In particular, we’ve had a long running split in the computer world between “compute it” and “capture it”. In my own work in texturing, it has often come down to “generate a procedural texture” or “scan a texture image”.
Yet like most dichotomies, that turns out to be a simplification. In practice, people will scan a texture image and then use that image as source material for a procedure.
For example, you might use Photoshop to paint an image of “here is where the forest should go.” Then the places where you painted green will be used by a computer program to grow synthetic trees.
So in the best cases it’s not really “procedure versus data,” but more “procedure using data.” Now we are just entering a new regime where this partnership is really taking off.
That’s because of recent rapid advances in machine learning. The beauty of machine learning is that it builds a procedure from data. The more examples of existing data you give it, the better will be the procedure that it can build.
Machine learning isn’t a panacea — it will only show answers to new things that are similar to the things you’ve already showed it. But it’s a lot better than anything we’ve had before.
For solving completely new problems, we still need human brains creating procedures. Computers don’t know how to do that yet. And maybe they never will.
Which may not be a bad thing. 🙂