Automatic Leveling for Science Games

Every day, millions of citizen volunteers help scientists answer fundamental research questions by playing games: Citizen science games turn research tasks like classifying galaxies, finding protein foldings, transcribing old weather reports, or mapping neurons, into game challenges. Because they work with real research tasks, citizen science games face a tough design problem – balancing.

A core part of what makes well-designed games engaging is that they balance and lay out the difficulty of challenges to follow the players’ growing skills in lockstep: neither boringly easy nor frustratingly hard. Yet while entertainment game designers can create challenges at whim, citizen science games have to work with the given tasks of researchers, which come in all shapes and sizes, with no labeling how difficult they may be.

To solve this design problem, this project brings together game design and computer science to develop a system that automatically estimates the skill of players and difficulty of given research tasks, selects the most fitting task for each player, and generates new, artificial tasks if there are no tasks matching the player’s skill available to keep the player engaged. The goal is to develop a suite of software modules that any citizen science game maker can easily implement to optimize engagement.