On Wed, Dec 05, 2001 at 09:30:43PM -0500, Mark Mielke wrote:
> > But that's not even the interesting point.  The interesting point is
> > that tactical perfection is only a small help in winning netrek games,
> > and another thing computers are bad at is dynamic strategizing, which
> > is the most important thing in winning netrek games.
> 
> You assume that Netrek is a complicated game like Life. It isn't,
> regardless as to how esteemed the 'clued' would like to feel. It is
> certainly an art, but believe it or not, it is possible to make a
> computer beat the best chess player in the world. It just takes a
> lot of investment into strategy and other such things up front.

Netrek is a lot more like go than chess, and go-playing programs suck.
Chess-playing computers don't have great strategy, they just brute-force
look ahead far enough that the strategy doesn't matter.  I remember
the deep blue team noting that if you replace deep blue's position
evaluator with a random number generator, it still plays pretty good
chess.

You can't do that kind of forward analysis with netrek.

> > > Other possibilities include "I'm going to plasma him, press him away
> > > from you and phaser him, and torp him so that we can distract him. The
> > > *instant* he phasers you, I'll launch the plasma."
> > Getting robots to do this level of communication is well beyond the
> > point of diminishing returns.  Maybe you get one extra plasma hit over
> > the course of an entire game.
> 
> This is one of among 1000 things I would complete my attempt with.

You're going to spend probably hundreds of hours of coding time to get
one more plasma hit, while the base is in the WRONG POSITION because
the computer doesn't know how to play netrek.  It's simply the wrong
thing to focus on, but perhaps the only thing you're capable of.

> > Your examples are all in tactical situations.  A tactically perfect
> > bot, using the best netrek robot strategy code ever written, would get
> > its butt kicked by any reasonable human team.  Although that's partly
> > because the people writing netrek robots tend to spend a lot of time
> > on stupid shit like trying to improve dogfighting and phaserlocking
> > and not enough time on interesting and difficult problems, like
> > whether to refuel in place, try to mutual and get a fresh ship, go to
> > a planet to refuel, or go back to core to pick armies.  (To give one
> > small example of the kind of decisions robots are extremely bad at).
> 
> This sounds like a dare. If I had some more spare time I'd ask you to
> put money on it.

There have been several netrek robots done already--we have a pretty
good idea of their flaws, and those don't have anything to do with 
how often their plasmas hit.
 -Tom