The Singapore MIT GAMBIT Game Lab has released
Improviso, a two-player improvisational performance game designed to capture data that will help train new, more human-like artifical intelligence systems.
First debuted at
a PAX East panel on dialogue in games last weekend,
Improviso asks players to take the role of either the Director or Lead Actor in three short scenes regarding a government cover-up of an alien landing. Players are matched up randomly online and interact with the scenes using a simple point-and-click interface and a text-based chat log.
Much like the team's previous project,
The Restaurant Game,
Improviso will use recorded data from these human performances to help train an AI system to play the role of one or more characters, interacting with human players in a lifelike manner.
“My interest is in establishing new genres of games,” said
Improviso creator and AI researcher Jeff Orkin in a statement. “We have this popular genre of role-playing games, but
Improviso is really about role-playing like you did were you were little: pretend play.”
In
a post on his blog, Orkin discussed the challenges in getting players, who might not be experienced with games or improvisation, to use the game to act out roles such as that of an Area 51 agent.
The development team used a detailed tutorial, sample gameplay videos, and a slow introduction of characters and scene elements to help acclimate players to the game's relatively unique gameplay mechanics and style.
Over 100 people have tried the game so far, and Orkin says the best time for players to find partners to help increase that number is between 7 and 10 p.m. in their time zone.
