To keep up with the current levels of publication, 200,000 humans would need read full-time every day 1 . This is to say nothing of them actually internalizing or understanding the information nor doing anything with it.
This seems to me to be an entirely unworkable situation. Even if you could have domain experts who fully understood only their domain - and kept up with it constantly - you would still have the problem of cross-domain information, solutions and meta-topics. Take a meta-domain Q that combines all (or elements from) domains X and Y. The experts in Q would need to be experts (and up-to-date) in both X and Y, or else they would need to selectively choose bits of X and Y to nominate for inclusion in Q. If the Q experts performed that selection themselves they would need to be expert in at least one of the subfields. Or they would have to rely on the X and Y experts to distill and digest their fields for inclusion in Q which itself is a complex task that requires either they know a lot about Q or that the Q-experts and X-experts devise a shared border area (call it J) where they can interface with each other and determine how best to select for X’s integration into Q. It’s unclear to me if this J interface also requires separate expertise!
The above problem is compounded exponentially if combining more fields into a Q-like meta structure and even more so if combining many Q-like structures into a QQ meta meta structure.
Assuming that detail is necessarily lost in the conglomeration, areas of interest that require QQ fields in order to define problem sets and solution spaces will be lacking those details. the problem formulations and possible solutions won’t have those details available to them and will thus be constrained - perhaps fatally (that is, the proper problem definitions may not arise, and/or their solutions may not be possible).
For simple systems this may not be an issue. But for complex problems (defined as those that involve many different fields) this seems like an insurmountable hurdle.
Also, as a simple system grows (even if it doesn’t involve many fields) this issue resurfaces (Complex information needs to be libraried in order to be useful sans a lifetime of study.)