levier17, on 2012-April-18, 06:50, said:
Can't there be a commission or something that we can write our complaints and they can try to advance GIB.
Are you asking for something different from the GIB Robot Discussion forum?
Anyway, nothing about fixing GIB is easy. It's a complicated program, written by an AI expert, but he's no longer involved. None of us at BBO have real AI talent (I took one AI class 30 years ago). We can tweak its bidding rules, and fix obvious bugs, but we're not really up to the task of redesigning its algorithms.
1. It knows, but sometimes it does simulations that make it think that leaving the double in will work out better. Since it's hard to write rules that tell when to convert takeout doubles to penalty, we depend on these simulations.
2. Watch when GIB is defending against your NT games, and notice that it often makes passive leads against NT, rather than attacking leads from its strong suit. In that case, returning the suit is usually wrong -- it needs to switch to capture declarer's honors. This style actually seems to work out pretty well for it.
Anyway, GIB's defense isn't based on rules, it's based on simulations. So it's hard for us to make it do anything specific, like return partner's suit.
3. When it jumps to 5 of trumps, it's basically saying that it has weak trumps, but a good hand otherwise -- bid slam if you have a good trump suit.