Last updated: November 2024
Overview
"Ben" is a bridge AI created by Lorand Dali and launched as an open-source project in 2022.
In 2023, Lorand Dali joined the BBO team as a backend engineer, leveraging BBO's comprehensive data archives to refine "Ben." Together with the tech team, Dali utilized multiple machine learning models to evolve "Ben-on-BBO."
The current version of "Ben" in use on BBO features an improved card-play engine, an updated bidding engine, and several new features, such as claims, bidding explanations, and a very fast simulation algorithm.
"Ben on BBO" aims to provide a bridge game experience that feels close to playing with real people, suitable for both casual and serious players.
Ben on BBO Highlights
The version of "Ben" developed for BBO differs from its open-source predecessor. BBO's "Ben" includes significant alterations and innovation, particularly in the card-play engine.
Card play engine
- New card play engine which does not rely on double dummy analysis
- Trained on human play
- Human opening leads
- Uses GIB's claim engine
Bidding engine
- Trained with up-to-date GIB system.
- For bidding explanation it uses GIB.
- Uses more search and simulation.
Efficiency
- Very efficient simulation algorithm makes the play very fast.
- Simulation algorithm allows to do a lot of simulation during bidding (e.g basic GIB doesn't use any simulation at all).
Play with Ben on BBO
Experience "Ben on BBO" firsthand and connect with the bridge community and with the development team:
Try Our AI Bridge Engine
- Go to Robot World on BBO.
- Click Try our AI Bridge Engine to play a simple total points game featuring the latest version of "Ben."
Ben & Friends Daily Game
- Participate in an 8-board free tournament using the latest version of "Ben".
- This is an invitational game.
- How to join: Post in the Ben on BBO BBO Forums thread to be added to the tournament. You only need to do this once.
Feedback
We highly value the bridge community's input and encourage you to share your experiences, suggestions, bug reports and any hands you find interesting or challenging. Here’s how you can contribute:
- Ben on BBO Forums Thread: This dedicated thread is the primary place for discussions, feedback, and sharing hands. Our development team regularly reviews this forum to understand your needs and experiences.
- Ben Feedback Form: Accessible in Robot World, in the "Try our AI Bridge Engine" game, this form is a direct line to us for your comments and suggestions.
- Email and Other Communications: While the forum and feedback form are preferred, we welcome your insights through any communication channel you find convenient.
Your insights drive the continuous improvement of "Ben on BBO," helping us create a better experience for all users.
Changelog
Version 1.2 Beta (2024-11-19)
- Bidding:
- Retrained the bidding model using data collected from Ben (previously, the model was trained to bid like GIB; now it is trained to bid like Ben).
- Adjustments made to the neural network responsible for bidding behavior.
- Card Play:
- In certain situations, Ben incorporates double-dummy analysis, when sufficient information about the opponents' hands is available. Examples include:
- Toward the end of the game.
- When an opponent fails to follow suit.
- After a highly descriptive bidding sequence where the opponents' hands are nearly known.
(Previously, double-dummy analysis was never used.)
Version 1.1 (2024-03-17)
- Added: Human opening leads.
- Added: Some rules for bidding, to make it obey conventions more (e.g Blackwood)
- Added: More search and simulation.
- Fixed: Bug fixes in competitive bidding.
Version 1.0 (2024-01-08)
- Initial release of "Ben on BBO".
- Added: New card play engine which does not rely on double dummy analysis.
- Added: Updated bidding engine, trained with the latest GIB version.
- Added: Bidding explanations, "borrowed" from GIB.
- Added: Ability to claim. Uses GIB's claim engine.
- Added: Human-like play, based on comprehensive training data from 100 million ACBL hands played in human pair games on BBO.
- Added: Efficient simulation algorithm.