Posted by Rob Knies

Project Waterloo

Have you heard of Colonel Blotto? Me, neither—at least not until a few days ago, when I ran into my friend and colleague Thore Graepel  in the atrium of Building 99 on Microsoft’s Redmond, Wash., campus, worldwide headquarters of Microsoft Research.

Graepel was telling me about Project Waterloo, the initial effort from the nascent Research Games project at Microsoft Research Cambridge. Project Waterloo is a Facebook game designed in the Colonel Blotto style, which means that two players are asked to distribute a finite set of resources over a collection of geographies. The player who has distributed the most resources over each geography is the winner of that one, and the player who wins the most geographies wins the game.

In the case of Project Waterloo, each player is allocated 100 “troops,” to be distributed over five “battlefields.” The player who wins the most battlefields wins the game.

The group’s website refers to Project Waterloo as “fun, engaging, and slightly addictive,” and that’s easy to believe, but still, why are Graepel and his colleagues pursuing such an endeavor?

As you might imagine, it’s not solely for the glory of Colonel Blotto.

“Social networks offer a unique opportunity for us to scale up research in behavioral game theory—and social science in general,” says Graepel, a principal researcher at the Cambridge facility. “Project Waterloo is our first attempt at dipping into this resource to potentially bring our research to planet scale.”

Classical game theory makes predictions on how rational agents would behave strategically, with implications in such scenarios as advertising, business interactions, and the job market. But predictions about human behavior often rely on strong assumptions, particularly in network domains. Agents in such studies are assumed to be entirely rational, maximizing utility, employing complex reasoning, and being equally rational as their adversary.

In the real world, such assumptions don’t hold true. Real people can act emotionally, can make decisions based on fairness or reciprocity, and might not be capable of complex reasoning. Hence, the need for a real-world platform to examine people’s strategic decision-making.

“Our research is all about comparing theory and practice,” Graepel explains. “Game theory predicts certain behaviors for interacting rational agents. But how rational are humans, really? This project will help us find out.”

Project Waterloo is planned to be the first foray of a Facebook Game Theory Lab at Microsoft Research Cambridge, one that will enable experiments involving people interacting with friends while playing games, examine how their behavior differs from predictions made by classical game theory, and devise tools to improve behavior prediction.

Opportunities available by using Facebook include:

• Viral marketing: recruiting subjects groups many times larger than in previous studies.
• Large scale: achieving a degree of demographic and behavioral resolution previously unimaginable.
• Social-graph embedded: supplying unprecedented types of analysis, such as distinguishing behavior toward friends and strangers.
• Natural habitat: As opposed to an artificial lab setting, players can engage in a setting where they feel comfortable.

“What is most exciting about this project,” Graepel concludes, “is the ability to recruit subjects from all walks of life for our research while, at the same time, offering them a fun and challenging experience.”

See for yourself.