Warren Spector has a dream. He calls it the “one city block” RPG idea, and he’s talked about it several times. The design consists of a relatively small area that is modelled with such detail and interactivity, and has NPCs that are so “alive”, that this small area can support as much gameplay and story as a “normal”, mission-driven game. Warren was certain that with Deus Ex he was going to get the chance to create this dream design of his – in fact, the original design of Deus Ex called for three such city-blocks – one in New York, one in Hong Kong and one in Paris.

As the development of Deus Ex continued, he discovered that the technology to implement his design still wasn’t there, even after all these years. So as the development of Deus Ex progressed, the game became a lot more linear and mission-driven, like other games.

But Warren’s attempt wasn’t a complete failure. The second mission of Deus Ex is set in New York, and comes very close to fulfilling his goal. Far too much stuff is scripted, but there are a lot of people on that map, all with different personalities and goals. There are a lot of very different places, and a metric ton of secrets (including a secret mission that is very easy to miss). All of these make Hell’s Kitchen feel more real than environments in most other games.

Hong Kong is almost as good, but by the time you get to Paris it becomes apparent that the incredible amount of work necessary to put a map like this together was beginning to wear on the designers. Paris is the last map with free-form elements in the game – all of the rest of the maps in the game are very traditional, mission-based maps. They still have secrets and alternate paths, but the conflicting goals, the complex NPCs and the scripted optional sequences are gone.

This is one reason that Warren has called for the development of tools that could generate game content automatically.

Content created through generation used to be extremely common in gaming. All of the maps of the original Populous were created simply by generating strings of random numbers (and Peter Molyneux admitted in his 2000 GDC presentation that the Promised Lands expansion pack to Populous simply consisted of those random numbers reversed).

Starflight is another older game that benefited greatly from generated content. Starflight attempted to give players the experience of their own Star Trek five year mission. In order to do so, it was necessary to give the player a very large number of star systems and planets to explore – too few and the player would feel too constrained and wouldn’t get the sense of grand exploration that the designers were going for. So the designers stuffed the game with over two hundred star systems to visit and over eight hundred individual planets to explore. But Starflight was made back in 1986, before hard drives were common and CD-ROMs even existed, so the game had to fit on floppies. The only way the designers could fit so much content on two 360k floppies was to generate each planet on the fly using algorithms.

So if generated content (both pregenerated and on-the-fly) was so common, why isn’t it used any more? Why is Warren’s request so difficult to fill?

If there is one thing the human brain is good at, it’s pattern recognition. In fact, the brain is so good at it, we humans often see patterns where none exist. The nature of the “game worlds” of Populous and Starflight were very simple – they weren’t hard to generate, and they weren’t large enough for patterns to develop to clue the player in to the fact that they weren’t hand-made.

That’s all changed. Generating a Shenmue-style world is a much tougher challenge than generating a 64×64 heightmap – but it may not be impossible. If this problem could be solved, it could allow game worlds to get bigger and more detailed without requiring teams to double in size and schedules to extend even farther than the industry standard eighteen months. It could also finally make Warren’s dream a reality.