To close this handout, I should mention that it is now evident that very few, if any, of the subprograms (such as the planner) built into PopBeast will scale up to larger problems.
To digress for a bit, it's interesting to note that there are precedents for such scaling problems in nature: the worst of Hollywood aside, a man-sized flea would not be able to jump over the Empire State Building. Insect respiration relies on oxygen being able to diffuse from pores in the exoskeleton to every part of the body, and this is only possible for small insects. Above a certain size, a complete redesign - lungs, with their fractally enormous surface area - was needed.
Perhaps it's not therefore surprising that AI methods which work on a very small scale should fail at a larger one. In fact, there is now an entire branch of computer science devoted to the way in which the time and space requirements of algorithms scale with the size of a problem.
And it's important not to underestimate the difference in scales between microworlds and the real world. A typical AI microworld, Eden included, contains perhaps ten to fifty ``objects'', where all objects of the same type are identical. Generally, they have no internal structure, no parts, and at most ten different attributes. In addition, the AI inhabiting such a microworld typically receives complete perceptual information about its environment. Finally, there is usually nothing corresponding to friction, weather, predators, or anything else which can cause the environment to behave unpredictably.