[MUD-Dev] OpenCyc, design implications of ontological systems?

Bruce Mitchener bruce at cubik.org
Mon Apr 15 19:22:10 New Zealand Standard Time 2002


I haven't had the time to follow up to Robert Zubek's reply quite
yet, but this appeared in the OpenCyc forums over the weekend:

   http://sourceforge.net/forum/message.php?msg_id=1538343

<EdNote: Reformatted for clarity, attribution added>

--- begin ---
From: Douglas Miles <dmiles at users.sourceforge.net>
Date: 14 April 2002
Subject: RE: Applications

I have plans to use OpenCyc in project shown at

  http://logicmoo.sourceforge.net

The project is based on hooking OpenCyc into an existing RPG system
called Jamud. (Java MUD)

This to accomplish three goals: (also the three stages of the
project)

Stage 1

  Goal: To test and do practical Knowledge Engineering about a
  virtual realm:

    1) Build an existing simple realm inside of Jamud.


    2) Create rules that interface to the realm Jamud mapping
    existing object types into cyc Collections

      a) Enumerate the possibilities and confirm the ontology of how
      things work or exist.
  
      b) Map/Align to Cyc's ontology

      c) Create a BroadMicrothery to reflect this

      d) Create an import routine to build the world into a
      DataMicrotheory

    3) Inference in Cyc about the DataMicrotheory and see how many
    commonsense rules we can make about our possible world.

    4) Complete the interface for querying/asserting to OpenCyc from
    Jamud telnet interface.

    5) Create synchronization: (choices)

      a) Each time a subscribed event occurs in Jamud, use the
      existing OpenCyc Java API to update Cyc (current)

      b) Make each object instance get/set methods query/assert into
      OpenCyc into a DataMicrotheory (could easily be done)

      c) Create HL interface predicates to query the existing Jamud
      objects into a DataMicrotheory (supported yet?)

Stage 2

  Goal: To animate non-player character agents from Cyc rules

    1) Create an ontology of non-player character roles and jobs.

    2) Leverage commonsense rules from stage 1 into rules about how
    to conduct actions

      a) Create cyc queries (and the rules to satisfy) that
      consequent in a form of (actorStateAction ?Actor ?State
      ?ActionsToGetThere)

      b) Map the agent categories into partitions of background
      commonsense required

      c) Create a Microtheory system for single agents based on
      'knows about' and 'know how to' that can instance individual's
      current knowledge states.

    3) Interface!

      a) Create methods in Jamud that builds and initializes a new
      Cyc microtheories per agent.

      b) genlMt the agent character ?Mt based on job types for their
      commonsense

      c) Agent character enumerates his intrinsic goals by making a
      query to OpenCyc

      d) Agent character per-tick initiates query about its goal and
      gets bindings back for his next action

      e) Perform action inside Jamud and update states

    4) Practice and refine knowledge engineering for stage 1 and 2

  Stage 3:    Expertly build rooms with objects based on rules.

    An example:

      (isa Mold FirstOrderCollection)
      (isa Mold ExistingObjectType)
      (isa Mold StuffType)
      (forAll ?Rooms
           (implies
              (and
                (isa ?Room Room)
                (hasProperty ?Room Damp))
              (thereExists ?Mold
                 (and
                    (isa ?Mold Mold)
                    (objectFoundInLocation ?Mold ?Room)))))
      
      (forAll ?Action
         (implies
           (and
             (isa ?Action TakingPartOfObject)
             (occursAt ?Action ?Room)
             (objectFoundInLocation ?Object ?Room)
             (rolePart ?Action Actor ?Actor )
             (rolePart ?Action Object ?Object)
             (rolePart ?Action Region ?Room)
             (isa ?Object StuffType))
        (thereExists ?PartStuff
             (and
               (part ?Object ?PartStuff)
               (objectControlledBy ?PartStuff ?Actor))))
      
    So we should be able to telnet into Jamud, enter a "damp room"
    and find and take parts of mold.

    1) Use the information from above stages to 'reflect back' into
    Jamud.

      a) Find which state-action-state rules from stage 2 are
      administrative tasks

      b) Write a Microtheory that contains rules that both antecede
      and consequent in the DataMicrotheory like the first of the
      rules shown above.

      c) create a robust colorful environment that may go beyond
      persisted Jamud

    Practice running queries until the results make good sense

    2)  Interface
  
      a) Jamud queries the DataMicrotheory (leveraging the above
      rules) and gets a map of the expected state.

      b) Confirm that each object exists and has the expected
      properties in Jamud.

      c) Make sure that the world transitions make sense to players
      in the MUD.

  Stage 4

    Goal: Apply virtual environments beyond this MUD/MOO/RPG
    analogy.

    1) Confirm the constancy, soundness and operational completeness
    of the above.

    2) Create agents/objects in Jamud that will propagate/query
    state to another application realm

    3) Do knowledge engineering like the above for the new
    application

    4) Evaluate and improve tools used.

The project is realized into finite number of non-vague tasks and
diverse enough.

ETAs?

  Stage 1, sometime at the end of next weekend.  We should have
  Jamud completed tonight.  (Alpha6) Alpha5 (build version) had too
  many tangled objects (not all reifiable to data). But is now
  converted to a very simple Node-by-Node object system that
  translates to a single DataMicrotheory.  We are on 'Stage 1 Step
  2d' and learning what was incomplete from 'b' and 'c' By the end
  of this week should be experimenting with 'Stage 1 Step 5b' (5c is
  not yet available)

  Stage 2 should be started (i find very exciting!) sometime late
  April (this month). But it is possible that other people could
  contribute ideas and methodologies to speed this up. This is
  mainly made up of Knowledge engineering since the java code time
  is only a couple days.  I am a programmer before a logician and
  there are people that are better Cyclists. So I will have to make
  sure they have a solid application framework to play with
  scripting NPCs.

  Stage 3 is a little vaporous since it depend on how we synchronize
  in this stage. But the knowledge engineering required should be
  something others can work on after this Friday once they are able
  to import the 'DataMicrotheory'.

  Stage 4 is of course is contingent upon success of the previous
  stages
---- end ----

_______________________________________________
MUD-Dev mailing list
MUD-Dev at kanga.nu
https://www.kanga.nu/lists/listinfo/mud-dev



More information about the MUD-Dev mailing list