Help
The GIB robot discussion could be more useful to programmers and players:
- This forum should be the only way for players to submit bug reports.
- Programmers should post comments whenever they tackle a posted problem, especially if they make any progress towards solving it.
Some suggestions:
- In-house development. BBO's own programmers should continue to develop GIB. Out-sourcing is fraught with worse problems .
- Outside Pro-Bots: Even if BBO remains loyal to GIB, however, it could still admit other robots (e.g. WBridge5 and Jack) as paid professionals. BBO could levy a percentage of their fees.
- Computer-Bridge-Championships. Again, if BBO provides a suitable interface, it could host Robot championships, with viewgraph and commentary by Robot-programmers and Bridge-experts. This might interest and benefit all participants.
- Full-disclosure (or Kungsgeten's simplification) should be resuscitated -- to improve GIB's explanations -- and to allow Gib to understand calls made by its partner and opponents.
- GIB heuristics might be improved. For example, as many critics point out, splinters will be ineffective, until GIB appreciates the idea of duplication. e.g. rules-of thumb like the LTC and the 30-34 HCP deck. Another useful heuristic is John McLaren's Ace-extra-test: Before making a non-forcing bid, GIB should check whether it would make the same bid with an ace fewer. If so, then it should consider stronger action.
- Play simulation during the auction, based on information gleaned from the bidding so far, might improve GIB's bidding. e.g. For its partner, GIB could try to construct the weakest hand consistent with the auction, but with well fitting shape and honours. At the other extreme it could consider the strongest hand, consistent with the bidding, but with badly fitting shape and honours. In the former case, it should be keen to explore when simulation shows that game/slam is still likely. In the latter, it should be cautious If game/slam is risky.