The problem with having computer separation be less than pilot separation is, what happens if the computers fail and the pilots have to take over? Then they're suddenly in a situation where they're already below whatever the minimum safe pilot separation is, and now you've potentially turned a recoverable failure into an unrecoverable one. If the whole point of having pilots is to be backups for the computers, the computer situation has to have more margin than the equivalent pilot one, even if efficiency is being left on the table.

(The only way around this is to not have pilots as backup, and turn the whole thing over to the computers, but we're not ready or willing to take that step yet)

Considering that only one runway (28R, IIRC) at SFO even winks in direction of supporting autoland, and there's no continental USA support for controller-pilot datalink (only used for oceanic flight), at least outside of test flights (correctme if I'm wrong, not an expert here).

On top of that current CPDLC doesn't seem to be as well integrated to autopilot support other than pilots acking each command, and then I do not think Mode S is mandated (or exact enough) to provide good enough separation.