Ben Gonshaw: Digital Media Theorist & Game Design Consultant
Home Latest Updates Design Rants Musings Links Comments

THE EMERGING MACHINE

2nd May 2005

Emergence has long been touted as the technology to transform the game experience. However, machine learning may play a more central role and provoke greater changes. A good way to examine this is by looking at the current darling of gameplay, the sandbox and comparing the impact that both technologies could have.   Defining the Sandbox A good place to start would be to establish what a sandbox is in this context. As I imagine Gary Penn of Denki would say, a sandbox is a place to play with toys. For me this is the leading characteristic of the play style: provide the player with a set of abilities and an environment to experiment with them.

Not having played the much-vaunted Katamari Damacy, I imagine that the ability to roll around and have things stick to you, making you bigger is a fun one. I can imagine having fun completing the challenges in the levels they set out for you, but I can also visualize just rolling around a vast world at my leisure, becoming larger and larger (and possibly getting smaller by rolling in water or some such so that the gameplay is open-ended). This strikes me as being fun. The idea of being able to test out the powers at your disposal, without punishment or artificial limits is a great draw.

It is fun to experiment, to set your own goals and to play with a virtual toy in a non-punitive environment. In a sense this is what made GTA great, because it blended traditional missions with the ability to play with the toys it gives you at your leisure.  There are three key factors to a good sandbox:   A Good Toy Ability There must be fun central ability that makes it a good toy, think yo-yo, hoop and stick, silly-putty: all easy to master experiences with satisfying tactile feedback.  A Good Playground What's the point in having a skateboard and ramps but making all the ramps the same? A good playground must not only give you the means to use your abilities, but it must supply a good variety of opportunities to use them in many different ways. Building the environment specifically so that the player can make combos (whatever they may be in your game) can be a hindrance as well as a help. You must guarantee that combos can be strung together with some creativity on the player's part, beyond simply finding the order to use scenery in. This will make the experience more involving and rewarding. If the player is a partner in authoring their play, then they will appreciate it much more.   Inexhaustible A successful sandbox is inexhaustible. MMO games are the archetypal sandboxes, as they are designed to be inexhaustible. Further to simply providing an unending challenge, a sandbox should also be a relatively safe environment. In the MMO you respawn or start a new character to continue. However, death or stuck scenarios can be avoided by clever use of the ability and the environment.

The Sandbox and Machine Learning It is in the final category of the world being inexhaustible that emergence comes to the fore. If the game world can change and adapt to the player, but not compromise points one or two then emergence can hold the key for extensible fun.

Emergence can prove to be a fun ability in itself, or even a great playground, but a similar effect to such adaptive content that moulds itself to the player can be acheived through other methods. For example, while the player is playing they are training neural nets, or creating fitness functions for GA's. These exemplars are then used to build the world around the player, or to alter the make-up of the content within it to maximise the player's enjoyment of their abilities. Rather than use AI to create a scaling difficulty curve that fits itself to the player's ability, the AI is used to maximise the player's enjoyment, to help them to wring out extra combos by making pedestrians walk into the correct position, or by slightly narrowing walls, shifting heights, subtly slowing or skewing the direction of player's or opponents' avatars to create the collisions to trigger the bonuses: there are a myriad of parameters that are available for adjustment. Machine learning in games could prove to be much more effective as a wow factor than emergence ever could be.

So while complexity based adaptation results in convincing, realistic movements and behaviours, machine learning can create similar but useful behaviours. This is a game after all and window dressing can be nice, like the physics in Halo, but if that window dressing is functional and adds to the gameplay, like the physics in Half Life 2, then it enhances the game and not just the graphics. The overall experience that is creates is a better one.

Write Comment : Back to Musings