We cannot perfectly replicate how real life worms behave because we don't have the necessary training data. We can create AI worms based on fake behavioral data, of course, but that will not result in prefect AI replicas of real world worms. We do have untold terabytes of the written word and that's sufficient to bootstrap intelligence with. It turns out that going to space, splitting the atom, and creating general superintelligence is easier than understanding insect behavior or replicating exactly how a bumblebee flies. Our intuitions about the relative difficulty of unsolved scientific questions are pretty much useless. Some seemingly easy things are in fact impossibly difficult and the inverse is true as well.