Might as well clarify how GIB works, as it is interesting:
What GIB (Goren in a box) did was to make each decision was do a monte carlo simulation of the possible deals, then solve them in double dummy fashion. That's as if everyone could see all the cards.
Worked quite well, but did introduce some downright odd elements into its play. In particular it is very bad at making people misguess when its defending (it assumes you'll guess right all the time!).
What I presume a fully modernised approach to it would do to proper opponent modelling. There wasn't either the data or the computational firepower to do that with GIB, but nowadays there's both in abundance. Bridge base online has every major event in its vugraph data, and this neural net stuff is great at this sort of thing.
You could even micro optimise it to play better vs specific opponents
(Either classes of them, or individuals!).
In principle, I think computers should be tremendously good at bridge. So much of the game is about retaining concentration for deal after deal after deal, and they're obviously entirely superhuman at that!