[MUD-Dev] Efficient AI?

[olag at ifi.uio.no]

I just thought about how to implement efficient markov chain based
AI for a MUD. Maybe this implementation strategy is unworkable, or
maybe it is well known, but it might be worth a discussion.

ASSUMPTIONS

  1. Selection the next ideal state is based a complex function F
  and a wide array of parameters

  2. You can create a hashing strategy which approximates F. (by a
  genetic algorithm for instance)

  3. Picking the wrong next state isn't critical.

  4. Life-like AI is more important than smart AI

  5. Efficiency is more important than accuracy.

PSEUDO CODE

  switch (cleverhash(npc.stats, npc.environment, npc.eventhistory)
           + npc.anger * random()) {
    case 0-9:   if this_state_is_fatal(npc.stats) select a safe state
                else set_state...; break;
    case 10-19: if this_state_is_fatal...
                  ...
    case 20-39: if this_state_is_fatal...
    case 40-45: ...
    case 46-100:
    default:
  }

EXPLANATION

  The basic idea is that you can find a clever hash function that
  will get you to a reasonable state most of the time, but that you
  sometimes will end up with the less perfect state. However, it is
  life-like to make errors, so that might not be a problem for a
  MUD.

  Each state is indexed by an interval. Similar states have index
  values that are close to eachother. The size of the interval
  reflects the probabilty of that state. A random factor is added to
  make behaviour less predictable and the more upset the NPC is
  (npc.anger) the less predictable he becomes.

  Is this workable? Is it possible to construct such hash-functions
  from a perfect heurtistics, by using a genetic algorithm? I should
  think so.

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