search

ensemble-engine / ensemble

A rules-based AI framework for social simulation
Other
50 stars 11 forks source link

"Reverse volitions" don't always work for determining if an action should be accepted #8

Open bsamuel opened 8 years ago

bsamuel commented 8 years ago

"Reverse volitions" don't always work for determining if something should be accepted.

Here's an example:

Consider volitions between a character who is ‘arrogant’ and a character who is sycophantic. Sycophant has high volition to praise arrogance. Arrogance has low volition to praise sycophant. ---> Does this mean that Arrogant is likely to refuse praise! I think so, but that doesn’t seem right!

This should be addressed somehow.

meldckn commented 4 years ago

Another example:

Say you have ApplyForJobSuccess and ApplyForJobFail actions. They're directed from applicant to potential employer.

But you don't want the default reverse volition-scoring to determine if applying for a job was successful or not. Because it would mean "I would get hired by you if you would get hired by me" if the roles were reversed. Or "You would hire me if I would hire you." Which is just not how applying for jobs works, unfortunately.

—Real example from @bluestar514's thesis project domain!

For now, for action rules that don't make sense to run backwards, just don't put the isAccept (either true or false) property in actions that you don't want to have the "reverse volition" scoring. This is assuming I'm correct in my understanding that the presence of the isAccept property is what specifies to Ensemble whether to run the volition scoring leading to this action in the other direction to decide if it's a successful action or not.

Ideally, authors would have more explicit control over what it means for this action to be "successful" or not. As in, what function/algorithm determines that.

One possibility is described in #43, of adding a field to volitions to specify for whom this rule is applied:

  • “initiatior” means that the rule pertains to a character wanting to engage in an action.
  • “responder” means that the rule pertains to a character determining if they will accept or reject the action if it is done to them.
  • “both” means that it affects both the initiator and the responder, as outlined above.
meldckn commented 4 years ago

Oh wait, in my tutorial slides I say that isAccept is true by default. Does that mean actions fulfilling intents about changing directed or reciprocal schema types are always impacted by the reverse-volition acceptance/rejection scoring? If so we don't have a good workaround and this is a higher-priority feature to add / concern to address.