Read the previous post before you read this one.
It is amusing that in a post that started by being about ways of communicating, I failed to make clear the difference between the various AI bits. Lemme 'splain:
Each of the AI bits uses a different algorithm. They aren't filters. They're full-blown systems which parse different data in radically different ways. The only thing they have in common is an emotional framework which affects all of them: if you are pissed, the way you look at the world changes at every level.
Other than that, they work on different algorithms entirely.
The "reality" system does short term goals, long term goals, and situational extrapolation. It does this by a long list of fuzzy rules which it trawls through. For example, "fights cause harm x 0.4" when harm is understood to be bad. The fuzzy multiplier is smaller the more powerful you are, because the fight is less likely to do you harm. Similarly, the enemy has a "visible power" which modifies your own multiplier.
By trawling through this rule set, the reality system can determine what basic actions the character needs to do in order to achieve or avoid a given result. This is a fairly basic system, but requires a fairly large number of explicit rules. Remember, this trawls, so it could go, "insults cause punching", "punching causes fight", and "fight causes harm".
The "social" system doesn't use logic or trawl through anything at all. It simply judges each person with a few stats. These stats, such as "attractiveness" and "power", are then used to determine what actions the system wants to perform to these people. These actions are passed to the reality engine, which determines exactly how to do them. This engine also handles conversation, because it has all the necessary emotional and relationship parameters already.
The social system is an active system. It doesn't simply judge the situation, it actively seeks out friends and enemies you know you have. The reality engine might have a sweet adaptive rule set that takes these things into effect, but that's pretty advanced...
(The social system could be done with a neural net or simple alife, if you wanted to be all snazzy and inefficient...)
The "convention" system uses a long list of simple rules. Unlike the reality system, there is no "tree" - no "this causes that". Instead, it is simply a list of situations and proper responses, paired with emotional feedback for achieving or missing the proper response. This list is the primary means of emotional change, the only other one being goal accomplishment and failure (remember, the social system causes goals to be made).
The convention system is similar to the "law" system. In fact, they could probably be compressed into the same system. It's not like I polished the concept.
These aren't differences in filters. The data they take in is different. The effects they efficate are different. And, most importantly, the algorithms they use to decide what data should be responded to in what way are radically different.
Not simply filters. Filters are a different concept entirely.