[MUD-Dev] Efficient AI?

[Brett Bibby]

Ola wrote:

> Just to avoid confusion: the main point was not to have good AI,
> but very fast AI (approx 100 cycles per decision?) with a small
> memory footprint (static code/tables).

Well, the fastest AI you could get, and be anywhere near 100 cycles,
is likely a lookup table and not an algorithm.  We worked on a
project recently with the same problem and came up with the
following idea: Let real human developers control the NPCs and
record each and every action by the "NPC" to files.  After many
sessions, parse the files and use some heuristics (using neural
nets, markov, etc.) to encode the behavior for an encounter into
decisions.  At runtime, in the real game, simply use your hash
function or whatever to lookup and playback the actions/decisions.

Theoretically this has the potential to also be encrypted and stored
on the client and at runtime you only need to send the key (which is
tiny in size) to decrypt the data for complex sequences.  This in
effect gives you mega-compression, thereby cutting bandwidth, etc.

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