NetNews Usenet Archive 1992 #31/NN_1992_31.iso/spool/rec/games/pbm/3453

Internet Message Format | 1992-12-31 | 9.2 KB

Path: sparky!uunet!spool.mu.edu!uwm.edu!cs.utexas.edu!gateway
From: [e-mail] (Luis Miguel Sequeira)
Newsgroups: rec.games.pbm
Subject: Re: Galaxy Results Format
Date: 31 Dec 1992 03:00:49 -0600
Organization: UTexas Mail-to-News Gateway
Lines: 178
Sender: [e-mail]
Message-ID: <[e-mail]>
References: <[e-mail]>
NNTP-Posting-Host: cs.utexas.edu

I wrote:

>Even a TOTALLY OBVIOUS game  (ie., one where all players  have all the
>rules & formulae of  the  game) has a   "non-obvious"  part:  that the
>players are human, and thus, unpredictable...

[e-mail] (Curt Welch) wrote:

>But with many games, like chess, go, etc., you don't really have to
>guess what the other player is going to do.  You basically just assume
>that he will make the best move possible.  There are no secrets in
>the game because you both can see the entire board.  If you are trying
>to guess what your opponent will do, then it's just because you
>can't figure out what your true "best" move is.
>
>But in Galaxy, guessing is required.  Your best move requires
>you to guess what your opponent will do.  There is no absolute
>"best" move like there is in chess.
>
>With chess, it's easy to write an algorithm to calcuate the best move -
>but imposible to find a machine that can run the algorthim in a
>resaonable time.  And the algorithm wouldn't have to guess what move
>the other palyer would make.  It would just find the best possible
>move based on any possible move from the opponent.
>
>With Galaxy, the algorithm would have to make a guess as to what the
>opponent would do.  It's like playing paper-scissors-rock.  The best
>move can't be calculated.

If you want to say that programming a chess player is _easier_ than an
automatic, artificially intelligent PBEM   player, you're  of   course
right! However, don't dismiss it as being "impossible to calculate the
best move".

I won't bother  you  with  lots  of technicalities.  I'm  no expert of
artificial intelligence; I've  studied it, and its possibilities;  but
if you really want some technically correct explanations, you'd better
ask them from someone qualified.

So, I'm just sticking with some basic issues:

You mentioned on  previous posts (sorry,  no quotations, I've  already
deleted them...)  the   way  you  play Galaxy   (and  with  which I've
agreed!): simply don't make gross  errors  (which one won't make if he
begins to understand the rules and their implications), have a goal in
mind (ie. conquering or developing this or that planet, attacking this
player, building  this or that...) and let  a  bit of  intuition guide
you.

Now, of course you  can't "win" all  the time; you're not judging your
moves based  on solid  calculations  and  facts (as  opposed  to David
desJardins approach...).  But you're  counting  on being impossible to
predict what the other players are  going  to do; so, your "best move"
will be the one which, in special circunstances (ie.  if other players
haven't done anything to prevent it),  give you  most success. Even if
the other    players make exceptionally  moves   - which you  couldn't
predict anyway - your own  move  was YOUR  best  move, as you couldn't
make a  better move WITHOUT knowing  what your opponents were going to
do...

What we're dealing with is incomplete knowledge.  There are artificial
intelligence "engines" (or programs) which are  able to deal with this
quite well. Perhaps you are familiar with fuzzy logic (from  the photo
industry, for example); but there are other ways (some quite simple!).

All this to get to the following  point: even if you can't "calculate"
the best move using a formula straight from  the spreadsheet, what you
can do is to assess an "heuristic"  value to your position.  Say, it's
"good" to have a lot of ships and a lot of planets (this increases the
heuristic value  of  the current position),  and  "bad"  to  have  low
industry. This would be  given  values points, for each characteristic
thought relevant. So, the automated Galaxy player  would try to make a
move that _improves_ the  heuristic value (ie., trying  a few  variant
moves, choosing randomly among them, calculate the new heuristic value
of the  resulting  situation, and    choose the one with  the  highest
possible value).

It also could    assess a few  heuristic  values   to other   players'
situations (from the few data you got on them), and make the move that
not only would improve your situation, but actually degrade  the other
players' situations.

Of course, this is more easily said than done.  But you would actually
have  a playing   algorithm mimicking  your  way of    playing: try to
evaluate your own situation, how to improve it  just by looking at the
data you have and "imagining" what other players would  do, and pick a
move.  You couldn't blame the computer  if his move  happened to  be a
terrible  one;  after   all,  it (just   like   yourself) has  limited
information on the "world" around him, and chooses  what is best under
such circunstances.

Perhaps  not your average  chess player program,  but  it  wouldn't be
_that_ difficult to implement.  Of course, on  a  _very slow_ machine,
you'd have to concern yourself  with the issue  of efficience (ie. the
computer should finish its moves before the next turn is due! :-)...


>>I _really_ think that a 2-month, one-week-turn game is _too_ short.
>
>Just to make this clear - I was talking about a 2 month, 5 turn per week
>game.

Uh... sorry...  misunderstood you! Ok,  this sounds fine.   However, I
must  admit that it's  too much for  me. What you propose is  almost a
cross-breed between a MUD and a PBEM :-)

But okay, go for it...


>I haven't really played any of the role-playing games, but I know what
>you mean about liking to build up your player.  And it's kinda neat playing
>a long Galaxy game.  You spend so much time studying the Galaxy, and
>talking to other players that they become somewhat real.  The game
>starts to feel like home.

Exactly my point! :-)

Perhaps  I'd add that,  if turns are very  close together,  unless you
spend  24   hrs   a day  on  the net,   diplomacy  would   be  reduced
dramatically... I  really don't  know. Twice a week   turns are for me
already too fast! Unless you play but one game  at the  time - which I
_don't_.


>But the problem (or just the fact) with Galaxy is that once you loose
>a big fight, you're pretty much out of the game.  There's no real way to
>start over and come back later and win.  One (major) mistake and you're
>dead.  Once you get too far behind in industry you can never catch up.
>
>It might be interesting to try and modify the game so palyers were a
>lot harder to knock out of the game.  Where you could loose more
>battles but not loose the war.  Then the concept of having longer
>games would make more sense.

You are _very_ right.  This is the  trouble  with  a lot  of different
games I've played on. Almost every one of it has this problem.

There is a single exception (a local PBM) I know about (there must be,
of  course, more!    But I  don't  know them...).   It's an open-ended
campaign.  Here diplomacy & politics are the  key to  survival  - more
true if you have lost your main army!  A "crippled"  player, without a
large   army, _isn't_ absolutely   unuseful!  He still  has  money and
resources.  So,  what he  does is   "buying"  protection: he enters  a
kingdom  (or   a similar  multi-player  organization)  and "pays"  for
protection  (really, he just contributes with  some cash and resources
to the kingdom, so that they  can buy a new army  to protect him until
he's able to  develop  his  own).  I know  of hundreds of  examples of
players starting with very poor positions, but that slowly (and I mean
SLOOOOWLY... as the game is running for two  years now...) cuild build
themselves up  again  to a   stable position.  This  is my   case, for
example... :-)

I really don't know how  to implement something  similar in Galaxy, as
there can be but ONE sole winner, and the game ends when this happens.
Perhaps the game could be changed so that the winner is  the  GROUP of
allies  which end the  game alone.  However, this  would mean that the
concept of allegiances would have  to be changed (if everyone declared
Peace on  everyone  else on the  game, it would be over  at the second
turn!). On the game above mentioned, players can  be  grouped together
in "political organizations",  ie. kingdoms, empires,  republics, etc.
You _could_ do something similar in Galaxy...

    - Luis Sequeira
  _________________________________________________________________________
 /
/
    _/     _/     _/  _/_/_/     Computer scientists do it byte by byte.
   _/     _/_/ _/_/ _/
  _/     _/ _/  _/  _/_/_/       "We don't ask for miracles to get the job
 _/     _/     _/       _/        done, we RELY upon them!"
_/_/_/ _/     _/  _/_/_/

  [e-mail]                        Luis Miguel Sequeira
                                  Laboratorio Nacional de Engenharia Civil
  Phone [phone] Ext. 2752   Centro Informatica/Grupo Sistemas Centrais
  "Don't call me, I'll call you"  Av. Brasil, 101 - 1700 Lisboa, Portugal
                                                                          /
_________________________________________________________________________/