As we all know, the project of artificial intelligence has been nothing but an abysmal disappointment from its beginning in the 60s, and anyone asserting otherwise should check his/her facts. Charles Csuri and James Shaffer wrote back in 1968:
At M.I.T. and Standford University considerable research is in progress which attempts to deal with artificial intelligence programs. Some researchers suggest that once we provide computer programs with sufficiently good learning techniques, these will improve to the point where they will become more intelligent than humans. [italics added]
More intelligent than humans, you say? Pretty bold statement, considering we are now 43 years later nowhere near replicating human intellect, or even of an earthworm for that matter.
So what makes this situation particularly interesting from a standpoint of real-time imagery is not that we have now, and will continue having, dumb enemies in first-person shooters. No, that’s secondary.
The real issue lies in procedurally generated content, and the fundamental nature of it. The thing is, as game worlds in general tend to grow larger and more detailed year after year, the workload behind them often increases logarithmically as a result. Developers have tackled this issue by using complementary procedural methods, algorithms, for their level and art asset creation, like in Fallout 3, Mass Effect 2, and other high-volume/density settings.
However, when using a certain algorithm, i.e. set of rules, as a means for artistic creation, it shines through like a supernova: Everything has this same, unifying feel to them – and I’m not talking about style here, which is completely different issue. I’m talking about the algorithm’s inability to produce anything meaningful, genuinely novel structures, which reduces to the failure of artificial intelligence discussed prior. Indeed, the human mind[1] is a sole artistically creative agent in the known universe, and as such, in a unique position.
Notice how the employed algorithm (NURMS subdivision) doesn’t add design to the geometry, but merely refines it according to certain general principles, causing them all to share the same look and feel.
For some reason, our emotional response to procedurally generated content differs fundamentally of the stuff originated by genuine design. We often find non-design uninteresting and boring, which I believe stems from the human’s inherent ability to recognize patterns, especially when one is bombarded with them like in a video game containing procedural material. And no algorithm will ever change that.
Interestingly, there are despite that cases in which algorithms are more suitable than design, which involve usually some sort of an undesigned natural occurrence. In theory, for instance clouds would be more than suitable for procedural generation, since clouds are generally not designed, but formed by natural forces. However, in practice, algorithms aren’t yet sophisticated enough to provide interesting, convincing results, and I have yet to come across procedural clouds that I would be happy with. Trees I have, but they indeed are fairly easy target for algorithmic creation.
Also, various simulations, such as of light or physics, are better to be carried out using an algorithmic approach, since they don’t include designed elements. In fact, before we had simulated, algorithmic physics, we had designed, animated physics that were obviously far inferior than simulated ones. Remember before ragdolls, how awkward it was to see a body laying down completely stiff on a staircase? That indeed was a horrible time in history.
Bottom line is, design is not replaceable. And algorithms have their place.
[1] One could indeed make a similar case about certain animal minds too, but I wouldn’t.
Pixel Counterrr 