How A.I. Can Solve the World’s Problems: Challenge It to a Game of Bridge

Why you should care

If artificial intelligence explains its learnings in this complex, collaborative card game, we could learn to trust it.

Artificial intelligence is increasingly making important decisions, but we can’t have faith in its choices unless we know how they’re made. When a decision about a mortgage, a health care policy or a medical treatment is challenged, we currently turn to a human for resolution. If and when artificial intelligence becomes the ultimate arbiter, its reasoning needs to be clear enough so that if it’s mistaken, we can intervene. Trouble is, AI learning is opaque — it involves building associations and relying on patterns that humans can’t understand, and the tech industry is struggling to make AI explain itself.

To solve this transparency problem, I’d like to challenge AI to a game of bridge.

If AI is able to map its decision-making process to the rules of a game, it could then lay out those rules for us to see. IBM’s Deep Blue learned the rules of chess so well that it was able to beat human champ Garry Kasparov. Google DeepMind’s AlphaGo, likewise, beat top-ranked Go player Lee Sedol. In these cases, we were able to understand how the AI made its choices because we spoke the same language: the rules of the game.

AI that could not only win the game but also explain the rationale for its moves in response to a string of unknown factors would have limitless applications.

The next step is to try to apply this shared decision-making language to nongame contexts. Could AI address real-world problems, ranging from energy efficiency to medical treatment, and tell us about its decisions using the rules of the game? 

Perhaps, but first it will require more-complex games to adequately span the complexity of real-world decisions. Chess and Go are, at their core, competitions between two parties based on an entirely known set of facts. Such scenarios are rarely found in the real world. Other games, then, may offer more potential.

A rivalry between Czech-Canadian collaboration DeepStack and Carnegie Mellon’s Libratus has begun to address some “real-world complexities” by exploring applications within the more complex game of poker. These products demonstrate that a computer can factor in possible “held” cards and the associated moves of multiple players, and they explore ways technology can account for incomplete information that remains constant in almost every decision humans make. But the noncollaborative nature of poker limits the technology, which is why we need to turn to bridge.

 

If you’re unfamiliar with bridge, what you need to know is that it’s an incomplete information partnership game, incorporating both competition and collaboration. A bridge player is compelled to disclose to his opponents the information he’s passing to his partner, whether in the bidding phase or during card play. Many argue that, because of this, real-life lessons can be learned from playing it.

AI that could not only win the game but also explain the rationale for its moves in response to a string of unknown factors would have limitless applications. The “black box” of AI-based decision-making would, theoretically, become clear, providing greater accountability and understandability of such decisions. And this would allow us, as a society, to apply AI to areas where we’re currently reluctant to fully trust it.

To get there, though, AI must first master the game of bridge … but can it? Cracking the game, technologically speaking, is not a novel quest. The first bridge-based computer program, Bridge Baron, was created in 1982. Various companies and developers have built bridge bots and competed against one another in World Computer-Bridge Championships since 1996. To date, not one of these computer programs has come close to outplaying a bridge expert, let alone beating a world champion.

Over the past year, though, the bridge technology market has seen a number of shifts that hold potential.  Recent market consolidation of the leading online bridge providers presents unique dynamics that lend themselves well to AI mastering bridge. On the one hand, the merger of GOTO Games with Bridge Base Online (BBO) reflects a centralization of talent: The companies bring together the expertise of two world-champion-level bridge players, Jérôme Rombaut of GOTO Games and Fred Gitelman of BBO, instrumental in the creation of the AI technology supporting their respective platforms. On the other hand, the merger of Black Ridge Acquisition Corp. with Ourgame International to form Allied Esports Entertainment reflects a direction of focus — building on their dominance in the Asian bridge market to become a leader in the gaming industry, online and otherwise. 

While these factors will contribute to a competitive ethos placing high value on constant technological improvement, the AI developed will be limited in nature to the game of bridge itself, leaving the question of alternative applications untouched. 

One French startup, NukkAI, is looking further ahead and seeking to build a technology that will not only win the game but will also explain its reasoning. By tapping the expertise of the world’s top bridge players with world-class researchers from different artificial intelligence companies, NukkAI is creating a hybrid AI technology that addresses different elements of the game of bridge, produces decisions that are explainable and could potentially be applied to problems requiring similar thinking.    

These recent advancements in the state of and approaches to AI offer a unique opportunity to finally unlock the game of bridge technologically and answer many of the most significant questions plaguing the future of AI. 

This may not make everyone comfortable with the technology, but it would help the average person better understand it.

So, who’s game?  

OZYOpinion

Interviews, op-eds, and analysis to help you make sense of the news of the day and the news of the future.