Intelligent Systems
And their practical application

Really facinating stuff, and very applicable to some things were trying to do here.

Not that you should expect anything to ever come of this, but it was very interesting to apply some of the problems we experience here and try and translate them into something OpenCyc could help us with.

Immideately the first thing that came to mind is relationships. You could define individual entities in the system, then make assertions about them, then it would draw conclusions about them for you, given the proper question.

For instance. I could start by saying that memebers of the same Gang like each other, and -hate- members of other gangs. Then I could say that Bob was a member of the Arteries, and Chayney was a member of the Sinners... and given the proper query (what's Bob's disposition to Chayney?) it would draw the conclusion for me that they hate each other.

This may seem really simplistic at first, but when you start to add other factors which affect these (like make the assertion that attacking someone is against the law, and people won't break the law in the presence of a judge, or that a knife will loose in a gun battle) and any Object Oriented framework which you might build to represent this quickly becomes hopelessly complex, as the rules, and more importantly the exceptions to the rules balloon out in complexity.

My reaction to this is if an initial set of assertions were made about how we expect people to react, and then communication happened from the MOO to the knoledge base at a rather high level (connect, disconnect, entering/exiting a room), some very interesting and realistic behavior might be implemented. Obviously they won't carry on an intelligent conversation with you, but they might react just a little more responsively. Further assertions would enhance this, as it would make them more 'reactive' to situtations. (Even though Bob hates Chayney, Bob's on Chayney's turf and dosn't have a weapon, and can see that Chayney does. Proper response? Run.)

Has anyone had any experience with using a product/framework of this nature/implementing it into an existing system, or have any other good ideas about how this might enhance the world? Not that I'm saying I will... mearly curious as to what your experience has been. If I'm not mistaken, we could (theoretically) implement new NPC behavior entirely through assertions in the knowledge base.

Looking forward to your input.

-Kevlar

And exactly the type of problem this is prepared to tackle in a very efficent and sexy manner.

Simply assert that TERRA agents don't like law breakers, and murder is against the law, and Bob the Ninja is known to have commited murder.

Then when the disposition of the Terra agent to Bob is queried, the response comes back they really don't like them.

In a nutshell:
The application of doing this type of thing is a facinating journey into a language called CycL. It makes a basic distinction between a collection or an individual (The US Marie Corps vs. A Member of the US Marine core) and a dichotamy of multiple inheritance via the concepts of 'Generalization Of' (which is transitive) and 'Is A' (which is not transitive, nor is it inheritantly reflexive, nor is it anti-reflexive). Against this model you make assertions (Murder is against the law, A Judge dosn't like law breakers, etc.), and with your assertions form Microtheories (A set of assertions in which no assertion in the microtheory contradicts another assertion in that microtheory) which allows you to have a knolwedge base which is globally contradictive (that is to say there's assertions which contradict other assertions) but locally coherent (within microtheories). Thus changing 'viewpoints' is as simple as applying a different set of microtheories... which comprises a theory. An example of a Theory would be a Judge Behavior Theory: A set of microtheories which makes up your assertions on the data... your view into the data if you will.

In this design lies the key to deducing 'implicit' knoledge from the information we have 'explicitly' provided. The most difficult task is structuring your microtheories in a useful heircharchy and building your theories from them. But the results are fucking amazing.

-Kevlar

The 'basic ability', as Max mentioned, is simply a character's stats. Assuming a character has no melee skill, for instance, he can still beat you senseless with a baseball bat by virtue of his Agi, Str, and other stats. Sure, he's not used to the weapon so he won't swing all that well, he might miss, or swing to slow, too fast, but he has a fair chance of thwacking you nonetheless.

In the case of Max's example,

If someone lunges at you, and you're brandishing a knife-

a)If he comes -straight at you- when you have a knife out, then he's not showing much consideration for his safety
b)If he's not showing much consideration for his safety, we can assume he's concetrating on hitting you

These being true, we can assume he's using the 'kamikaze' posture. Whether you have the skill or not, unless you're particularly slow, you will probably strike him with the knife as he comes at you, skill or not, especially if you make an effort to stab him (i.e. posture offensive).

People have to stop thinking of game mechanics as game mechanics but rather as to what they represent.

In the above example again, the guy wants to kick your ass, but he sees your knife and doesn't want to get cut. What does he do? He attacks you more carefully, circling maybe, looking for an opening before trying to strike (i.e.  posture neutral), or, if he's really cautious, he might make a move to attack you, while still staying back, conentrating on avoiding your counter-blows until he finds a proper opening to hit you (postures defensive and guarded). You don't need any special code for this, its already there as far as I can see.

Frankly, it's insulting.

As for game mechanics, in my opinion, this is -entirely- about game mechanics and how to make those mechanics -more accurately- portray what should happen and how it should happen in the fictional reality that is Sindome.

We would do this by injecting 'common sense' as defined by a common body of knowledge, actions and reactions shared among people or groups with specific similar charateristics. �This, would make the mechanics seem less mechanical.

Maybe Bob the Ninja sees the knife and because (common sense tells him) the 50 chyen he might get isn't worth getting stabbed, (a situation that, in my opinion, is much more realistic,) he doesn't mug the victim.

Or maybe Bob has to make a descision about the intended victim that will determine whether he circles, attacks all out, whatever.

Should much thought be given as to what those mechanics represent, of course.

So in this case, I will be thinking about game mechanics.

From the website:
The Cyc inference engine handles modus ponens and modus tollens (contrapositive) inferencing, universal and existential quantification, and mathematical inferencing.

Modus ponens is defined as:
A rule of inference under which, given a knowledge base which contains the formulas "A" and "A implies B", one may conclude "B".

Modus tollens:
A rule of inference which can be derived from modus ponens under which, given a knowledge base which contains the formulas "Not B" and "A implies B", one may conclude "Not A".

Universal quantification is just a fancy way of saying you can make broad statements about things. Such as: All leaves are green.

Existential quantification means you can declare 'instances' of things. That is to say you can assert that man exists, and he can be Unmarried, or Married. You can then state that there -exists- a man who is unmarried (and we'll call him Fred). Note the distinction. In the first case, we made a universal qualification. In the second case, we made a existential qualification.

In other words, it supports the 'concepts' and 'objects' existing within the conext of concept(s). And it supports the concepts of explicit AND implicit deductions.

This allows for enormous potential. But if your not careful, it can bite you in the ass. If your sloppy about your assertions, it can 'infer' the wrong conclusion about things. For instance, if you say that judges don't like murders, and that law abiding citizens like judges, it could infer from that that law-abiding citizens don't like murderers, and they should be attacked on sight. The concept of 'something known' and 'environmental' factors are key in getting useful information out of this. To this extent I believe it's usefullness to the game is quite interesting because that's exactly the problem we sometimes face: The 'something known' and the 'taking everything else into consideration' problem. In a model like this that could move us away from having to explicitly track every nuance of everything and allowing 'known facts' and 'common-sence inferances' to take it's place.

Everything I've read tells me that the most time consuming part is finding existing concepts on which to base new concepts, and that there's no substution (presently) for familiarity.

Fortunately this thing comes out of the box with over a million assertions concerning human behavior and how it relates to things (money, environment, proximity, known facts, rumors, impressions, prior experiences, etc.).

Maybe someday...

-Kevlar