THEORY
To me it seems that the AI actually has a FORM of rational when deciding whether or not to accept a cease-fire. Number one is it's treasury: The lower it's treasury, the less likely it will seek peace. This sounds counter intuitive, but think about it this way. The computer is trying to fund it's war, money's tight, and it has few options, except to capture more territory. It realizes, however dimly, that a two front war could easily lead to a multifront war, which would cause it to stagnant simply trying to defend itself. You'll see this inaction when it's allied to all it's neighbors. So it decides to just stick with you and hope for the best. It also figures in what it can VISIBLY see of your forces, and thus bases its actions on said forces, even if they may be 1/5 of your overall available non-garrison forces. Third, it takes into account your own diplomatic finesse. It does this by the agent you use, the agent's experience, your past history of relations, length of time spent allied or neutral, and your influence. Based on all this information and its own flawed rational, it determines whether or not to accept.
THEORY