James Earle, ASA Marketing and Communications Coordinator

From left: Eric Rose, Nick Meyer, Longshaokan “Marshall” Wang, and Maria Jahja of the NC State Laser Foxes. (Photo courtesy of Laber Labs)

In April 2016, the American Statistical Association shared its Discovery Through Statistics booth at the USA Science & Engineering Festival with a team of statistics students from Laber Labs, the creation of North Carolina State University’s Eric Laber. They were in Washington, DC, to showcase their latest project, a video game called Laser Foxes.

With its retro graphics and simple controls, Laser Foxes has the look and feel of an arcade classic. In the game, you control a laser-shooting fox with the goal of shooting down your opponent, an AI-controlled fox that not only shoots back, but adapts to your playstyle. As you move about the screen, taking cover behind colored blocks and gathering power-ups, the AI classifies your behavior and adopts the strategy best suited to counter your own.

Among adults and children alike, Laser Foxes was a hit—and thanks to the game’s design and the team’s explanations of its underlying statistical methods, each contender left with a better understanding of the different applications of statistics.

The focus of Laber Labs is the development of methodology for data-driven decision making in complex environments, with major application areas in precision medicine, management of infectious diseases, artificial intelligence for gaming, and STEM outreach. Their games are tools to study, implement, and test various machine learning and optimization algorithms that can be used to solve real-world problems. Laser Foxes is one such tool.

“Developing Laser Foxes was a lot of work, and they did all of it,” Laber jokes. He is Laber Labs’ eponymous principal investigator and the team’s faculty supervisor at North Carolina State University. Laber acts as a sounding board and facilitates brainstorming sessions on the artificial intelligence and educational components of the students’ work, and Laser Foxes emerged after he suggested the students add a competitive component to their games. In the hands of Maria Jahja, Nick Meyer, Eric Rose, and Marshall Wang, the idea took off quickly.

“I’m a big gamer,” Wang remarks, “and I’ve found the best games have simple rules, but are designed to allow different play styles and strategies.” Using the AI from precursors Zombies and Treadmills and LaserCat as starting points, the team worked to define those strategies, designing the game’s four deterministic AIs and a new tracking algorithm.

First, the computer player uses multinomial logistic regression to classify the human player as one of four behaviors: Forager, Aggressor, Camper, or Evader. The result is what they call the “FACE” of the player. Foragers hunt for power-ups, aggressors chase after their opponent, campers stay in one spot, and evaders run away. Using the estimated human behavior and other environmental variables gathered in real time, the computer player then decides which of the four behaviors to follow itself. Training that decision rule is done using policy search.

The result is an AI that challenges players to recognize its methods to beat it. Most commonly, the player exhibited traits of the Camper, keeping a safe distance and waiting until it better understood how to attack the opponent. But as Jahja noted, “Once people learned how the game worked, how it used statistics to classify their behavior, they would try to confuse the AI by erratically changing their strategy.”

Communicating those statistics was a design challenge in itself, and the team had to strike a balance between entertainment and education. “You just don’t have a lot of time to get people involved in a game or explain the statistics behind it,” Meyer explains, “so you have to ask, ‘What are the points we’re trying to get across, and how quickly can we do it?’”

Their solution is to show players their FACE—the extent to which their behavior fits into each of the four defined playstyles—as four bars on the side of the screen. As the bars shift, players can see how they’re playing, learn what the AI is ‘thinking,’ and change their behavior to beat it.

This interaction between player and AI was a new challenge for Laber Labs, but the applications are exciting. “We’re working on more adversarial games that adapt to what the player is doing using adaptive search, and that opens up a lot of possibilities for behavioral shaping,” Laber explains. “Can the AI push you to behave in a certain way? If we can get you to be more aggressive in the game until you’ve overextended yourself, for example, we can take those lessons into the educational materials we’re working on next. We can use statistics to prompt kids to behave in a way that makes it easier to learn faster or recall information better.”

The next game in development will put these ideas to the test by pitting the player against an infestation of termites. The game’s AI aims to inform the player of weaknesses in their strategy and provide them with suggestions to improve their approach.

Other projects in the pipeline include a library installation that will allow patrons to play one of their games while a computer attempts to imitate them in real time and an animated web series about artificial intelligence and gaming. Targeted at middle- and high-school students, the web series’ goal will be to inspire the best minds for the next generation of statisticians and data scientists.

Games like Laser Foxes are stepping stones to bigger applications and tests for valuable statistical methods. As Wang concludes, “Coming back from the festival, you could see good gaming and good graphics are very useful for popularizing statistical methods and attracting young audiences. That’s the goal; improve the gaming experience and graphics while improving statistical algorithms.”