Most of the time, when I talk about the crowd and games on this blog, it’s a discussion of the benefits of citizen science, and how the playing of games by the public can support scientific efforts.  A recent effort by the games company GiantOtter attempts to employ the crowd for something more traditional.

They were tired of the relatively simple way the AI in most games operated.  The computer opponents in many games were often following relatively fixed and rigid scripts, which made them unworthy opponents when stacked against the potential of an unpredictable, human opponent.

Most AI behaviour in games is programmed in by the developers themselves.  GiantOtter believe they can make things more realistic by employing the crowd to get involved in this process.  They believe that crowdsourcing this step can provide the game environment with a potentially endless supply of possible responses and behaviours.

The company is currently in the process of pulling together inputs from the crowd for a number of player response situations for a series of mini fantasy games that the company has developed.  The participants converse with one another online as they discuss each scenario, and their conversations are recorded and transcribed using speech recognition software.  The conversations are then tagged with keywords so that the game AI can employ the right speech and behaviour in the appropriate circumstances.

The first game that they hope to deploy this approach in is called The Restaurant Game, which is due to go live in November.  For the game, they are recording the behaviour of thousands of people in a virtual restaurant to try and make it as realistic as possible.

The aim is that with this rich array of scenarios to choose from, it’s unlikely that the AI in the game will ever act the same way twice, thus making the game itself much deeper and more varied for players.

Suffice to say, even these advanced algorithms are no match for the human opponents found in massive multiplayer games, but the developers believe their approach offers a significant improvement on what is currently available.