Exam reading: “Games with a purpose”

People teaching computers to do stuff. Actually, I’ve played one of these games (though a 1-person variety). It was kind of fun…

Luis Von Ahn and Laura Dabbish. “Designing Games With A Purpose.” Communications of the ACM. 51(8): 58-67, 2008.

Summary: Games with a purpose (GWAPs) involve people performing tasks that can’t be automated, e.g., image tagging, collecting facts, etc. Related to open-source software movement, non-game crowdsourcing, and gamelike interfaces in business apps. The authors first describe three categories for these games: 1) “output-agreement:” players see same input & must produce same output (e.g., give same label to photo); 2) “inversion-problem:” one player describes something, other guesses it (sim. to 20 Questions); 3) “input-agreement:” players are given inputs and must describe them to see if they have the same input or not (e.g., both describe a music clip and then guess if the other player’s clip matches yours). For enjoyable play, need to add features to these templates: time limits, scorekeeping, high scores, randomness, leveling up. They describe mechanisms to guard against player collusion, e.g., cross-checking, random matching of players; games can also be modified for n¹2 players. Games are evaluated by throughput * enjoyability (average lifetime play)= “expected contribution;” this doesn’t capture popularity or word of mouth. They point out that their examples focus on similarity/matching- need a different template for gathering diversity.

The goal of GWAPs is to capture large datasets for developing programs with advanced perceptual capabilities.

Comments: Focus here is on machine learning, rather than human (“useful computation as a side effect of enjoyable game play”-61), but one could potentially link such a system to an educational tool, for a crowdsourced informational resource.

Links to: Brown & Adler (general online learning); Howe (crowdsourcing)