular level.