I have noticed, then and now, a great deal of discussion about the need
to expand the capabilities of the agents (NPC's) of interactive fiction.
For at _least_ two years, the majority consensus is that IF won't be
able to make any real progress until believable agents are possible.
My problem is that this appears to be all talk and no action. The few
papers I've read about believable agents have all been smoke... "Here's
what you need to do; here's what advances we need to make" but nobody's
actually DONE any of these things as far as I can tell. (And some of
these papers are a few years old.)
The Oz project is working on this, and I know the guys at MIT's AI lab
are interested in it, but has anybody turned out anything that's
directly usable in our current IF efforts? It's nice for all these
papers to dream and predict, but it'd be infinitely better if someone
would actually MAKE these breakthroughs. Eliza and Racter have been
around for years... is the current reasonable-compute-cycles technology
for believable conversation no more advanced than this?
Just what _has_ the Oz project accomplished? (All their papers are in
PS... anybody have text versions? Maybe I'll have to set up
GhostScript.) I've seen a script of Lyotard and it didn't show me
anything I couldn't do myself in TADS in a couple of hours or less.
(That is, I saw one specific series of actions and reactions... the
script showed no evidence of flexibility.) The same for the other
scripts I've seen. Is Oz actually getting somewhere, but refusing to
publish their "real" advances to keep it from being stolen? Or is Oz
still in the baby-step phase... making progress, but producing nothing
significantly usable at this time?
In summary, the IF community has gabbed about believable agents for
years, but has it actually done anything about it? Or is the level of
knowledge required up there in the "get a doctorate in AI first" range?
Or is this technology still entirely out of our reach?
-- Carl (ravenpub@southwind.net) * Contentsoftaglinemaysettleduri.