While the role of the understanding phase is to describe the problem in a more explicit fashion, it should not attempt to explain the solution. The descriptions should be purposefully left informal (that is, in English) and high level, to influence the AI design decisions the least possible. Naturally, we will have some ideas while we describe the problem, but we should try to keep it unbiased; this is a good premise to help us come up with the best solution.
Another advantage of the intuitive nature of the understanding phase is that anyone can do it. All it takes is a bit of experience with computer games and some terminology to describe the situations. Then, all we need is a small amount of research to put us on the same page, and place these problems into context with respect to existing work. This is of the kind of understanding the previous two sections provide—although with a few more details than necessary. At this stage, we should be more than capable of communicating with other developers about technological requirements, and also dealing with vague requests from artists or designers.
Generally, the list of criteria will be relatively ambiguous—a curse associated with all natural languages. Ambiguity can be considered as an advantage because it translates into flexibility for the solution. However, misinterpretation is a nasty side effect of this flexibility, which may break assumptions with other aspects of the design. Pay particular attention to the criteria that have strong dependencies to other aspects of the design, removing ambiguity as much as possible.