Monday, January 4, 2010

The Devil in the Details



The value of simulating as many elements of a world as possible is, shall we say, non-obvious. One is inclined to say that simulating more elements will naturally make the simulation more realistic, but there are particular reasons for this, and if you don't simulate the rules and physics well enough, you're not doing much simulating.

What I'm driving at here is the idea of complexity. The interactions between microscopic objects in an environment can often yield macroscopic changes. As one example, consider a breakable wall. In a simple 3D environment like Duke Nukem, that wall might be represented by a single polygon, which can receive a certain amount of damage from weapons before the whole thing breaks. In real life, that wall is composed of a series of bonded molecules, and it is going to break differently depending on what type of wall it is. If it is made of rock, forming a crack at one particular part of the wall, and focusing one's attention on that crack, could be enough to bring down an otherwise impenetrable wall. Similarly, if the wall is made of wood, forming one hole in it can be enough to destroy the rest of it.

For the given example of simulating a world for one man, in order to learn his secrets, these things might not be very important (unless you have to make sure he doesn't catch on that he is in a simulation). But if your attempt is to make a kind of "what if" hypothetical in a computer model, and get as good of an idea as possible what would happen in the real world, you need your simulation to be as isometric as possible. The main uses for supercomputers are exactly these kinds of situations, where the details are phenomenally important: quantum physics; molecular modeling; weather forecasting; and physical simulations of things like airplane wind tunnels, and nuclear explosions.

For any given simulation, there can be a value in simulating as many things as possible. But whoever designs the simulation needs to decide where the "diminishing returns" set in, where you have a level of accuracy that is not actually going to be more informative. In our simulation where we want to know where the bodies are, simulating anything beyond Earth is going to be unnecessary. We also don't need to simulate any of the people that our "player" is never going to come into contact with.

If you wanted to simulate the interactions between a population of people, though, you would want to simulate everything about those people, down to their bodily functions, because the fate of the world could very well rest on whether John Hammersmith of Liverpool, England needs to take a poo at 12:54 pm of next Tuesday.

No comments: